1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE 1 & 2 Instructions Classes
19 //===----------------------------------------------------------------------===//
21 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
22 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
23 RegisterClass RC, X86MemOperand x86memop,
25 let isCommutable = 1 in {
26 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
28 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
30 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
32 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
34 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
35 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
36 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
39 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
40 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
41 string asm, string SSEVer, string FPSizeStr,
42 Operand memopr, ComplexPattern mem_cpat,
44 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
46 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
47 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
48 [(set RC:$dst, (!cast<Intrinsic>(
49 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
50 RC:$src1, RC:$src2))]>;
51 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
53 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
54 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
55 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
56 SSEVer, "_", OpcodeStr, FPSizeStr))
57 RC:$src1, mem_cpat:$src2))]>;
60 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
61 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
62 RegisterClass RC, ValueType vt,
63 X86MemOperand x86memop, PatFrag mem_frag,
64 Domain d, bit Is2Addr = 1> {
65 let isCommutable = 1 in
66 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
68 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
69 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
70 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_DEFAULT, d>;
72 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
74 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
75 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
76 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
80 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
81 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
82 string OpcodeStr, X86MemOperand x86memop,
83 list<dag> pat_rr, list<dag> pat_rm,
85 bit rr_hasSideEffects = 0> {
86 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
87 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
89 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
90 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
91 pat_rr, IIC_DEFAULT, d>;
92 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
94 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
95 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
96 pat_rm, IIC_DEFAULT, d>;
99 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
100 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
101 string asm, string SSEVer, string FPSizeStr,
102 X86MemOperand x86memop, PatFrag mem_frag,
103 Domain d, bit Is2Addr = 1> {
104 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
106 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
107 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
108 [(set RC:$dst, (!cast<Intrinsic>(
109 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
110 RC:$src1, RC:$src2))], IIC_DEFAULT, d>;
111 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
113 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
114 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
115 [(set RC:$dst, (!cast<Intrinsic>(
116 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
117 RC:$src1, (mem_frag addr:$src2)))], IIC_DEFAULT, d>;
120 //===----------------------------------------------------------------------===//
121 // Non-instruction patterns
122 //===----------------------------------------------------------------------===//
124 // A vector extract of the first f32/f64 position is a subregister copy
125 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
126 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
127 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
128 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
130 // A 128-bit subvector extract from the first 256-bit vector position
131 // is a subregister copy that needs no instruction.
132 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
133 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
134 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
135 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
137 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
138 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
139 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
140 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
142 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
143 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
144 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
145 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
147 // A 128-bit subvector insert to the first 256-bit vector position
148 // is a subregister copy that needs no instruction.
149 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
150 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
151 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
152 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
153 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
154 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
155 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
156 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
157 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
158 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
159 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
160 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
162 // Implicitly promote a 32-bit scalar to a vector.
163 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
164 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
165 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
166 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
167 // Implicitly promote a 64-bit scalar to a vector.
168 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
169 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
170 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
171 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
173 // Bitcasts between 128-bit vector types. Return the original type since
174 // no instruction is needed for the conversion
175 let Predicates = [HasSSE2] in {
176 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
177 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
178 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
179 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
180 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
181 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
182 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
183 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
184 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
185 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
186 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
187 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
188 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
189 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
190 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
191 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
192 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
193 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
194 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
195 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
196 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
197 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
198 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
199 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
200 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
201 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
202 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
203 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
204 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
205 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
208 // Bitcasts between 256-bit vector types. Return the original type since
209 // no instruction is needed for the conversion
210 let Predicates = [HasAVX] in {
211 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
212 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
213 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
214 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
215 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
216 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
217 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
218 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
219 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
220 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
221 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
222 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
223 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
224 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
225 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
226 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
227 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
228 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
229 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
230 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
231 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
232 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
233 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
234 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
235 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
236 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
237 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
238 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
239 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
240 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
243 // Alias instructions that map fld0 to pxor for sse.
244 // This is expanded by ExpandPostRAPseudos.
245 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
247 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
248 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
249 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
250 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
253 //===----------------------------------------------------------------------===//
254 // AVX & SSE - Zero/One Vectors
255 //===----------------------------------------------------------------------===//
257 // Alias instruction that maps zero vector to pxor / xorp* for sse.
258 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
259 // swizzled by ExecutionDepsFix to pxor.
260 // We set canFoldAsLoad because this can be converted to a constant-pool
261 // load of an all-zeros value if folding it would be beneficial.
262 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
263 isPseudo = 1, neverHasSideEffects = 1 in {
264 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
267 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
268 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
269 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
270 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
271 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
272 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
275 // The same as done above but for AVX. The 256-bit ISA does not support PI,
276 // and doesn't need it because on sandy bridge the register is set to zero
277 // at the rename stage without using any execution unit, so SET0PSY
278 // and SET0PDY can be used for vector int instructions without penalty
279 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
280 // JIT implementatioan, it does not expand the instructions below like
281 // X86MCInstLower does.
282 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
283 isCodeGenOnly = 1 in {
284 let Predicates = [HasAVX] in {
285 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
286 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
287 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
288 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
290 let Predicates = [HasAVX2], neverHasSideEffects = 1 in
291 def AVX2_SET0 : PDI<0xef, MRMInitReg, (outs VR256:$dst), (ins), "",
295 let Predicates = [HasAVX2], AddedComplexity = 5 in {
296 def : Pat<(v4i64 immAllZerosV), (AVX2_SET0)>;
297 def : Pat<(v8i32 immAllZerosV), (AVX2_SET0)>;
298 def : Pat<(v16i16 immAllZerosV), (AVX2_SET0)>;
299 def : Pat<(v32i8 immAllZerosV), (AVX2_SET0)>;
302 // AVX has no support for 256-bit integer instructions, but since the 128-bit
303 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
304 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
305 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
306 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
308 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
309 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
310 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
312 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
313 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
314 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
316 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
317 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
318 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
320 // We set canFoldAsLoad because this can be converted to a constant-pool
321 // load of an all-ones value if folding it would be beneficial.
322 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
323 // JIT implementation, it does not expand the instructions below like
324 // X86MCInstLower does.
325 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
326 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
327 let Predicates = [HasAVX] in
328 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
329 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
330 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
331 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
332 let Predicates = [HasAVX2] in
333 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
334 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
338 //===----------------------------------------------------------------------===//
339 // SSE 1 & 2 - Move FP Scalar Instructions
341 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
342 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
343 // is used instead. Register-to-register movss/movsd is not modeled as an
344 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
345 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
346 //===----------------------------------------------------------------------===//
348 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
349 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
350 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
352 // Loading from memory automatically zeroing upper bits.
353 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
354 PatFrag mem_pat, string OpcodeStr> :
355 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
356 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
357 [(set RC:$dst, (mem_pat addr:$src))]>;
360 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
361 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
363 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
364 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
367 // For the disassembler
368 let isCodeGenOnly = 1 in {
369 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
370 (ins VR128:$src1, FR32:$src2),
371 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
373 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
374 (ins VR128:$src1, FR64:$src2),
375 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
379 let canFoldAsLoad = 1, isReMaterializable = 1 in {
380 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
382 let AddedComplexity = 20 in
383 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
387 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
388 "movss\t{$src, $dst|$dst, $src}",
389 [(store FR32:$src, addr:$dst)]>, XS, VEX, VEX_LIG;
390 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
391 "movsd\t{$src, $dst|$dst, $src}",
392 [(store FR64:$src, addr:$dst)]>, XD, VEX, VEX_LIG;
395 let Constraints = "$src1 = $dst" in {
396 def MOVSSrr : sse12_move_rr<FR32, v4f32,
397 "movss\t{$src2, $dst|$dst, $src2}">, XS;
398 def MOVSDrr : sse12_move_rr<FR64, v2f64,
399 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
401 // For the disassembler
402 let isCodeGenOnly = 1 in {
403 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
404 (ins VR128:$src1, FR32:$src2),
405 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
406 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
407 (ins VR128:$src1, FR64:$src2),
408 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
412 let canFoldAsLoad = 1, isReMaterializable = 1 in {
413 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
415 let AddedComplexity = 20 in
416 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
419 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
420 "movss\t{$src, $dst|$dst, $src}",
421 [(store FR32:$src, addr:$dst)]>;
422 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
423 "movsd\t{$src, $dst|$dst, $src}",
424 [(store FR64:$src, addr:$dst)]>;
427 let Predicates = [HasAVX] in {
428 let AddedComplexity = 15 in {
429 // Extract the low 32-bit value from one vector and insert it into another.
430 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
431 (VMOVSSrr (v4f32 VR128:$src1),
432 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
433 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
434 (VMOVSSrr (v4i32 VR128:$src1),
435 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
437 // Extract the low 64-bit value from one vector and insert it into another.
438 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
439 (VMOVSDrr (v2f64 VR128:$src1),
440 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
441 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
442 (VMOVSDrr (v2i64 VR128:$src1),
443 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
445 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
446 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
447 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
448 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
449 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
451 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
452 // MOVS{S,D} to the lower bits.
453 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
454 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
455 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
456 (VMOVSSrr (v4f32 (V_SET0)),
457 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
458 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
459 (VMOVSSrr (v4i32 (V_SET0)),
460 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
461 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
462 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
464 // Move low f32 and clear high bits.
465 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
466 (SUBREG_TO_REG (i32 0),
467 (VMOVSSrr (v4f32 (V_SET0)),
468 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
469 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
470 (SUBREG_TO_REG (i32 0),
471 (VMOVSSrr (v4i32 (V_SET0)),
472 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
475 let AddedComplexity = 20 in {
476 // MOVSSrm zeros the high parts of the register; represent this
477 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
478 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
479 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
480 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
481 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
482 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
483 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
485 // MOVSDrm zeros the high parts of the register; represent this
486 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
487 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
488 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
489 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
490 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
491 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
492 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
493 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
494 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
495 def : Pat<(v2f64 (X86vzload addr:$src)),
496 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
498 // Represent the same patterns above but in the form they appear for
500 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
501 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
502 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
503 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
504 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
505 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
506 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
507 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
508 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
510 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
511 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
512 (SUBREG_TO_REG (i32 0),
513 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
515 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
516 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
517 (SUBREG_TO_REG (i64 0),
518 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
520 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
521 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
522 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
524 // Move low f64 and clear high bits.
525 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
526 (SUBREG_TO_REG (i32 0),
527 (VMOVSDrr (v2f64 (V_SET0)),
528 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
530 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
531 (SUBREG_TO_REG (i32 0),
532 (VMOVSDrr (v2i64 (V_SET0)),
533 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
535 // Extract and store.
536 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
539 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
540 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
543 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
545 // Shuffle with VMOVSS
546 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
547 (VMOVSSrr VR128:$src1, FR32:$src2)>;
548 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
549 (VMOVSSrr (v4i32 VR128:$src1),
550 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
551 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
552 (VMOVSSrr (v4f32 VR128:$src1),
553 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
556 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
557 (SUBREG_TO_REG (i32 0),
558 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
559 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
560 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
561 (SUBREG_TO_REG (i32 0),
562 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
563 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
565 // Shuffle with VMOVSD
566 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
567 (VMOVSDrr VR128:$src1, FR64:$src2)>;
568 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
569 (VMOVSDrr (v2i64 VR128:$src1),
570 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
571 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
572 (VMOVSDrr (v2f64 VR128:$src1),
573 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
574 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
575 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
577 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
578 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
582 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
583 (SUBREG_TO_REG (i32 0),
584 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
585 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
586 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
587 (SUBREG_TO_REG (i32 0),
588 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
589 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
592 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
593 // is during lowering, where it's not possible to recognize the fold cause
594 // it has two uses through a bitcast. One use disappears at isel time and the
595 // fold opportunity reappears.
596 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
597 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
599 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
600 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
602 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
603 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
605 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
606 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
610 let Predicates = [HasSSE1] in {
611 let AddedComplexity = 15 in {
612 // Extract the low 32-bit value from one vector and insert it into another.
613 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
614 (MOVSSrr (v4f32 VR128:$src1),
615 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
616 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
617 (MOVSSrr (v4i32 VR128:$src1),
618 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
620 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
621 // MOVSS to the lower bits.
622 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
623 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
624 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
625 (MOVSSrr (v4f32 (V_SET0)),
626 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
627 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
628 (MOVSSrr (v4i32 (V_SET0)),
629 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
632 let AddedComplexity = 20 in {
633 // MOVSSrm zeros the high parts of the register; represent this
634 // with SUBREG_TO_REG.
635 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
636 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
637 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
638 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
639 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
640 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
643 // Extract and store.
644 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
647 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
649 // Shuffle with MOVSS
650 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
651 (MOVSSrr VR128:$src1, FR32:$src2)>;
652 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
653 (MOVSSrr (v4i32 VR128:$src1),
654 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
655 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
656 (MOVSSrr (v4f32 VR128:$src1),
657 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
660 let Predicates = [HasSSE2] in {
661 let AddedComplexity = 15 in {
662 // Extract the low 64-bit value from one vector and insert it into another.
663 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
664 (MOVSDrr (v2f64 VR128:$src1),
665 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
666 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
667 (MOVSDrr (v2i64 VR128:$src1),
668 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
670 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
671 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
672 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
673 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
674 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
676 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
677 // MOVSD to the lower bits.
678 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
679 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
682 let AddedComplexity = 20 in {
683 // MOVSDrm zeros the high parts of the register; represent this
684 // with SUBREG_TO_REG.
685 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
686 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
687 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
688 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
689 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
690 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
691 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
692 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
693 def : Pat<(v2f64 (X86vzload addr:$src)),
694 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
697 // Extract and store.
698 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
701 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
703 // Shuffle with MOVSD
704 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
705 (MOVSDrr VR128:$src1, FR64:$src2)>;
706 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
707 (MOVSDrr (v2i64 VR128:$src1),
708 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
709 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
710 (MOVSDrr (v2f64 VR128:$src1),
711 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
712 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
713 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
714 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
715 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
717 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
718 // is during lowering, where it's not possible to recognize the fold cause
719 // it has two uses through a bitcast. One use disappears at isel time and the
720 // fold opportunity reappears.
721 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
722 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
723 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
724 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
725 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
726 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
727 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
728 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
731 //===----------------------------------------------------------------------===//
732 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
733 //===----------------------------------------------------------------------===//
735 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
736 X86MemOperand x86memop, PatFrag ld_frag,
737 string asm, Domain d,
738 bit IsReMaterializable = 1> {
739 let neverHasSideEffects = 1 in
740 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
741 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], IIC_DEFAULT, d>;
742 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
743 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
744 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
745 [(set RC:$dst, (ld_frag addr:$src))], IIC_DEFAULT, d>;
748 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
749 "movaps", SSEPackedSingle>, TB, VEX;
750 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
751 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
752 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
753 "movups", SSEPackedSingle>, TB, VEX;
754 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
755 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
757 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
758 "movaps", SSEPackedSingle>, TB, VEX;
759 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
760 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
761 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
762 "movups", SSEPackedSingle>, TB, VEX;
763 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
764 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
765 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
766 "movaps", SSEPackedSingle>, TB;
767 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
768 "movapd", SSEPackedDouble>, TB, OpSize;
769 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
770 "movups", SSEPackedSingle>, TB;
771 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
772 "movupd", SSEPackedDouble, 0>, TB, OpSize;
774 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
775 "movaps\t{$src, $dst|$dst, $src}",
776 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
777 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
778 "movapd\t{$src, $dst|$dst, $src}",
779 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
780 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
781 "movups\t{$src, $dst|$dst, $src}",
782 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
783 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
784 "movupd\t{$src, $dst|$dst, $src}",
785 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
786 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
787 "movaps\t{$src, $dst|$dst, $src}",
788 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
789 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
790 "movapd\t{$src, $dst|$dst, $src}",
791 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
792 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
793 "movups\t{$src, $dst|$dst, $src}",
794 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
795 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
796 "movupd\t{$src, $dst|$dst, $src}",
797 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
800 let isCodeGenOnly = 1 in {
801 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
803 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
804 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
806 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
807 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
809 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
810 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
812 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
813 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
815 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
816 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
818 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
819 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
821 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
822 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
824 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
827 let Predicates = [HasAVX] in {
828 def : Pat<(v8i32 (X86vzmovl
829 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
830 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
831 def : Pat<(v4i64 (X86vzmovl
832 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
833 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
834 def : Pat<(v8f32 (X86vzmovl
835 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
836 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
837 def : Pat<(v4f64 (X86vzmovl
838 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
839 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
843 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
844 (VMOVUPSYmr addr:$dst, VR256:$src)>;
845 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
846 (VMOVUPDYmr addr:$dst, VR256:$src)>;
848 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
849 "movaps\t{$src, $dst|$dst, $src}",
850 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
851 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
852 "movapd\t{$src, $dst|$dst, $src}",
853 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
854 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
855 "movups\t{$src, $dst|$dst, $src}",
856 [(store (v4f32 VR128:$src), addr:$dst)]>;
857 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
858 "movupd\t{$src, $dst|$dst, $src}",
859 [(store (v2f64 VR128:$src), addr:$dst)]>;
862 let isCodeGenOnly = 1 in {
863 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
864 "movaps\t{$src, $dst|$dst, $src}", []>;
865 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
866 "movapd\t{$src, $dst|$dst, $src}", []>;
867 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
868 "movups\t{$src, $dst|$dst, $src}", []>;
869 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
870 "movupd\t{$src, $dst|$dst, $src}", []>;
873 let Predicates = [HasAVX] in {
874 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
875 (VMOVUPSmr addr:$dst, VR128:$src)>;
876 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
877 (VMOVUPDmr addr:$dst, VR128:$src)>;
880 let Predicates = [HasSSE1] in
881 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
882 (MOVUPSmr addr:$dst, VR128:$src)>;
883 let Predicates = [HasSSE2] in
884 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
885 (MOVUPDmr addr:$dst, VR128:$src)>;
887 // Use vmovaps/vmovups for AVX integer load/store.
888 let Predicates = [HasAVX] in {
889 // 128-bit load/store
890 def : Pat<(alignedloadv2i64 addr:$src),
891 (VMOVAPSrm addr:$src)>;
892 def : Pat<(loadv2i64 addr:$src),
893 (VMOVUPSrm addr:$src)>;
895 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
896 (VMOVAPSmr addr:$dst, VR128:$src)>;
897 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
898 (VMOVAPSmr addr:$dst, VR128:$src)>;
899 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
900 (VMOVAPSmr addr:$dst, VR128:$src)>;
901 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
902 (VMOVAPSmr addr:$dst, VR128:$src)>;
903 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
904 (VMOVUPSmr addr:$dst, VR128:$src)>;
905 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
906 (VMOVUPSmr addr:$dst, VR128:$src)>;
907 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
908 (VMOVUPSmr addr:$dst, VR128:$src)>;
909 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
910 (VMOVUPSmr addr:$dst, VR128:$src)>;
912 // 256-bit load/store
913 def : Pat<(alignedloadv4i64 addr:$src),
914 (VMOVAPSYrm addr:$src)>;
915 def : Pat<(loadv4i64 addr:$src),
916 (VMOVUPSYrm addr:$src)>;
917 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
918 (VMOVAPSYmr addr:$dst, VR256:$src)>;
919 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
920 (VMOVAPSYmr addr:$dst, VR256:$src)>;
921 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
922 (VMOVAPSYmr addr:$dst, VR256:$src)>;
923 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
924 (VMOVAPSYmr addr:$dst, VR256:$src)>;
925 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
926 (VMOVUPSYmr addr:$dst, VR256:$src)>;
927 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
928 (VMOVUPSYmr addr:$dst, VR256:$src)>;
929 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
930 (VMOVUPSYmr addr:$dst, VR256:$src)>;
931 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
932 (VMOVUPSYmr addr:$dst, VR256:$src)>;
935 // Use movaps / movups for SSE integer load / store (one byte shorter).
936 // The instructions selected below are then converted to MOVDQA/MOVDQU
937 // during the SSE domain pass.
938 let Predicates = [HasSSE1] in {
939 def : Pat<(alignedloadv2i64 addr:$src),
940 (MOVAPSrm addr:$src)>;
941 def : Pat<(loadv2i64 addr:$src),
942 (MOVUPSrm addr:$src)>;
944 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
945 (MOVAPSmr addr:$dst, VR128:$src)>;
946 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
947 (MOVAPSmr addr:$dst, VR128:$src)>;
948 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
949 (MOVAPSmr addr:$dst, VR128:$src)>;
950 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
951 (MOVAPSmr addr:$dst, VR128:$src)>;
952 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
953 (MOVUPSmr addr:$dst, VR128:$src)>;
954 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
955 (MOVUPSmr addr:$dst, VR128:$src)>;
956 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
957 (MOVUPSmr addr:$dst, VR128:$src)>;
958 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
959 (MOVUPSmr addr:$dst, VR128:$src)>;
962 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
963 // bits are disregarded. FIXME: Set encoding to pseudo!
964 let neverHasSideEffects = 1 in {
965 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
966 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
967 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
968 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
969 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
970 "movaps\t{$src, $dst|$dst, $src}", []>;
971 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
972 "movapd\t{$src, $dst|$dst, $src}", []>;
975 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
976 // bits are disregarded. FIXME: Set encoding to pseudo!
977 let canFoldAsLoad = 1, isReMaterializable = 1 in {
978 let isCodeGenOnly = 1 in {
979 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
980 "movaps\t{$src, $dst|$dst, $src}",
981 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
982 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
983 "movapd\t{$src, $dst|$dst, $src}",
984 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
986 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
987 "movaps\t{$src, $dst|$dst, $src}",
988 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
989 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
990 "movapd\t{$src, $dst|$dst, $src}",
991 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
994 //===----------------------------------------------------------------------===//
995 // SSE 1 & 2 - Move Low packed FP Instructions
996 //===----------------------------------------------------------------------===//
998 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
999 PatFrag mov_frag, string base_opc,
1001 def PSrm : PI<opc, MRMSrcMem,
1002 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1003 !strconcat(base_opc, "s", asm_opr),
1006 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1007 IIC_DEFAULT, SSEPackedSingle>, TB;
1009 def PDrm : PI<opc, MRMSrcMem,
1010 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1011 !strconcat(base_opc, "d", asm_opr),
1012 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
1013 (scalar_to_vector (loadf64 addr:$src2)))))],
1014 IIC_DEFAULT, SSEPackedDouble>, TB, OpSize;
1017 let AddedComplexity = 20 in {
1018 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1019 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1021 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1022 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1023 "\t{$src2, $dst|$dst, $src2}">;
1026 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1027 "movlps\t{$src, $dst|$dst, $src}",
1028 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1029 (iPTR 0))), addr:$dst)]>, VEX;
1030 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1031 "movlpd\t{$src, $dst|$dst, $src}",
1032 [(store (f64 (vector_extract (v2f64 VR128:$src),
1033 (iPTR 0))), addr:$dst)]>, VEX;
1034 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1035 "movlps\t{$src, $dst|$dst, $src}",
1036 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1037 (iPTR 0))), addr:$dst)]>;
1038 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1039 "movlpd\t{$src, $dst|$dst, $src}",
1040 [(store (f64 (vector_extract (v2f64 VR128:$src),
1041 (iPTR 0))), addr:$dst)]>;
1043 let Predicates = [HasAVX] in {
1044 let AddedComplexity = 20 in {
1045 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1046 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1047 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1048 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1049 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1050 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1051 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1052 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1053 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1054 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1057 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1058 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1059 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1060 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1061 VR128:$src2)), addr:$src1),
1062 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1064 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1065 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1066 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1067 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1068 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1070 // Shuffle with VMOVLPS
1071 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1072 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1073 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1074 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1075 def : Pat<(X86Movlps VR128:$src1,
1076 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1077 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1079 // Shuffle with VMOVLPD
1080 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1081 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1082 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1083 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1084 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1085 (scalar_to_vector (loadf64 addr:$src2)))),
1086 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1089 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1091 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1092 def : Pat<(store (v4i32 (X86Movlps
1093 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1094 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1095 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1097 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1098 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1100 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1103 let Predicates = [HasSSE1] in {
1104 let AddedComplexity = 20 in {
1105 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1106 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1107 (MOVLPSrm VR128:$src1, addr:$src2)>;
1108 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1109 (MOVLPSrm VR128:$src1, addr:$src2)>;
1112 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1113 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1114 (iPTR 0))), addr:$src1),
1115 (MOVLPSmr addr:$src1, VR128:$src2)>;
1116 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1117 (MOVLPSmr addr:$src1, VR128:$src2)>;
1118 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1119 VR128:$src2)), addr:$src1),
1120 (MOVLPSmr addr:$src1, VR128:$src2)>;
1122 // Shuffle with MOVLPS
1123 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1124 (MOVLPSrm VR128:$src1, addr:$src2)>;
1125 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1126 (MOVLPSrm VR128:$src1, addr:$src2)>;
1127 def : Pat<(X86Movlps VR128:$src1,
1128 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1129 (MOVLPSrm VR128:$src1, addr:$src2)>;
1130 def : Pat<(X86Movlps VR128:$src1,
1131 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1132 (MOVLPSrm VR128:$src1, addr:$src2)>;
1135 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1137 (MOVLPSmr addr:$src1, VR128:$src2)>;
1138 def : Pat<(store (v4i32 (X86Movlps
1139 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1141 (MOVLPSmr addr:$src1, VR128:$src2)>;
1144 let Predicates = [HasSSE2] in {
1145 let AddedComplexity = 20 in {
1146 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1147 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1148 (MOVLPDrm VR128:$src1, addr:$src2)>;
1149 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1150 (MOVLPDrm VR128:$src1, addr:$src2)>;
1153 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1154 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1155 (MOVLPDmr addr:$src1, VR128:$src2)>;
1156 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1157 (MOVLPDmr addr:$src1, VR128:$src2)>;
1159 // Shuffle with MOVLPD
1160 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1161 (MOVLPDrm VR128:$src1, addr:$src2)>;
1162 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1163 (MOVLPDrm VR128:$src1, addr:$src2)>;
1164 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1165 (scalar_to_vector (loadf64 addr:$src2)))),
1166 (MOVLPDrm VR128:$src1, addr:$src2)>;
1169 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1171 (MOVLPDmr addr:$src1, VR128:$src2)>;
1172 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1174 (MOVLPDmr addr:$src1, VR128:$src2)>;
1177 //===----------------------------------------------------------------------===//
1178 // SSE 1 & 2 - Move Hi packed FP Instructions
1179 //===----------------------------------------------------------------------===//
1181 let AddedComplexity = 20 in {
1182 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1183 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1185 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1186 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1187 "\t{$src2, $dst|$dst, $src2}">;
1190 // v2f64 extract element 1 is always custom lowered to unpack high to low
1191 // and extract element 0 so the non-store version isn't too horrible.
1192 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1193 "movhps\t{$src, $dst|$dst, $src}",
1194 [(store (f64 (vector_extract
1195 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1196 (bc_v2f64 (v4f32 VR128:$src))),
1197 (iPTR 0))), addr:$dst)]>, VEX;
1198 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1199 "movhpd\t{$src, $dst|$dst, $src}",
1200 [(store (f64 (vector_extract
1201 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1202 (iPTR 0))), addr:$dst)]>, VEX;
1203 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1204 "movhps\t{$src, $dst|$dst, $src}",
1205 [(store (f64 (vector_extract
1206 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1207 (bc_v2f64 (v4f32 VR128:$src))),
1208 (iPTR 0))), addr:$dst)]>;
1209 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1210 "movhpd\t{$src, $dst|$dst, $src}",
1211 [(store (f64 (vector_extract
1212 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1213 (iPTR 0))), addr:$dst)]>;
1215 let Predicates = [HasAVX] in {
1217 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1218 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1219 def : Pat<(X86Movlhps VR128:$src1,
1220 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1221 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1222 def : Pat<(X86Movlhps VR128:$src1,
1223 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1224 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1225 def : Pat<(X86Movlhps VR128:$src1,
1226 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1227 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1229 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1230 // is during lowering, where it's not possible to recognize the load fold
1231 // cause it has two uses through a bitcast. One use disappears at isel time
1232 // and the fold opportunity reappears.
1233 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1234 (scalar_to_vector (loadf64 addr:$src2)))),
1235 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1237 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1238 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1239 (scalar_to_vector (loadf64 addr:$src2)))),
1240 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1243 let Predicates = [HasSSE1] in {
1245 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1246 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1247 def : Pat<(X86Movlhps VR128:$src1,
1248 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1249 (MOVHPSrm VR128:$src1, addr:$src2)>;
1250 def : Pat<(X86Movlhps VR128:$src1,
1251 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1252 (MOVHPSrm VR128:$src1, addr:$src2)>;
1253 def : Pat<(X86Movlhps VR128:$src1,
1254 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1255 (MOVHPSrm VR128:$src1, addr:$src2)>;
1258 let Predicates = [HasSSE2] in {
1259 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1260 // is during lowering, where it's not possible to recognize the load fold
1261 // cause it has two uses through a bitcast. One use disappears at isel time
1262 // and the fold opportunity reappears.
1263 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1264 (scalar_to_vector (loadf64 addr:$src2)))),
1265 (MOVHPDrm VR128:$src1, addr:$src2)>;
1267 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1268 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1269 (scalar_to_vector (loadf64 addr:$src2)))),
1270 (MOVHPDrm VR128:$src1, addr:$src2)>;
1273 //===----------------------------------------------------------------------===//
1274 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1275 //===----------------------------------------------------------------------===//
1277 let AddedComplexity = 20 in {
1278 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1279 (ins VR128:$src1, VR128:$src2),
1280 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1282 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))]>,
1284 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1285 (ins VR128:$src1, VR128:$src2),
1286 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1288 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))]>,
1291 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1292 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1293 (ins VR128:$src1, VR128:$src2),
1294 "movlhps\t{$src2, $dst|$dst, $src2}",
1296 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))]>;
1297 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1298 (ins VR128:$src1, VR128:$src2),
1299 "movhlps\t{$src2, $dst|$dst, $src2}",
1301 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))]>;
1304 let Predicates = [HasAVX] in {
1306 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1307 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1308 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1309 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1312 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1313 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1316 let Predicates = [HasSSE1] in {
1318 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1319 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1320 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1321 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1324 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1325 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1328 //===----------------------------------------------------------------------===//
1329 // SSE 1 & 2 - Conversion Instructions
1330 //===----------------------------------------------------------------------===//
1332 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1333 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1335 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1336 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1337 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1338 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1341 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1342 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1343 string asm, Domain d> {
1344 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1345 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1347 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1348 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1352 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1353 X86MemOperand x86memop, string asm> {
1354 let neverHasSideEffects = 1 in {
1355 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1356 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1358 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1359 (ins DstRC:$src1, x86memop:$src),
1360 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1361 } // neverHasSideEffects = 1
1364 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1365 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1367 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1368 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1370 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1371 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1373 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1374 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1375 VEX, VEX_W, VEX_LIG;
1377 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1378 // register, but the same isn't true when only using memory operands,
1379 // provide other assembly "l" and "q" forms to address this explicitly
1380 // where appropriate to do so.
1381 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1383 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1384 VEX_4V, VEX_W, VEX_LIG;
1385 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1387 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1389 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1390 VEX_4V, VEX_W, VEX_LIG;
1392 let Predicates = [HasAVX], AddedComplexity = 1 in {
1393 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1394 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1395 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1396 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1397 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1398 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1399 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1400 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1402 def : Pat<(f32 (sint_to_fp GR32:$src)),
1403 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1404 def : Pat<(f32 (sint_to_fp GR64:$src)),
1405 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1406 def : Pat<(f64 (sint_to_fp GR32:$src)),
1407 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1408 def : Pat<(f64 (sint_to_fp GR64:$src)),
1409 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1412 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1413 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1414 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1415 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1416 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1417 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1418 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1419 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1420 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1421 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1422 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1423 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1424 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1425 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1426 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1427 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1429 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1430 // and/or XMM operand(s).
1432 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1433 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1435 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1436 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1437 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1438 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1439 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1440 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1443 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1444 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1445 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1446 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1448 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1449 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1450 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1451 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1452 (ins DstRC:$src1, x86memop:$src2),
1454 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1455 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1456 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1459 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1460 f128mem, load, "cvtsd2si">, XD, VEX, VEX_LIG;
1461 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1462 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1463 XD, VEX, VEX_W, VEX_LIG;
1465 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1466 f128mem, load, "cvtsd2si{l}">, XD;
1467 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1468 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1471 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1472 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1473 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1474 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1476 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1477 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1478 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1479 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1482 let Constraints = "$src1 = $dst" in {
1483 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1484 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1486 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1487 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1488 "cvtsi2ss{q}">, XS, REX_W;
1489 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1490 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1492 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1493 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1494 "cvtsi2sd">, XD, REX_W;
1499 // Aliases for intrinsics
1500 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1501 f32mem, load, "cvttss2si">, XS, VEX;
1502 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1503 int_x86_sse_cvttss2si64, f32mem, load,
1504 "cvttss2si">, XS, VEX, VEX_W;
1505 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1506 f128mem, load, "cvttsd2si">, XD, VEX;
1507 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1508 int_x86_sse2_cvttsd2si64, f128mem, load,
1509 "cvttsd2si">, XD, VEX, VEX_W;
1510 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1511 f32mem, load, "cvttss2si">, XS;
1512 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1513 int_x86_sse_cvttss2si64, f32mem, load,
1514 "cvttss2si{q}">, XS, REX_W;
1515 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1516 f128mem, load, "cvttsd2si">, XD;
1517 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1518 int_x86_sse2_cvttsd2si64, f128mem, load,
1519 "cvttsd2si{q}">, XD, REX_W;
1521 let Pattern = []<dag> in {
1522 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1523 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1525 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1526 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1528 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1529 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1530 SSEPackedSingle>, TB, VEX;
1531 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1532 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1533 SSEPackedSingle>, TB, VEX;
1536 let Pattern = []<dag> in {
1537 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1538 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1539 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1540 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1541 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1542 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1543 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1546 let Predicates = [HasAVX] in {
1547 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1548 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1549 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1550 (VCVTSS2SIrm addr:$src)>;
1551 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1552 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1553 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1554 (VCVTSS2SI64rm addr:$src)>;
1557 let Predicates = [HasSSE1] in {
1558 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1559 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1560 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1561 (CVTSS2SIrm addr:$src)>;
1562 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1563 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1564 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1565 (CVTSS2SI64rm addr:$src)>;
1570 // Convert scalar double to scalar single
1571 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1572 (ins FR64:$src1, FR64:$src2),
1573 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1576 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1577 (ins FR64:$src1, f64mem:$src2),
1578 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1579 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1581 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1584 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1585 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1586 [(set FR32:$dst, (fround FR64:$src))]>;
1587 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1588 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1589 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1590 Requires<[HasSSE2, OptForSize]>;
1592 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1593 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1595 let Constraints = "$src1 = $dst" in
1596 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1597 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1599 // Convert scalar single to scalar double
1600 // SSE2 instructions with XS prefix
1601 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1602 (ins FR32:$src1, FR32:$src2),
1603 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1604 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1606 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1607 (ins FR32:$src1, f32mem:$src2),
1608 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1609 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1611 let Predicates = [HasAVX] in {
1612 def : Pat<(f64 (fextend FR32:$src)),
1613 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1614 def : Pat<(fextend (loadf32 addr:$src)),
1615 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1616 def : Pat<(extloadf32 addr:$src),
1617 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1620 def : Pat<(extloadf32 addr:$src),
1621 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1622 Requires<[HasAVX, OptForSpeed]>;
1624 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1625 "cvtss2sd\t{$src, $dst|$dst, $src}",
1626 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1627 Requires<[HasSSE2]>;
1628 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1629 "cvtss2sd\t{$src, $dst|$dst, $src}",
1630 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1631 Requires<[HasSSE2, OptForSize]>;
1633 // extload f32 -> f64. This matches load+fextend because we have a hack in
1634 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1636 // Since these loads aren't folded into the fextend, we have to match it
1638 def : Pat<(fextend (loadf32 addr:$src)),
1639 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1640 def : Pat<(extloadf32 addr:$src),
1641 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1643 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1644 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1645 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1646 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1647 VR128:$src2))]>, XS, VEX_4V,
1649 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1650 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1651 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1652 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1653 (load addr:$src2)))]>, XS, VEX_4V,
1655 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1656 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1657 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1658 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1659 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1660 VR128:$src2))]>, XS,
1661 Requires<[HasSSE2]>;
1662 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1663 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1664 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1665 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1666 (load addr:$src2)))]>, XS,
1667 Requires<[HasSSE2]>;
1670 // Convert doubleword to packed single/double fp
1671 // SSE2 instructions without OpSize prefix
1672 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1673 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1674 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1675 TB, VEX, Requires<[HasAVX]>;
1676 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1677 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1678 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1679 (bitconvert (memopv2i64 addr:$src))))]>,
1680 TB, VEX, Requires<[HasAVX]>;
1681 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1682 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1683 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1684 TB, Requires<[HasSSE2]>;
1685 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1686 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1687 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1688 (bitconvert (memopv2i64 addr:$src))))]>,
1689 TB, Requires<[HasSSE2]>;
1691 // FIXME: why the non-intrinsic version is described as SSE3?
1692 // SSE2 instructions with XS prefix
1693 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1694 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1695 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1696 XS, VEX, Requires<[HasAVX]>;
1697 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1698 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1699 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1700 (bitconvert (memopv2i64 addr:$src))))]>,
1701 XS, VEX, Requires<[HasAVX]>;
1702 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1703 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1704 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1705 XS, Requires<[HasSSE2]>;
1706 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1707 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1708 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1709 (bitconvert (memopv2i64 addr:$src))))]>,
1710 XS, Requires<[HasSSE2]>;
1713 // Convert packed single/double fp to doubleword
1714 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1715 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1716 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1717 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1718 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1719 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1720 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1721 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1722 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1723 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1724 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1725 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1727 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1728 "cvtps2dq\t{$src, $dst|$dst, $src}",
1729 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1731 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1733 "cvtps2dq\t{$src, $dst|$dst, $src}",
1734 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1735 (memop addr:$src)))]>, VEX;
1736 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1737 "cvtps2dq\t{$src, $dst|$dst, $src}",
1738 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1739 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1740 "cvtps2dq\t{$src, $dst|$dst, $src}",
1741 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1742 (memop addr:$src)))]>;
1744 // SSE2 packed instructions with XD prefix
1745 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1746 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1747 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1748 XD, VEX, Requires<[HasAVX]>;
1749 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1750 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1751 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1752 (memop addr:$src)))]>,
1753 XD, VEX, Requires<[HasAVX]>;
1754 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1755 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1756 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1757 XD, Requires<[HasSSE2]>;
1758 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1759 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1760 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1761 (memop addr:$src)))]>,
1762 XD, Requires<[HasSSE2]>;
1765 // Convert with truncation packed single/double fp to doubleword
1766 // SSE2 packed instructions with XS prefix
1767 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1768 "cvttps2dq\t{$src, $dst|$dst, $src}",
1770 (int_x86_sse2_cvttps2dq VR128:$src))]>, VEX;
1771 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1772 "cvttps2dq\t{$src, $dst|$dst, $src}",
1773 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1774 (memop addr:$src)))]>, VEX;
1775 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1776 "cvttps2dq\t{$src, $dst|$dst, $src}",
1778 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))]>, VEX;
1779 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1780 "cvttps2dq\t{$src, $dst|$dst, $src}",
1781 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1782 (memopv8f32 addr:$src)))]>, VEX;
1784 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1785 "cvttps2dq\t{$src, $dst|$dst, $src}",
1787 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1788 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1789 "cvttps2dq\t{$src, $dst|$dst, $src}",
1791 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1793 let Predicates = [HasAVX] in {
1794 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1795 (Int_VCVTDQ2PSrr VR128:$src)>;
1796 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1797 (Int_VCVTDQ2PSrm addr:$src)>;
1799 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1800 (VCVTTPS2DQrr VR128:$src)>;
1801 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1802 (VCVTTPS2DQrm addr:$src)>;
1804 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1805 (VCVTDQ2PSYrr VR256:$src)>;
1806 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1807 (VCVTDQ2PSYrm addr:$src)>;
1809 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1810 (VCVTTPS2DQYrr VR256:$src)>;
1811 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1812 (VCVTTPS2DQYrm addr:$src)>;
1815 let Predicates = [HasSSE2] in {
1816 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1817 (Int_CVTDQ2PSrr VR128:$src)>;
1818 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1819 (Int_CVTDQ2PSrm addr:$src)>;
1821 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1822 (CVTTPS2DQrr VR128:$src)>;
1823 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1824 (CVTTPS2DQrm addr:$src)>;
1827 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1828 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1830 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1831 let isCodeGenOnly = 1 in
1832 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1833 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1834 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1835 (memop addr:$src)))]>, VEX;
1836 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1837 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1838 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1839 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1840 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1841 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1842 (memop addr:$src)))]>;
1844 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1845 // register, but the same isn't true when using memory operands instead.
1846 // Provide other assembly rr and rm forms to address this explicitly.
1847 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1848 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1851 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1852 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1853 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1854 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1857 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1858 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1859 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1860 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1862 // Convert packed single to packed double
1863 let Predicates = [HasAVX] in {
1864 // SSE2 instructions without OpSize prefix
1865 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1866 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1867 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1868 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1869 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1870 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1871 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1872 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1874 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1875 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1876 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1877 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1879 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1880 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1881 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1882 TB, VEX, Requires<[HasAVX]>;
1883 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1884 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1885 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1886 (load addr:$src)))]>,
1887 TB, VEX, Requires<[HasAVX]>;
1888 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1889 "cvtps2pd\t{$src, $dst|$dst, $src}",
1890 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1891 TB, Requires<[HasSSE2]>;
1892 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1893 "cvtps2pd\t{$src, $dst|$dst, $src}",
1894 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1895 (load addr:$src)))]>,
1896 TB, Requires<[HasSSE2]>;
1898 // Convert packed double to packed single
1899 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1900 // register, but the same isn't true when using memory operands instead.
1901 // Provide other assembly rr and rm forms to address this explicitly.
1902 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1903 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1904 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1905 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1908 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1909 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1910 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1911 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1914 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1915 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1916 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1917 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1918 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1919 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1920 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1921 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1924 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1925 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1926 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1927 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1929 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1930 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1931 (memop addr:$src)))]>;
1932 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1933 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1934 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1935 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1936 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1937 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1938 (memop addr:$src)))]>;
1940 // AVX 256-bit register conversion intrinsics
1941 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1942 // whenever possible to avoid declaring two versions of each one.
1943 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1944 (VCVTDQ2PSYrr VR256:$src)>;
1945 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
1946 (VCVTDQ2PSYrm addr:$src)>;
1948 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1949 (VCVTPD2PSYrr VR256:$src)>;
1950 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1951 (VCVTPD2PSYrm addr:$src)>;
1953 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1954 (VCVTPS2DQYrr VR256:$src)>;
1955 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1956 (VCVTPS2DQYrm addr:$src)>;
1958 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1959 (VCVTPS2PDYrr VR128:$src)>;
1960 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1961 (VCVTPS2PDYrm addr:$src)>;
1963 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1964 (VCVTTPD2DQYrr VR256:$src)>;
1965 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1966 (VCVTTPD2DQYrm addr:$src)>;
1968 // Match fround and fextend for 128/256-bit conversions
1969 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1970 (VCVTPD2PSYrr VR256:$src)>;
1971 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1972 (VCVTPD2PSYrm addr:$src)>;
1974 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1975 (VCVTPS2PDYrr VR128:$src)>;
1976 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1977 (VCVTPS2PDYrm addr:$src)>;
1979 //===----------------------------------------------------------------------===//
1980 // SSE 1 & 2 - Compare Instructions
1981 //===----------------------------------------------------------------------===//
1983 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1984 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1985 SDNode OpNode, ValueType VT, PatFrag ld_frag,
1986 string asm, string asm_alt> {
1987 def rr : SIi8<0xC2, MRMSrcReg,
1988 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
1989 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
1990 def rm : SIi8<0xC2, MRMSrcMem,
1991 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
1992 [(set RC:$dst, (OpNode (VT RC:$src1),
1993 (ld_frag addr:$src2), imm:$cc))]>;
1995 // Accept explicit immediate argument form instead of comparison code.
1996 let neverHasSideEffects = 1 in {
1997 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
1998 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2000 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2001 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2005 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2006 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2007 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2008 XS, VEX_4V, VEX_LIG;
2009 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2010 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2011 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2012 XD, VEX_4V, VEX_LIG;
2014 let Constraints = "$src1 = $dst" in {
2015 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2016 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2017 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2019 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2020 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2021 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2025 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2026 Intrinsic Int, string asm> {
2027 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2028 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2029 [(set VR128:$dst, (Int VR128:$src1,
2030 VR128:$src, imm:$cc))]>;
2031 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2032 (ins VR128:$src1, x86memop:$src, SSECC:$cc), asm,
2033 [(set VR128:$dst, (Int VR128:$src1,
2034 (load addr:$src), imm:$cc))]>;
2037 // Aliases to match intrinsics which expect XMM operand(s).
2038 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2039 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2041 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2042 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2044 let Constraints = "$src1 = $dst" in {
2045 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2046 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2047 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2048 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2052 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2053 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2054 ValueType vt, X86MemOperand x86memop,
2055 PatFrag ld_frag, string OpcodeStr, Domain d> {
2056 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2057 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2058 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2060 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2061 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2062 [(set EFLAGS, (OpNode (vt RC:$src1),
2063 (ld_frag addr:$src2)))],
2067 let Defs = [EFLAGS] in {
2068 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2069 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2070 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2071 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2073 let Pattern = []<dag> in {
2074 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2075 "comiss", SSEPackedSingle>, TB, VEX,
2077 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2078 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2082 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2083 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2084 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2085 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2087 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2088 load, "comiss", SSEPackedSingle>, TB, VEX;
2089 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2090 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2091 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2092 "ucomiss", SSEPackedSingle>, TB;
2093 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2094 "ucomisd", SSEPackedDouble>, TB, OpSize;
2096 let Pattern = []<dag> in {
2097 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2098 "comiss", SSEPackedSingle>, TB;
2099 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2100 "comisd", SSEPackedDouble>, TB, OpSize;
2103 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2104 load, "ucomiss", SSEPackedSingle>, TB;
2105 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2106 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2108 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2109 "comiss", SSEPackedSingle>, TB;
2110 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2111 "comisd", SSEPackedDouble>, TB, OpSize;
2112 } // Defs = [EFLAGS]
2114 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2115 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2116 Intrinsic Int, string asm, string asm_alt,
2118 let isAsmParserOnly = 1 in {
2119 def rri : PIi8<0xC2, MRMSrcReg,
2120 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2121 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2123 def rmi : PIi8<0xC2, MRMSrcMem,
2124 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2125 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2129 // Accept explicit immediate argument form instead of comparison code.
2130 def rri_alt : PIi8<0xC2, MRMSrcReg,
2131 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2132 asm_alt, [], IIC_DEFAULT, d>;
2133 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2134 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2135 asm_alt, [], IIC_DEFAULT, d>;
2138 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2139 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2140 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2141 SSEPackedSingle>, TB, VEX_4V;
2142 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2143 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2144 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2145 SSEPackedDouble>, TB, OpSize, VEX_4V;
2146 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2147 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2148 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2149 SSEPackedSingle>, TB, VEX_4V;
2150 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2151 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2152 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2153 SSEPackedDouble>, TB, OpSize, VEX_4V;
2154 let Constraints = "$src1 = $dst" in {
2155 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2156 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2157 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2158 SSEPackedSingle>, TB;
2159 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2160 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2161 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2162 SSEPackedDouble>, TB, OpSize;
2165 let Predicates = [HasAVX] in {
2166 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2167 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2168 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2169 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2170 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2171 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2172 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2173 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2175 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2176 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2177 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2178 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2179 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2180 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2181 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2182 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2185 let Predicates = [HasSSE1] in {
2186 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2187 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2188 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2189 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2192 let Predicates = [HasSSE2] in {
2193 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2194 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2195 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2196 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2199 //===----------------------------------------------------------------------===//
2200 // SSE 1 & 2 - Shuffle Instructions
2201 //===----------------------------------------------------------------------===//
2203 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2204 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2205 ValueType vt, string asm, PatFrag mem_frag,
2206 Domain d, bit IsConvertibleToThreeAddress = 0> {
2207 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2208 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2209 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2210 (i8 imm:$src3))))], IIC_DEFAULT, d>;
2211 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2212 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2213 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2214 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2215 (i8 imm:$src3))))], IIC_DEFAULT, d>;
2218 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2219 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2220 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2221 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2222 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2223 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2224 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2225 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2226 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2227 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2228 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2229 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2231 let Constraints = "$src1 = $dst" in {
2232 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2233 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2234 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2236 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2237 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2238 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2242 let Predicates = [HasAVX] in {
2243 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2244 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2245 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2246 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2247 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2249 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2250 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2251 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2252 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2253 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2256 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2257 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2258 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2259 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2260 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2262 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2263 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2264 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2265 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2266 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2269 let Predicates = [HasSSE1] in {
2270 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2271 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2272 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2273 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2274 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2277 let Predicates = [HasSSE2] in {
2278 // Generic SHUFPD patterns
2279 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2280 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2281 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2282 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2283 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2286 //===----------------------------------------------------------------------===//
2287 // SSE 1 & 2 - Unpack Instructions
2288 //===----------------------------------------------------------------------===//
2290 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2291 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2292 PatFrag mem_frag, RegisterClass RC,
2293 X86MemOperand x86memop, string asm,
2295 def rr : PI<opc, MRMSrcReg,
2296 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2298 (vt (OpNode RC:$src1, RC:$src2)))],
2300 def rm : PI<opc, MRMSrcMem,
2301 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2303 (vt (OpNode RC:$src1,
2304 (mem_frag addr:$src2))))],
2308 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2309 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2310 SSEPackedSingle>, TB, VEX_4V;
2311 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2312 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2313 SSEPackedDouble>, TB, OpSize, VEX_4V;
2314 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2315 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2316 SSEPackedSingle>, TB, VEX_4V;
2317 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2318 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2319 SSEPackedDouble>, TB, OpSize, VEX_4V;
2321 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2322 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2323 SSEPackedSingle>, TB, VEX_4V;
2324 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2325 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2326 SSEPackedDouble>, TB, OpSize, VEX_4V;
2327 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2328 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2329 SSEPackedSingle>, TB, VEX_4V;
2330 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2331 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2332 SSEPackedDouble>, TB, OpSize, VEX_4V;
2334 let Constraints = "$src1 = $dst" in {
2335 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2336 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2337 SSEPackedSingle>, TB;
2338 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2339 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2340 SSEPackedDouble>, TB, OpSize;
2341 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2342 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2343 SSEPackedSingle>, TB;
2344 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2345 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2346 SSEPackedDouble>, TB, OpSize;
2347 } // Constraints = "$src1 = $dst"
2349 let Predicates = [HasAVX], AddedComplexity = 1 in {
2350 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2351 // problem is during lowering, where it's not possible to recognize the load
2352 // fold cause it has two uses through a bitcast. One use disappears at isel
2353 // time and the fold opportunity reappears.
2354 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2355 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2358 let Predicates = [HasSSE2] in {
2359 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2360 // problem is during lowering, where it's not possible to recognize the load
2361 // fold cause it has two uses through a bitcast. One use disappears at isel
2362 // time and the fold opportunity reappears.
2363 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2364 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2367 //===----------------------------------------------------------------------===//
2368 // SSE 1 & 2 - Extract Floating-Point Sign mask
2369 //===----------------------------------------------------------------------===//
2371 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2372 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2374 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2375 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2376 [(set GR32:$dst, (Int RC:$src))], IIC_DEFAULT, d>;
2377 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2378 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2379 IIC_DEFAULT, d>, REX_W;
2382 let Predicates = [HasAVX] in {
2383 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2384 "movmskps", SSEPackedSingle>, TB, VEX;
2385 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2386 "movmskpd", SSEPackedDouble>, TB,
2388 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2389 "movmskps", SSEPackedSingle>, TB, VEX;
2390 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2391 "movmskpd", SSEPackedDouble>, TB,
2394 def : Pat<(i32 (X86fgetsign FR32:$src)),
2395 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2397 def : Pat<(i64 (X86fgetsign FR32:$src)),
2398 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2400 def : Pat<(i32 (X86fgetsign FR64:$src)),
2401 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2403 def : Pat<(i64 (X86fgetsign FR64:$src)),
2404 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2408 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2409 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2410 SSEPackedSingle>, TB, VEX;
2411 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2412 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2413 SSEPackedDouble>, TB,
2415 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2416 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2417 SSEPackedSingle>, TB, VEX;
2418 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2419 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2420 SSEPackedDouble>, TB,
2424 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2425 SSEPackedSingle>, TB;
2426 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2427 SSEPackedDouble>, TB, OpSize;
2429 def : Pat<(i32 (X86fgetsign FR32:$src)),
2430 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2431 sub_ss))>, Requires<[HasSSE1]>;
2432 def : Pat<(i64 (X86fgetsign FR32:$src)),
2433 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2434 sub_ss))>, Requires<[HasSSE1]>;
2435 def : Pat<(i32 (X86fgetsign FR64:$src)),
2436 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2437 sub_sd))>, Requires<[HasSSE2]>;
2438 def : Pat<(i64 (X86fgetsign FR64:$src)),
2439 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2440 sub_sd))>, Requires<[HasSSE2]>;
2442 //===---------------------------------------------------------------------===//
2443 // SSE2 - Packed Integer Logical Instructions
2444 //===---------------------------------------------------------------------===//
2446 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2448 /// PDI_binop_rm - Simple SSE2 binary operator.
2449 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2450 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2451 X86MemOperand x86memop, bit IsCommutable = 0,
2453 let isCommutable = IsCommutable in
2454 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2455 (ins RC:$src1, RC:$src2),
2457 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2458 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2459 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2460 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2461 (ins RC:$src1, x86memop:$src2),
2463 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2464 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2465 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2466 (bitconvert (memop_frag addr:$src2)))))]>;
2468 } // ExeDomain = SSEPackedInt
2470 // These are ordered here for pattern ordering requirements with the fp versions
2472 let Predicates = [HasAVX] in {
2473 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2474 i128mem, 1, 0>, VEX_4V;
2475 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2476 i128mem, 1, 0>, VEX_4V;
2477 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2478 i128mem, 1, 0>, VEX_4V;
2479 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2480 i128mem, 0, 0>, VEX_4V;
2483 let Constraints = "$src1 = $dst" in {
2484 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2486 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2488 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2490 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2492 } // Constraints = "$src1 = $dst"
2494 let Predicates = [HasAVX2] in {
2495 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2496 i256mem, 1, 0>, VEX_4V;
2497 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2498 i256mem, 1, 0>, VEX_4V;
2499 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2500 i256mem, 1, 0>, VEX_4V;
2501 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2502 i256mem, 0, 0>, VEX_4V;
2505 //===----------------------------------------------------------------------===//
2506 // SSE 1 & 2 - Logical Instructions
2507 //===----------------------------------------------------------------------===//
2509 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2511 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2513 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2514 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2516 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2517 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2519 let Constraints = "$src1 = $dst" in {
2520 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2521 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2523 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2524 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2528 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2529 let mayLoad = 0 in {
2530 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2531 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2532 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2535 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2536 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2538 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2540 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2542 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2543 // are all promoted to v2i64, and the patterns are covered by the int
2544 // version. This is needed in SSE only, because v2i64 isn't supported on
2545 // SSE1, but only on SSE2.
2546 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2547 !strconcat(OpcodeStr, "ps"), f128mem, [],
2548 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2549 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2551 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2552 !strconcat(OpcodeStr, "pd"), f128mem,
2553 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2554 (bc_v2i64 (v2f64 VR128:$src2))))],
2555 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2556 (memopv2i64 addr:$src2)))], 0>,
2558 let Constraints = "$src1 = $dst" in {
2559 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2560 !strconcat(OpcodeStr, "ps"), f128mem,
2561 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2562 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2563 (memopv2i64 addr:$src2)))]>, TB;
2565 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2566 !strconcat(OpcodeStr, "pd"), f128mem,
2567 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2568 (bc_v2i64 (v2f64 VR128:$src2))))],
2569 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2570 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2574 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2576 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2578 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2579 !strconcat(OpcodeStr, "ps"), f256mem,
2580 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2581 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2582 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2584 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2585 !strconcat(OpcodeStr, "pd"), f256mem,
2586 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2587 (bc_v4i64 (v4f64 VR256:$src2))))],
2588 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2589 (memopv4i64 addr:$src2)))], 0>,
2593 // AVX 256-bit packed logical ops forms
2594 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2595 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2596 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2597 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2599 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2600 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2601 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2602 let isCommutable = 0 in
2603 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2605 //===----------------------------------------------------------------------===//
2606 // SSE 1 & 2 - Arithmetic Instructions
2607 //===----------------------------------------------------------------------===//
2609 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2612 /// In addition, we also have a special variant of the scalar form here to
2613 /// represent the associated intrinsic operation. This form is unlike the
2614 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2615 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2617 /// These three forms can each be reg+reg or reg+mem.
2620 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2622 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2624 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2625 OpNode, FR32, f32mem, Is2Addr>, XS;
2626 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2627 OpNode, FR64, f64mem, Is2Addr>, XD;
2630 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2632 let mayLoad = 0 in {
2633 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2634 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2635 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2636 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2640 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2642 let mayLoad = 0 in {
2643 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2644 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2645 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2646 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2650 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2652 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2653 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2654 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2655 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2658 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2660 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2661 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2662 SSEPackedSingle, Is2Addr>, TB;
2664 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2665 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2666 SSEPackedDouble, Is2Addr>, TB, OpSize;
2669 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2670 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2671 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2672 SSEPackedSingle, 0>, TB;
2674 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2675 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2676 SSEPackedDouble, 0>, TB, OpSize;
2679 // Binary Arithmetic instructions
2680 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2681 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2682 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2683 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2684 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2685 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2686 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2687 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2689 let isCommutable = 0 in {
2690 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2691 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2692 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2693 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2694 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2695 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2696 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2697 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2698 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2699 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2700 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2701 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2702 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2703 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2704 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2705 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2706 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2707 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2708 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2709 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2712 let Constraints = "$src1 = $dst" in {
2713 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2714 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2715 basic_sse12_fp_binop_s_int<0x58, "add">;
2716 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2717 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2718 basic_sse12_fp_binop_s_int<0x59, "mul">;
2720 let isCommutable = 0 in {
2721 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2722 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2723 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2724 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2725 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2726 basic_sse12_fp_binop_s_int<0x5E, "div">;
2727 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2728 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2729 basic_sse12_fp_binop_s_int<0x5F, "max">,
2730 basic_sse12_fp_binop_p_int<0x5F, "max">;
2731 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2732 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2733 basic_sse12_fp_binop_s_int<0x5D, "min">,
2734 basic_sse12_fp_binop_p_int<0x5D, "min">;
2739 /// In addition, we also have a special variant of the scalar form here to
2740 /// represent the associated intrinsic operation. This form is unlike the
2741 /// plain scalar form, in that it takes an entire vector (instead of a
2742 /// scalar) and leaves the top elements undefined.
2744 /// And, we have a special variant form for a full-vector intrinsic form.
2746 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2747 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2748 SDNode OpNode, Intrinsic F32Int> {
2749 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2750 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2751 [(set FR32:$dst, (OpNode FR32:$src))]>;
2752 // For scalar unary operations, fold a load into the operation
2753 // only in OptForSize mode. It eliminates an instruction, but it also
2754 // eliminates a whole-register clobber (the load), so it introduces a
2755 // partial register update condition.
2756 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2757 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2758 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2759 Requires<[HasSSE1, OptForSize]>;
2760 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2761 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2762 [(set VR128:$dst, (F32Int VR128:$src))]>;
2763 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2764 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2765 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2768 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2769 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2770 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2771 !strconcat(OpcodeStr,
2772 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2774 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2775 !strconcat(OpcodeStr,
2776 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2777 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2778 (ins VR128:$src1, ssmem:$src2),
2779 !strconcat(OpcodeStr,
2780 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2783 /// sse1_fp_unop_p - SSE1 unops in packed form.
2784 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2785 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2786 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2787 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2788 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2789 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2790 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2793 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2794 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2795 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2796 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2797 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2798 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2799 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2800 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2803 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2804 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2805 Intrinsic V4F32Int> {
2806 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2807 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2808 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2809 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2810 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2811 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2814 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2815 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2816 Intrinsic V4F32Int> {
2817 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2818 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2819 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2820 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2821 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2822 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2825 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2826 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2827 SDNode OpNode, Intrinsic F64Int> {
2828 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2829 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2830 [(set FR64:$dst, (OpNode FR64:$src))]>;
2831 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2832 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2833 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2834 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2835 Requires<[HasSSE2, OptForSize]>;
2836 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2837 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2838 [(set VR128:$dst, (F64Int VR128:$src))]>;
2839 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2840 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2841 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2844 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2845 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2846 let neverHasSideEffects = 1 in {
2847 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2848 !strconcat(OpcodeStr,
2849 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2851 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2852 !strconcat(OpcodeStr,
2853 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2855 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2856 (ins VR128:$src1, sdmem:$src2),
2857 !strconcat(OpcodeStr,
2858 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2861 /// sse2_fp_unop_p - SSE2 unops in vector forms.
2862 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
2864 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2865 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2866 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
2867 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2868 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2869 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
2872 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
2873 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2874 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2875 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2876 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
2877 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2878 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2879 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
2882 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
2883 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2884 Intrinsic V2F64Int> {
2885 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2886 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2887 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
2888 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2889 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2890 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
2893 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
2894 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2895 Intrinsic V2F64Int> {
2896 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2897 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2898 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
2899 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2900 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2901 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
2904 let Predicates = [HasAVX] in {
2906 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
2907 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
2909 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
2910 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
2911 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2912 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2913 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
2914 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
2915 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
2916 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
2919 // Reciprocal approximations. Note that these typically require refinement
2920 // in order to obtain suitable precision.
2921 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
2922 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
2923 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
2924 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
2925 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
2927 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
2928 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
2929 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
2930 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
2931 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
2934 let AddedComplexity = 1 in {
2935 def : Pat<(f32 (fsqrt FR32:$src)),
2936 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2937 def : Pat<(f32 (fsqrt (load addr:$src))),
2938 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2939 Requires<[HasAVX, OptForSize]>;
2940 def : Pat<(f64 (fsqrt FR64:$src)),
2941 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
2942 def : Pat<(f64 (fsqrt (load addr:$src))),
2943 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
2944 Requires<[HasAVX, OptForSize]>;
2946 def : Pat<(f32 (X86frsqrt FR32:$src)),
2947 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2948 def : Pat<(f32 (X86frsqrt (load addr:$src))),
2949 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2950 Requires<[HasAVX, OptForSize]>;
2952 def : Pat<(f32 (X86frcp FR32:$src)),
2953 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2954 def : Pat<(f32 (X86frcp (load addr:$src))),
2955 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2956 Requires<[HasAVX, OptForSize]>;
2959 let Predicates = [HasAVX], AddedComplexity = 1 in {
2960 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
2961 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2962 (VSQRTSSr (f32 (IMPLICIT_DEF)),
2963 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2965 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
2966 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2968 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
2969 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
2970 (VSQRTSDr (f64 (IMPLICIT_DEF)),
2971 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
2973 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
2974 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
2976 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
2977 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2978 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
2979 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2981 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
2982 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2984 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
2985 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2986 (VRCPSSr (f32 (IMPLICIT_DEF)),
2987 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2989 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
2990 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2994 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
2995 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
2996 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
2997 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
2998 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
2999 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3001 // Reciprocal approximations. Note that these typically require refinement
3002 // in order to obtain suitable precision.
3003 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3004 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3005 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3006 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3007 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3008 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3010 // There is no f64 version of the reciprocal approximation instructions.
3012 //===----------------------------------------------------------------------===//
3013 // SSE 1 & 2 - Non-temporal stores
3014 //===----------------------------------------------------------------------===//
3016 let AddedComplexity = 400 in { // Prefer non-temporal versions
3017 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3018 (ins f128mem:$dst, VR128:$src),
3019 "movntps\t{$src, $dst|$dst, $src}",
3020 [(alignednontemporalstore (v4f32 VR128:$src),
3022 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3023 (ins f128mem:$dst, VR128:$src),
3024 "movntpd\t{$src, $dst|$dst, $src}",
3025 [(alignednontemporalstore (v2f64 VR128:$src),
3028 let ExeDomain = SSEPackedInt in
3029 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3030 (ins f128mem:$dst, VR128:$src),
3031 "movntdq\t{$src, $dst|$dst, $src}",
3032 [(alignednontemporalstore (v2i64 VR128:$src),
3035 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3036 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3038 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3039 (ins f256mem:$dst, VR256:$src),
3040 "movntps\t{$src, $dst|$dst, $src}",
3041 [(alignednontemporalstore (v8f32 VR256:$src),
3043 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3044 (ins f256mem:$dst, VR256:$src),
3045 "movntpd\t{$src, $dst|$dst, $src}",
3046 [(alignednontemporalstore (v4f64 VR256:$src),
3048 let ExeDomain = SSEPackedInt in
3049 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3050 (ins f256mem:$dst, VR256:$src),
3051 "movntdq\t{$src, $dst|$dst, $src}",
3052 [(alignednontemporalstore (v4i64 VR256:$src),
3056 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3057 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3058 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3059 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3060 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3061 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3063 let AddedComplexity = 400 in { // Prefer non-temporal versions
3064 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3065 "movntps\t{$src, $dst|$dst, $src}",
3066 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3067 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3068 "movntpd\t{$src, $dst|$dst, $src}",
3069 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3071 let ExeDomain = SSEPackedInt in
3072 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3073 "movntdq\t{$src, $dst|$dst, $src}",
3074 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3076 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3077 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3079 // There is no AVX form for instructions below this point
3080 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3081 "movnti{l}\t{$src, $dst|$dst, $src}",
3082 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3083 TB, Requires<[HasSSE2]>;
3084 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3085 "movnti{q}\t{$src, $dst|$dst, $src}",
3086 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3087 TB, Requires<[HasSSE2]>;
3090 //===----------------------------------------------------------------------===//
3091 // SSE 1 & 2 - Prefetch and memory fence
3092 //===----------------------------------------------------------------------===//
3094 // Prefetch intrinsic.
3095 let Predicates = [HasSSE1] in {
3096 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3097 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3098 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3099 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3100 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3101 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3102 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3103 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3107 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3108 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3109 TB, Requires<[HasSSE2]>;
3111 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3112 // was introduced with SSE2, it's backward compatible.
3113 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3115 // Load, store, and memory fence
3116 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3117 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3118 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3119 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3120 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3121 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3123 def : Pat<(X86SFence), (SFENCE)>;
3124 def : Pat<(X86LFence), (LFENCE)>;
3125 def : Pat<(X86MFence), (MFENCE)>;
3127 //===----------------------------------------------------------------------===//
3128 // SSE 1 & 2 - Load/Store XCSR register
3129 //===----------------------------------------------------------------------===//
3131 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3132 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3133 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3134 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3136 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3137 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3138 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3139 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3141 //===---------------------------------------------------------------------===//
3142 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3143 //===---------------------------------------------------------------------===//
3145 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3147 let neverHasSideEffects = 1 in {
3148 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3149 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3150 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3151 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3153 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3154 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3155 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3156 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3159 let isCodeGenOnly = 1 in {
3160 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3161 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3162 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3163 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3164 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3165 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3166 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3167 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3170 let canFoldAsLoad = 1, mayLoad = 1 in {
3171 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3172 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3173 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3174 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3175 let Predicates = [HasAVX] in {
3176 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3177 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3178 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3179 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3183 let mayStore = 1 in {
3184 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3185 (ins i128mem:$dst, VR128:$src),
3186 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3187 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3188 (ins i256mem:$dst, VR256:$src),
3189 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3190 let Predicates = [HasAVX] in {
3191 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3192 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3193 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3194 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3198 let neverHasSideEffects = 1 in
3199 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3200 "movdqa\t{$src, $dst|$dst, $src}", []>;
3202 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3203 "movdqu\t{$src, $dst|$dst, $src}",
3204 []>, XS, Requires<[HasSSE2]>;
3207 let isCodeGenOnly = 1 in {
3208 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3209 "movdqa\t{$src, $dst|$dst, $src}", []>;
3211 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3212 "movdqu\t{$src, $dst|$dst, $src}",
3213 []>, XS, Requires<[HasSSE2]>;
3216 let canFoldAsLoad = 1, mayLoad = 1 in {
3217 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3218 "movdqa\t{$src, $dst|$dst, $src}",
3219 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3220 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3221 "movdqu\t{$src, $dst|$dst, $src}",
3222 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3223 XS, Requires<[HasSSE2]>;
3226 let mayStore = 1 in {
3227 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3228 "movdqa\t{$src, $dst|$dst, $src}",
3229 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3230 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3231 "movdqu\t{$src, $dst|$dst, $src}",
3232 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3233 XS, Requires<[HasSSE2]>;
3236 // Intrinsic forms of MOVDQU load and store
3237 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3238 "vmovdqu\t{$src, $dst|$dst, $src}",
3239 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3240 XS, VEX, Requires<[HasAVX]>;
3242 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3243 "movdqu\t{$src, $dst|$dst, $src}",
3244 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3245 XS, Requires<[HasSSE2]>;
3247 } // ExeDomain = SSEPackedInt
3249 let Predicates = [HasAVX] in {
3250 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3251 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3254 //===---------------------------------------------------------------------===//
3255 // SSE2 - Packed Integer Arithmetic Instructions
3256 //===---------------------------------------------------------------------===//
3258 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3260 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3261 RegisterClass RC, PatFrag memop_frag,
3262 X86MemOperand x86memop, bit IsCommutable = 0,
3264 let isCommutable = IsCommutable in
3265 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3266 (ins RC:$src1, RC:$src2),
3268 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3269 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3270 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3271 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3272 (ins RC:$src1, x86memop:$src2),
3274 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3275 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3276 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3279 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3280 string OpcodeStr, SDNode OpNode,
3281 SDNode OpNode2, RegisterClass RC,
3282 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3284 // src2 is always 128-bit
3285 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3286 (ins RC:$src1, VR128:$src2),
3288 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3289 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3290 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))]>;
3291 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3292 (ins RC:$src1, i128mem:$src2),
3294 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3295 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3296 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3297 (bc_frag (memopv2i64 addr:$src2)))))]>;
3298 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3299 (ins RC:$src1, i32i8imm:$src2),
3301 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3302 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3303 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))]>;
3306 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3307 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3308 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3309 PatFrag memop_frag, X86MemOperand x86memop,
3310 bit IsCommutable = 0, bit Is2Addr = 1> {
3311 let isCommutable = IsCommutable in
3312 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3313 (ins RC:$src1, RC:$src2),
3315 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3316 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3317 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3318 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3319 (ins RC:$src1, x86memop:$src2),
3321 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3322 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3323 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3324 (bitconvert (memop_frag addr:$src2)))))]>;
3326 } // ExeDomain = SSEPackedInt
3328 // 128-bit Integer Arithmetic
3330 let Predicates = [HasAVX] in {
3331 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3332 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3333 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3334 i128mem, 1, 0>, VEX_4V;
3335 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3336 i128mem, 1, 0>, VEX_4V;
3337 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3338 i128mem, 1, 0>, VEX_4V;
3339 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3340 i128mem, 1, 0>, VEX_4V;
3341 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3342 i128mem, 0, 0>, VEX_4V;
3343 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3344 i128mem, 0, 0>, VEX_4V;
3345 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3346 i128mem, 0, 0>, VEX_4V;
3347 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3348 i128mem, 0, 0>, VEX_4V;
3349 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3350 memopv2i64, i128mem, 1, 0>, VEX_4V;
3353 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3354 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3355 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3356 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3357 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3358 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3359 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3360 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3361 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3362 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3363 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3364 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3365 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3366 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3367 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3368 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3369 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3370 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3371 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3372 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3373 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3374 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3375 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3376 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3377 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3378 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3379 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3380 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3381 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3382 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3383 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3384 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3385 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3386 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3387 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3388 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3391 let Predicates = [HasAVX2] in {
3392 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3393 i256mem, 1, 0>, VEX_4V;
3394 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3395 i256mem, 1, 0>, VEX_4V;
3396 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3397 i256mem, 1, 0>, VEX_4V;
3398 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3399 i256mem, 1, 0>, VEX_4V;
3400 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3401 i256mem, 1, 0>, VEX_4V;
3402 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3403 i256mem, 0, 0>, VEX_4V;
3404 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3405 i256mem, 0, 0>, VEX_4V;
3406 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3407 i256mem, 0, 0>, VEX_4V;
3408 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3409 i256mem, 0, 0>, VEX_4V;
3410 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3411 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3414 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3415 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3416 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3417 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3418 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3419 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3420 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3421 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3422 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3423 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3424 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3425 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3426 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3427 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3428 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3429 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3430 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3431 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3432 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3433 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3434 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3435 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3436 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3437 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3438 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3439 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3440 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3441 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3442 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3443 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3444 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3445 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3446 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3447 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3448 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3449 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3452 let Constraints = "$src1 = $dst" in {
3453 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3455 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3457 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3459 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3461 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3463 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3465 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3467 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3469 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3471 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3472 memopv2i64, i128mem, 1>;
3475 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3476 VR128, memopv2i64, i128mem>;
3477 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3478 VR128, memopv2i64, i128mem>;
3479 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3480 VR128, memopv2i64, i128mem>;
3481 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3482 VR128, memopv2i64, i128mem>;
3483 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3484 VR128, memopv2i64, i128mem, 1>;
3485 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3486 VR128, memopv2i64, i128mem, 1>;
3487 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3488 VR128, memopv2i64, i128mem, 1>;
3489 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3490 VR128, memopv2i64, i128mem, 1>;
3491 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3492 VR128, memopv2i64, i128mem, 1>;
3493 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3494 VR128, memopv2i64, i128mem, 1>;
3495 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3496 VR128, memopv2i64, i128mem, 1>;
3497 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3498 VR128, memopv2i64, i128mem, 1>;
3499 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3500 VR128, memopv2i64, i128mem, 1>;
3501 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3502 VR128, memopv2i64, i128mem, 1>;
3503 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3504 VR128, memopv2i64, i128mem, 1>;
3505 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3506 VR128, memopv2i64, i128mem, 1>;
3507 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3508 VR128, memopv2i64, i128mem, 1>;
3509 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3510 VR128, memopv2i64, i128mem, 1>;
3512 } // Constraints = "$src1 = $dst"
3514 //===---------------------------------------------------------------------===//
3515 // SSE2 - Packed Integer Logical Instructions
3516 //===---------------------------------------------------------------------===//
3518 let Predicates = [HasAVX] in {
3519 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3520 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3521 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3522 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3523 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3524 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3526 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3527 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3528 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3529 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3530 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3531 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3533 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3534 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3535 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3536 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3538 let ExeDomain = SSEPackedInt in {
3539 // 128-bit logical shifts.
3540 def VPSLLDQri : PDIi8<0x73, MRM7r,
3541 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3542 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3544 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3546 def VPSRLDQri : PDIi8<0x73, MRM3r,
3547 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3548 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3550 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3552 // PSRADQri doesn't exist in SSE[1-3].
3554 } // Predicates = [HasAVX]
3556 let Predicates = [HasAVX2] in {
3557 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3558 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3559 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3560 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3561 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3562 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3564 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3565 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3566 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3567 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3568 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3569 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3571 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3572 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3573 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3574 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3576 let ExeDomain = SSEPackedInt in {
3577 // 256-bit logical shifts.
3578 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3579 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3580 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3582 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3584 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3585 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3586 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3588 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3590 // PSRADQYri doesn't exist in SSE[1-3].
3592 } // Predicates = [HasAVX2]
3594 let Constraints = "$src1 = $dst" in {
3595 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3596 VR128, v8i16, v8i16, bc_v8i16>;
3597 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3598 VR128, v4i32, v4i32, bc_v4i32>;
3599 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3600 VR128, v2i64, v2i64, bc_v2i64>;
3602 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3603 VR128, v8i16, v8i16, bc_v8i16>;
3604 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3605 VR128, v4i32, v4i32, bc_v4i32>;
3606 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3607 VR128, v2i64, v2i64, bc_v2i64>;
3609 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3610 VR128, v8i16, v8i16, bc_v8i16>;
3611 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3612 VR128, v4i32, v4i32, bc_v4i32>;
3614 let ExeDomain = SSEPackedInt in {
3615 // 128-bit logical shifts.
3616 def PSLLDQri : PDIi8<0x73, MRM7r,
3617 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3618 "pslldq\t{$src2, $dst|$dst, $src2}",
3620 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3621 def PSRLDQri : PDIi8<0x73, MRM3r,
3622 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3623 "psrldq\t{$src2, $dst|$dst, $src2}",
3625 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3626 // PSRADQri doesn't exist in SSE[1-3].
3628 } // Constraints = "$src1 = $dst"
3630 let Predicates = [HasAVX] in {
3631 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3632 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3633 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3634 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3635 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3636 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3638 // Shift up / down and insert zero's.
3639 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3640 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3641 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3642 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3645 let Predicates = [HasAVX2] in {
3646 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3647 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3648 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3649 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3652 let Predicates = [HasSSE2] in {
3653 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3654 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3655 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3656 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3657 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3658 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3660 // Shift up / down and insert zero's.
3661 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3662 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3663 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3664 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3667 //===---------------------------------------------------------------------===//
3668 // SSE2 - Packed Integer Comparison Instructions
3669 //===---------------------------------------------------------------------===//
3671 let Predicates = [HasAVX] in {
3672 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
3673 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3674 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
3675 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3676 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
3677 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3678 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
3679 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3680 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
3681 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3682 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
3683 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3686 let Predicates = [HasAVX2] in {
3687 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
3688 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3689 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
3690 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3691 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
3692 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3693 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
3694 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3695 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
3696 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3697 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
3698 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3701 let Constraints = "$src1 = $dst" in {
3702 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
3703 VR128, memopv2i64, i128mem, 1>;
3704 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
3705 VR128, memopv2i64, i128mem, 1>;
3706 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
3707 VR128, memopv2i64, i128mem, 1>;
3708 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
3709 VR128, memopv2i64, i128mem>;
3710 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
3711 VR128, memopv2i64, i128mem>;
3712 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
3713 VR128, memopv2i64, i128mem>;
3714 } // Constraints = "$src1 = $dst"
3716 //===---------------------------------------------------------------------===//
3717 // SSE2 - Packed Integer Pack Instructions
3718 //===---------------------------------------------------------------------===//
3720 let Predicates = [HasAVX] in {
3721 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3722 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3723 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3724 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3725 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3726 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3729 let Predicates = [HasAVX2] in {
3730 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
3731 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3732 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
3733 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3734 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
3735 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3738 let Constraints = "$src1 = $dst" in {
3739 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3740 VR128, memopv2i64, i128mem>;
3741 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3742 VR128, memopv2i64, i128mem>;
3743 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
3744 VR128, memopv2i64, i128mem>;
3745 } // Constraints = "$src1 = $dst"
3747 //===---------------------------------------------------------------------===//
3748 // SSE2 - Packed Integer Shuffle Instructions
3749 //===---------------------------------------------------------------------===//
3751 let ExeDomain = SSEPackedInt in {
3752 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
3753 def ri : Ii8<0x70, MRMSrcReg,
3754 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3755 !strconcat(OpcodeStr,
3756 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3757 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))]>;
3758 def mi : Ii8<0x70, MRMSrcMem,
3759 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3760 !strconcat(OpcodeStr,
3761 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3763 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
3764 (i8 imm:$src2))))]>;
3767 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
3768 def Yri : Ii8<0x70, MRMSrcReg,
3769 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
3770 !strconcat(OpcodeStr,
3771 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3772 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
3773 def Ymi : Ii8<0x70, MRMSrcMem,
3774 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
3775 !strconcat(OpcodeStr,
3776 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3778 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
3779 (i8 imm:$src2))))]>;
3781 } // ExeDomain = SSEPackedInt
3783 let Predicates = [HasAVX] in {
3784 let AddedComplexity = 5 in
3785 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
3787 // SSE2 with ImmT == Imm8 and XS prefix.
3788 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
3790 // SSE2 with ImmT == Imm8 and XD prefix.
3791 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
3793 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3794 (VPSHUFDmi addr:$src1, imm:$imm)>;
3795 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3796 (VPSHUFDri VR128:$src1, imm:$imm)>;
3799 let Predicates = [HasAVX2] in {
3800 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
3801 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
3802 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
3805 let Predicates = [HasSSE2] in {
3806 let AddedComplexity = 5 in
3807 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
3809 // SSE2 with ImmT == Imm8 and XS prefix.
3810 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
3812 // SSE2 with ImmT == Imm8 and XD prefix.
3813 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
3815 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3816 (PSHUFDmi addr:$src1, imm:$imm)>;
3817 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3818 (PSHUFDri VR128:$src1, imm:$imm)>;
3821 //===---------------------------------------------------------------------===//
3822 // SSE2 - Packed Integer Unpack Instructions
3823 //===---------------------------------------------------------------------===//
3825 let ExeDomain = SSEPackedInt in {
3826 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3827 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3828 def rr : PDI<opc, MRMSrcReg,
3829 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3831 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3832 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3833 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
3834 def rm : PDI<opc, MRMSrcMem,
3835 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3837 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3838 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3839 [(set VR128:$dst, (OpNode VR128:$src1,
3840 (bc_frag (memopv2i64
3844 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
3845 SDNode OpNode, PatFrag bc_frag> {
3846 def Yrr : PDI<opc, MRMSrcReg,
3847 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
3848 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3849 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
3850 def Yrm : PDI<opc, MRMSrcMem,
3851 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
3852 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3853 [(set VR256:$dst, (OpNode VR256:$src1,
3854 (bc_frag (memopv4i64 addr:$src2))))]>;
3857 let Predicates = [HasAVX] in {
3858 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
3859 bc_v16i8, 0>, VEX_4V;
3860 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
3861 bc_v8i16, 0>, VEX_4V;
3862 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
3863 bc_v4i32, 0>, VEX_4V;
3864 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
3865 bc_v2i64, 0>, VEX_4V;
3867 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
3868 bc_v16i8, 0>, VEX_4V;
3869 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
3870 bc_v8i16, 0>, VEX_4V;
3871 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
3872 bc_v4i32, 0>, VEX_4V;
3873 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
3874 bc_v2i64, 0>, VEX_4V;
3877 let Predicates = [HasAVX2] in {
3878 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
3880 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
3882 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
3884 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
3887 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
3889 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
3891 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
3893 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
3897 let Constraints = "$src1 = $dst" in {
3898 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
3900 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
3902 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
3904 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
3907 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
3909 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
3911 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
3913 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
3916 } // ExeDomain = SSEPackedInt
3918 // Patterns for using AVX1 instructions with integer vectors
3919 // Here to give AVX2 priority
3920 let Predicates = [HasAVX] in {
3921 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
3922 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
3923 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
3924 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
3925 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
3926 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
3927 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
3928 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
3930 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
3931 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
3932 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
3933 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
3934 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
3935 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
3936 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
3937 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
3940 //===---------------------------------------------------------------------===//
3941 // SSE2 - Packed Integer Extract and Insert
3942 //===---------------------------------------------------------------------===//
3944 let ExeDomain = SSEPackedInt in {
3945 multiclass sse2_pinsrw<bit Is2Addr = 1> {
3946 def rri : Ii8<0xC4, MRMSrcReg,
3947 (outs VR128:$dst), (ins VR128:$src1,
3948 GR32:$src2, i32i8imm:$src3),
3950 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3951 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3953 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
3954 def rmi : Ii8<0xC4, MRMSrcMem,
3955 (outs VR128:$dst), (ins VR128:$src1,
3956 i16mem:$src2, i32i8imm:$src3),
3958 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3959 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3961 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
3966 let Predicates = [HasAVX] in
3967 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
3968 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3969 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3970 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3971 imm:$src2))]>, TB, OpSize, VEX;
3972 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
3973 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3974 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3975 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3979 let Predicates = [HasAVX] in {
3980 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
3981 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
3982 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
3983 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
3984 []>, TB, OpSize, VEX_4V;
3987 let Constraints = "$src1 = $dst" in
3988 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
3990 } // ExeDomain = SSEPackedInt
3992 //===---------------------------------------------------------------------===//
3993 // SSE2 - Packed Mask Creation
3994 //===---------------------------------------------------------------------===//
3996 let ExeDomain = SSEPackedInt in {
3998 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3999 "pmovmskb\t{$src, $dst|$dst, $src}",
4000 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4001 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4002 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4004 let Predicates = [HasAVX2] in {
4005 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4006 "pmovmskb\t{$src, $dst|$dst, $src}",
4007 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4008 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4009 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4012 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4013 "pmovmskb\t{$src, $dst|$dst, $src}",
4014 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4016 } // ExeDomain = SSEPackedInt
4018 //===---------------------------------------------------------------------===//
4019 // SSE2 - Conditional Store
4020 //===---------------------------------------------------------------------===//
4022 let ExeDomain = SSEPackedInt in {
4025 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4026 (ins VR128:$src, VR128:$mask),
4027 "maskmovdqu\t{$mask, $src|$src, $mask}",
4028 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4030 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4031 (ins VR128:$src, VR128:$mask),
4032 "maskmovdqu\t{$mask, $src|$src, $mask}",
4033 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4036 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4037 "maskmovdqu\t{$mask, $src|$src, $mask}",
4038 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4040 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4041 "maskmovdqu\t{$mask, $src|$src, $mask}",
4042 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4044 } // ExeDomain = SSEPackedInt
4046 //===---------------------------------------------------------------------===//
4047 // SSE2 - Move Doubleword
4048 //===---------------------------------------------------------------------===//
4050 //===---------------------------------------------------------------------===//
4051 // Move Int Doubleword to Packed Double Int
4053 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4054 "movd\t{$src, $dst|$dst, $src}",
4056 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4057 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4058 "movd\t{$src, $dst|$dst, $src}",
4060 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4062 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4063 "mov{d|q}\t{$src, $dst|$dst, $src}",
4065 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4066 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4067 "mov{d|q}\t{$src, $dst|$dst, $src}",
4068 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4070 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4071 "movd\t{$src, $dst|$dst, $src}",
4073 (v4i32 (scalar_to_vector GR32:$src)))]>;
4074 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4075 "movd\t{$src, $dst|$dst, $src}",
4077 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4078 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4079 "mov{d|q}\t{$src, $dst|$dst, $src}",
4081 (v2i64 (scalar_to_vector GR64:$src)))]>;
4082 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4083 "mov{d|q}\t{$src, $dst|$dst, $src}",
4084 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4086 //===---------------------------------------------------------------------===//
4087 // Move Int Doubleword to Single Scalar
4089 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4090 "movd\t{$src, $dst|$dst, $src}",
4091 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4093 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4094 "movd\t{$src, $dst|$dst, $src}",
4095 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4097 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4098 "movd\t{$src, $dst|$dst, $src}",
4099 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4101 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4102 "movd\t{$src, $dst|$dst, $src}",
4103 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4105 //===---------------------------------------------------------------------===//
4106 // Move Packed Doubleword Int to Packed Double Int
4108 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4109 "movd\t{$src, $dst|$dst, $src}",
4110 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4112 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4113 (ins i32mem:$dst, VR128:$src),
4114 "movd\t{$src, $dst|$dst, $src}",
4115 [(store (i32 (vector_extract (v4i32 VR128:$src),
4116 (iPTR 0))), addr:$dst)]>, VEX;
4117 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4118 "movd\t{$src, $dst|$dst, $src}",
4119 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4121 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4122 "movd\t{$src, $dst|$dst, $src}",
4123 [(store (i32 (vector_extract (v4i32 VR128:$src),
4124 (iPTR 0))), addr:$dst)]>;
4126 //===---------------------------------------------------------------------===//
4127 // Move Packed Doubleword Int first element to Doubleword Int
4129 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4130 "mov{d|q}\t{$src, $dst|$dst, $src}",
4131 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4133 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4135 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4136 "mov{d|q}\t{$src, $dst|$dst, $src}",
4137 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4140 //===---------------------------------------------------------------------===//
4141 // Bitcast FR64 <-> GR64
4143 let Predicates = [HasAVX] in
4144 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4145 "vmovq\t{$src, $dst|$dst, $src}",
4146 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4148 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4149 "mov{d|q}\t{$src, $dst|$dst, $src}",
4150 [(set GR64:$dst, (bitconvert FR64:$src))]>, VEX;
4151 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4152 "movq\t{$src, $dst|$dst, $src}",
4153 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>,
4156 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4157 "movq\t{$src, $dst|$dst, $src}",
4158 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4159 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4160 "mov{d|q}\t{$src, $dst|$dst, $src}",
4161 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4162 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4163 "movq\t{$src, $dst|$dst, $src}",
4164 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4166 //===---------------------------------------------------------------------===//
4167 // Move Scalar Single to Double Int
4169 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4170 "movd\t{$src, $dst|$dst, $src}",
4171 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4172 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4173 "movd\t{$src, $dst|$dst, $src}",
4174 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4175 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4176 "movd\t{$src, $dst|$dst, $src}",
4177 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4178 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4179 "movd\t{$src, $dst|$dst, $src}",
4180 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4182 //===---------------------------------------------------------------------===//
4183 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4185 let AddedComplexity = 15 in {
4186 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4187 "movd\t{$src, $dst|$dst, $src}",
4188 [(set VR128:$dst, (v4i32 (X86vzmovl
4189 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4191 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4192 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4193 [(set VR128:$dst, (v2i64 (X86vzmovl
4194 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4197 let AddedComplexity = 15 in {
4198 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4199 "movd\t{$src, $dst|$dst, $src}",
4200 [(set VR128:$dst, (v4i32 (X86vzmovl
4201 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4202 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4203 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4204 [(set VR128:$dst, (v2i64 (X86vzmovl
4205 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4208 let AddedComplexity = 20 in {
4209 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4210 "movd\t{$src, $dst|$dst, $src}",
4212 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4213 (loadi32 addr:$src))))))]>,
4215 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4216 "movd\t{$src, $dst|$dst, $src}",
4218 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4219 (loadi32 addr:$src))))))]>;
4222 let Predicates = [HasAVX] in {
4223 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4224 let AddedComplexity = 20 in {
4225 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4226 (VMOVZDI2PDIrm addr:$src)>;
4227 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4228 (VMOVZDI2PDIrm addr:$src)>;
4230 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4231 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4232 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4233 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4234 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4235 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4236 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4239 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4240 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4241 (MOVZDI2PDIrm addr:$src)>;
4242 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4243 (MOVZDI2PDIrm addr:$src)>;
4246 // These are the correct encodings of the instructions so that we know how to
4247 // read correct assembly, even though we continue to emit the wrong ones for
4248 // compatibility with Darwin's buggy assembler.
4249 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4250 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4251 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4252 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4253 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4254 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4255 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4256 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4257 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4258 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4259 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4260 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4262 //===---------------------------------------------------------------------===//
4263 // SSE2 - Move Quadword
4264 //===---------------------------------------------------------------------===//
4266 //===---------------------------------------------------------------------===//
4267 // Move Quadword Int to Packed Quadword Int
4269 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4270 "vmovq\t{$src, $dst|$dst, $src}",
4272 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4273 VEX, Requires<[HasAVX]>;
4274 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4275 "movq\t{$src, $dst|$dst, $src}",
4277 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4278 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4280 //===---------------------------------------------------------------------===//
4281 // Move Packed Quadword Int to Quadword Int
4283 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4284 "movq\t{$src, $dst|$dst, $src}",
4285 [(store (i64 (vector_extract (v2i64 VR128:$src),
4286 (iPTR 0))), addr:$dst)]>, VEX;
4287 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4288 "movq\t{$src, $dst|$dst, $src}",
4289 [(store (i64 (vector_extract (v2i64 VR128:$src),
4290 (iPTR 0))), addr:$dst)]>;
4292 //===---------------------------------------------------------------------===//
4293 // Store / copy lower 64-bits of a XMM register.
4295 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4296 "movq\t{$src, $dst|$dst, $src}",
4297 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4298 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4299 "movq\t{$src, $dst|$dst, $src}",
4300 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4302 let AddedComplexity = 20 in
4303 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4304 "vmovq\t{$src, $dst|$dst, $src}",
4306 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4307 (loadi64 addr:$src))))))]>,
4308 XS, VEX, Requires<[HasAVX]>;
4310 let AddedComplexity = 20 in
4311 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4312 "movq\t{$src, $dst|$dst, $src}",
4314 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4315 (loadi64 addr:$src))))))]>,
4316 XS, Requires<[HasSSE2]>;
4318 let Predicates = [HasAVX], AddedComplexity = 20 in {
4319 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4320 (VMOVZQI2PQIrm addr:$src)>;
4321 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4322 (VMOVZQI2PQIrm addr:$src)>;
4323 def : Pat<(v2i64 (X86vzload addr:$src)),
4324 (VMOVZQI2PQIrm addr:$src)>;
4327 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4328 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4329 (MOVZQI2PQIrm addr:$src)>;
4330 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4331 (MOVZQI2PQIrm addr:$src)>;
4332 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4335 let Predicates = [HasAVX] in {
4336 def : Pat<(v4i64 (X86vzload addr:$src)),
4337 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4340 //===---------------------------------------------------------------------===//
4341 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4342 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4344 let AddedComplexity = 15 in
4345 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4346 "vmovq\t{$src, $dst|$dst, $src}",
4347 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4348 XS, VEX, Requires<[HasAVX]>;
4349 let AddedComplexity = 15 in
4350 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4351 "movq\t{$src, $dst|$dst, $src}",
4352 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4353 XS, Requires<[HasSSE2]>;
4355 let AddedComplexity = 20 in
4356 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4357 "vmovq\t{$src, $dst|$dst, $src}",
4358 [(set VR128:$dst, (v2i64 (X86vzmovl
4359 (loadv2i64 addr:$src))))]>,
4360 XS, VEX, Requires<[HasAVX]>;
4361 let AddedComplexity = 20 in {
4362 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4363 "movq\t{$src, $dst|$dst, $src}",
4364 [(set VR128:$dst, (v2i64 (X86vzmovl
4365 (loadv2i64 addr:$src))))]>,
4366 XS, Requires<[HasSSE2]>;
4369 let AddedComplexity = 20 in {
4370 let Predicates = [HasAVX] in {
4371 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4372 (VMOVZPQILo2PQIrm addr:$src)>;
4373 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4374 (VMOVZPQILo2PQIrr VR128:$src)>;
4376 let Predicates = [HasSSE2] in {
4377 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4378 (MOVZPQILo2PQIrm addr:$src)>;
4379 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4380 (MOVZPQILo2PQIrr VR128:$src)>;
4384 // Instructions to match in the assembler
4385 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4386 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4387 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4388 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4389 // Recognize "movd" with GR64 destination, but encode as a "movq"
4390 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4391 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4393 // Instructions for the disassembler
4394 // xr = XMM register
4397 let Predicates = [HasAVX] in
4398 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4399 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4400 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4401 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4403 //===---------------------------------------------------------------------===//
4404 // SSE3 - Conversion Instructions
4405 //===---------------------------------------------------------------------===//
4407 // Convert Packed Double FP to Packed DW Integers
4408 let Predicates = [HasAVX] in {
4409 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4410 // register, but the same isn't true when using memory operands instead.
4411 // Provide other assembly rr and rm forms to address this explicitly.
4412 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4413 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4414 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4415 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4418 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4419 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4420 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4421 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4424 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4425 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4426 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4427 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4430 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4431 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4432 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4433 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4435 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4436 (VCVTTPD2DQYrr VR256:$src)>;
4437 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4438 (VCVTTPD2DQYrm addr:$src)>;
4440 // Convert Packed DW Integers to Packed Double FP
4441 let Predicates = [HasAVX] in {
4442 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4443 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4444 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4445 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4446 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4447 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4448 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4449 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4452 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4453 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4454 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4455 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4457 // AVX 256-bit register conversion intrinsics
4458 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4459 (VCVTDQ2PDYrr VR128:$src)>;
4460 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4461 (VCVTDQ2PDYrm addr:$src)>;
4463 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4464 (VCVTPD2DQYrr VR256:$src)>;
4465 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4466 (VCVTPD2DQYrm addr:$src)>;
4468 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4469 (VCVTDQ2PDYrr VR128:$src)>;
4470 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4471 (VCVTDQ2PDYrm addr:$src)>;
4473 //===---------------------------------------------------------------------===//
4474 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4475 //===---------------------------------------------------------------------===//
4476 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4477 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4478 X86MemOperand x86memop> {
4479 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4480 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4481 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4482 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4483 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4484 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4487 let Predicates = [HasAVX] in {
4488 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4489 v4f32, VR128, memopv4f32, f128mem>, VEX;
4490 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4491 v4f32, VR128, memopv4f32, f128mem>, VEX;
4492 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4493 v8f32, VR256, memopv8f32, f256mem>, VEX;
4494 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4495 v8f32, VR256, memopv8f32, f256mem>, VEX;
4497 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4498 memopv4f32, f128mem>;
4499 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4500 memopv4f32, f128mem>;
4502 let Predicates = [HasAVX] in {
4503 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4504 (VMOVSHDUPrr VR128:$src)>;
4505 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4506 (VMOVSHDUPrm addr:$src)>;
4507 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4508 (VMOVSLDUPrr VR128:$src)>;
4509 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4510 (VMOVSLDUPrm addr:$src)>;
4511 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4512 (VMOVSHDUPYrr VR256:$src)>;
4513 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4514 (VMOVSHDUPYrm addr:$src)>;
4515 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4516 (VMOVSLDUPYrr VR256:$src)>;
4517 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4518 (VMOVSLDUPYrm addr:$src)>;
4521 let Predicates = [HasSSE3] in {
4522 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4523 (MOVSHDUPrr VR128:$src)>;
4524 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4525 (MOVSHDUPrm addr:$src)>;
4526 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4527 (MOVSLDUPrr VR128:$src)>;
4528 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4529 (MOVSLDUPrm addr:$src)>;
4532 //===---------------------------------------------------------------------===//
4533 // SSE3 - Replicate Double FP - MOVDDUP
4534 //===---------------------------------------------------------------------===//
4536 multiclass sse3_replicate_dfp<string OpcodeStr> {
4537 let neverHasSideEffects = 1 in
4538 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4539 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4541 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4542 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4545 (scalar_to_vector (loadf64 addr:$src)))))]>;
4548 // FIXME: Merge with above classe when there're patterns for the ymm version
4549 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4550 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4551 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4552 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4553 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4554 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4557 (scalar_to_vector (loadf64 addr:$src)))))]>;
4560 let Predicates = [HasAVX] in {
4561 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4562 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4565 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4567 let Predicates = [HasAVX] in {
4568 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4569 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4570 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4571 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4572 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4573 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4574 def : Pat<(X86Movddup (bc_v2f64
4575 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4576 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4579 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4580 (VMOVDDUPYrm addr:$src)>;
4581 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4582 (VMOVDDUPYrm addr:$src)>;
4583 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4584 (VMOVDDUPYrm addr:$src)>;
4585 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4586 (VMOVDDUPYrr VR256:$src)>;
4589 let Predicates = [HasSSE3] in {
4590 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4591 (MOVDDUPrm addr:$src)>;
4592 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4593 (MOVDDUPrm addr:$src)>;
4594 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4595 (MOVDDUPrm addr:$src)>;
4596 def : Pat<(X86Movddup (bc_v2f64
4597 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4598 (MOVDDUPrm addr:$src)>;
4601 //===---------------------------------------------------------------------===//
4602 // SSE3 - Move Unaligned Integer
4603 //===---------------------------------------------------------------------===//
4605 let Predicates = [HasAVX] in {
4606 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4607 "vlddqu\t{$src, $dst|$dst, $src}",
4608 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4609 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4610 "vlddqu\t{$src, $dst|$dst, $src}",
4611 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4613 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4614 "lddqu\t{$src, $dst|$dst, $src}",
4615 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4617 //===---------------------------------------------------------------------===//
4618 // SSE3 - Arithmetic
4619 //===---------------------------------------------------------------------===//
4621 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4622 X86MemOperand x86memop, bit Is2Addr = 1> {
4623 def rr : I<0xD0, MRMSrcReg,
4624 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4626 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4627 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4628 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4629 def rm : I<0xD0, MRMSrcMem,
4630 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4632 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4633 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4634 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4637 let Predicates = [HasAVX] in {
4638 let ExeDomain = SSEPackedSingle in {
4639 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4640 f128mem, 0>, TB, XD, VEX_4V;
4641 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4642 f256mem, 0>, TB, XD, VEX_4V;
4644 let ExeDomain = SSEPackedDouble in {
4645 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4646 f128mem, 0>, TB, OpSize, VEX_4V;
4647 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4648 f256mem, 0>, TB, OpSize, VEX_4V;
4651 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
4652 let ExeDomain = SSEPackedSingle in
4653 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4655 let ExeDomain = SSEPackedDouble in
4656 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4657 f128mem>, TB, OpSize;
4660 //===---------------------------------------------------------------------===//
4661 // SSE3 Instructions
4662 //===---------------------------------------------------------------------===//
4665 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4666 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4667 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4669 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4670 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4671 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4673 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4675 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4676 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4677 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4679 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4680 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4681 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4683 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4684 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4685 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4687 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4689 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4690 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4691 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4694 let Predicates = [HasAVX] in {
4695 let ExeDomain = SSEPackedSingle in {
4696 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4697 X86fhadd, 0>, VEX_4V;
4698 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4699 X86fhsub, 0>, VEX_4V;
4700 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4701 X86fhadd, 0>, VEX_4V;
4702 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4703 X86fhsub, 0>, VEX_4V;
4705 let ExeDomain = SSEPackedDouble in {
4706 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4707 X86fhadd, 0>, VEX_4V;
4708 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4709 X86fhsub, 0>, VEX_4V;
4710 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4711 X86fhadd, 0>, VEX_4V;
4712 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4713 X86fhsub, 0>, VEX_4V;
4717 let Constraints = "$src1 = $dst" in {
4718 let ExeDomain = SSEPackedSingle in {
4719 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
4720 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
4722 let ExeDomain = SSEPackedDouble in {
4723 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
4724 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
4728 //===---------------------------------------------------------------------===//
4729 // SSSE3 - Packed Absolute Instructions
4730 //===---------------------------------------------------------------------===//
4733 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4734 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4735 Intrinsic IntId128> {
4736 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4738 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4739 [(set VR128:$dst, (IntId128 VR128:$src))]>,
4742 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4744 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4747 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
4750 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4751 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
4752 Intrinsic IntId256> {
4753 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4755 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4756 [(set VR256:$dst, (IntId256 VR256:$src))]>,
4759 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4761 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4764 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
4767 let Predicates = [HasAVX] in {
4768 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
4769 int_x86_ssse3_pabs_b_128>, VEX;
4770 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
4771 int_x86_ssse3_pabs_w_128>, VEX;
4772 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
4773 int_x86_ssse3_pabs_d_128>, VEX;
4776 let Predicates = [HasAVX2] in {
4777 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
4778 int_x86_avx2_pabs_b>, VEX;
4779 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
4780 int_x86_avx2_pabs_w>, VEX;
4781 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
4782 int_x86_avx2_pabs_d>, VEX;
4785 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
4786 int_x86_ssse3_pabs_b_128>;
4787 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
4788 int_x86_ssse3_pabs_w_128>;
4789 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
4790 int_x86_ssse3_pabs_d_128>;
4792 //===---------------------------------------------------------------------===//
4793 // SSSE3 - Packed Binary Operator Instructions
4794 //===---------------------------------------------------------------------===//
4796 /// SS3I_binop_rm - Simple SSSE3 bin op
4797 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4798 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
4799 X86MemOperand x86memop, bit Is2Addr = 1> {
4800 let isCommutable = 1 in
4801 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
4802 (ins RC:$src1, RC:$src2),
4804 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4805 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4806 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
4808 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
4809 (ins RC:$src1, x86memop:$src2),
4811 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4812 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4814 (OpVT (OpNode RC:$src1,
4815 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
4818 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4819 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4820 Intrinsic IntId128, bit Is2Addr = 1> {
4821 let isCommutable = 1 in
4822 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4823 (ins VR128:$src1, VR128:$src2),
4825 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4826 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4827 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4829 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4830 (ins VR128:$src1, i128mem:$src2),
4832 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4833 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4835 (IntId128 VR128:$src1,
4836 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
4839 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
4840 Intrinsic IntId256> {
4841 let isCommutable = 1 in
4842 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4843 (ins VR256:$src1, VR256:$src2),
4844 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4845 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
4847 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4848 (ins VR256:$src1, i256mem:$src2),
4849 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4851 (IntId256 VR256:$src1,
4852 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
4855 let ImmT = NoImm, Predicates = [HasAVX] in {
4856 let isCommutable = 0 in {
4857 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
4858 memopv2i64, i128mem, 0>, VEX_4V;
4859 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
4860 memopv2i64, i128mem, 0>, VEX_4V;
4861 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
4862 memopv2i64, i128mem, 0>, VEX_4V;
4863 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
4864 memopv2i64, i128mem, 0>, VEX_4V;
4865 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
4866 memopv2i64, i128mem, 0>, VEX_4V;
4867 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
4868 memopv2i64, i128mem, 0>, VEX_4V;
4869 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
4870 memopv2i64, i128mem, 0>, VEX_4V;
4871 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
4872 memopv2i64, i128mem, 0>, VEX_4V;
4873 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
4874 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
4875 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
4876 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
4877 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
4878 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
4880 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
4881 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
4884 let ImmT = NoImm, Predicates = [HasAVX2] in {
4885 let isCommutable = 0 in {
4886 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
4887 memopv4i64, i256mem, 0>, VEX_4V;
4888 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
4889 memopv4i64, i256mem, 0>, VEX_4V;
4890 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
4891 memopv4i64, i256mem, 0>, VEX_4V;
4892 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
4893 memopv4i64, i256mem, 0>, VEX_4V;
4894 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
4895 memopv4i64, i256mem, 0>, VEX_4V;
4896 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
4897 memopv4i64, i256mem, 0>, VEX_4V;
4898 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
4899 memopv4i64, i256mem, 0>, VEX_4V;
4900 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
4901 memopv4i64, i256mem, 0>, VEX_4V;
4902 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
4903 int_x86_avx2_phadd_sw>, VEX_4V;
4904 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
4905 int_x86_avx2_phsub_sw>, VEX_4V;
4906 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
4907 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
4909 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
4910 int_x86_avx2_pmul_hr_sw>, VEX_4V;
4913 // None of these have i8 immediate fields.
4914 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
4915 let isCommutable = 0 in {
4916 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
4917 memopv2i64, i128mem>;
4918 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
4919 memopv2i64, i128mem>;
4920 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
4921 memopv2i64, i128mem>;
4922 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
4923 memopv2i64, i128mem>;
4924 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
4925 memopv2i64, i128mem>;
4926 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
4927 memopv2i64, i128mem>;
4928 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
4929 memopv2i64, i128mem>;
4930 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
4931 memopv2i64, i128mem>;
4932 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
4933 int_x86_ssse3_phadd_sw_128>;
4934 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
4935 int_x86_ssse3_phsub_sw_128>;
4936 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
4937 int_x86_ssse3_pmadd_ub_sw_128>;
4939 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
4940 int_x86_ssse3_pmul_hr_sw_128>;
4943 //===---------------------------------------------------------------------===//
4944 // SSSE3 - Packed Align Instruction Patterns
4945 //===---------------------------------------------------------------------===//
4947 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
4948 let neverHasSideEffects = 1 in {
4949 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
4950 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4952 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4954 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4957 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
4958 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4960 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4962 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4967 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
4968 let neverHasSideEffects = 1 in {
4969 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
4970 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
4972 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4975 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
4976 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
4978 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4983 let Predicates = [HasAVX] in
4984 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
4985 let Predicates = [HasAVX2] in
4986 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
4987 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
4988 defm PALIGN : ssse3_palign<"palignr">;
4990 let Predicates = [HasAVX2] in {
4991 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
4992 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
4993 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
4994 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
4995 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
4996 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
4997 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
4998 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5001 let Predicates = [HasAVX] in {
5002 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5003 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5004 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5005 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5006 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5007 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5008 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5009 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5012 let Predicates = [HasSSSE3] in {
5013 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5014 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5015 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5016 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5017 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5018 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5019 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5020 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5023 //===---------------------------------------------------------------------===//
5024 // SSSE3 - Thread synchronization
5025 //===---------------------------------------------------------------------===//
5027 let usesCustomInserter = 1 in {
5028 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5029 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5030 Requires<[HasSSE3]>;
5031 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5032 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5033 Requires<[HasSSE3]>;
5036 let Uses = [EAX, ECX, EDX] in
5037 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5038 Requires<[HasSSE3]>;
5039 let Uses = [ECX, EAX] in
5040 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5041 Requires<[HasSSE3]>;
5043 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5044 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5046 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5047 Requires<[In32BitMode]>;
5048 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5049 Requires<[In64BitMode]>;
5051 //===----------------------------------------------------------------------===//
5052 // SSE4.1 - Packed Move with Sign/Zero Extend
5053 //===----------------------------------------------------------------------===//
5055 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5056 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5057 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5058 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5060 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5061 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5063 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5067 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5069 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5070 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5071 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5073 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5074 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5075 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5078 let Predicates = [HasAVX] in {
5079 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5081 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5083 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5085 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5087 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5089 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5093 let Predicates = [HasAVX2] in {
5094 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5095 int_x86_avx2_pmovsxbw>, VEX;
5096 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5097 int_x86_avx2_pmovsxwd>, VEX;
5098 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5099 int_x86_avx2_pmovsxdq>, VEX;
5100 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5101 int_x86_avx2_pmovzxbw>, VEX;
5102 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5103 int_x86_avx2_pmovzxwd>, VEX;
5104 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5105 int_x86_avx2_pmovzxdq>, VEX;
5108 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5109 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5110 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5111 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5112 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5113 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5115 let Predicates = [HasAVX] in {
5116 // Common patterns involving scalar load.
5117 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5118 (VPMOVSXBWrm addr:$src)>;
5119 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5120 (VPMOVSXBWrm addr:$src)>;
5122 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5123 (VPMOVSXWDrm addr:$src)>;
5124 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5125 (VPMOVSXWDrm addr:$src)>;
5127 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5128 (VPMOVSXDQrm addr:$src)>;
5129 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5130 (VPMOVSXDQrm addr:$src)>;
5132 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5133 (VPMOVZXBWrm addr:$src)>;
5134 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5135 (VPMOVZXBWrm addr:$src)>;
5137 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5138 (VPMOVZXWDrm addr:$src)>;
5139 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5140 (VPMOVZXWDrm addr:$src)>;
5142 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5143 (VPMOVZXDQrm addr:$src)>;
5144 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5145 (VPMOVZXDQrm addr:$src)>;
5148 let Predicates = [HasSSE41] in {
5149 // Common patterns involving scalar load.
5150 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5151 (PMOVSXBWrm addr:$src)>;
5152 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5153 (PMOVSXBWrm addr:$src)>;
5155 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5156 (PMOVSXWDrm addr:$src)>;
5157 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5158 (PMOVSXWDrm addr:$src)>;
5160 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5161 (PMOVSXDQrm addr:$src)>;
5162 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5163 (PMOVSXDQrm addr:$src)>;
5165 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5166 (PMOVZXBWrm addr:$src)>;
5167 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5168 (PMOVZXBWrm addr:$src)>;
5170 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5171 (PMOVZXWDrm addr:$src)>;
5172 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5173 (PMOVZXWDrm addr:$src)>;
5175 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5176 (PMOVZXDQrm addr:$src)>;
5177 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5178 (PMOVZXDQrm addr:$src)>;
5181 let Predicates = [HasAVX] in {
5182 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5183 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5186 let Predicates = [HasSSE41] in {
5187 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5188 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5192 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5193 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5194 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5195 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5197 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5198 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5200 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5204 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5206 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5207 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5208 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5210 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5211 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5213 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5217 let Predicates = [HasAVX] in {
5218 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5220 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5222 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5224 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5228 let Predicates = [HasAVX2] in {
5229 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5230 int_x86_avx2_pmovsxbd>, VEX;
5231 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5232 int_x86_avx2_pmovsxwq>, VEX;
5233 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5234 int_x86_avx2_pmovzxbd>, VEX;
5235 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5236 int_x86_avx2_pmovzxwq>, VEX;
5239 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5240 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5241 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5242 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5244 let Predicates = [HasAVX] in {
5245 // Common patterns involving scalar load
5246 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5247 (VPMOVSXBDrm addr:$src)>;
5248 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5249 (VPMOVSXWQrm addr:$src)>;
5251 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5252 (VPMOVZXBDrm addr:$src)>;
5253 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5254 (VPMOVZXWQrm addr:$src)>;
5257 let Predicates = [HasSSE41] in {
5258 // Common patterns involving scalar load
5259 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5260 (PMOVSXBDrm addr:$src)>;
5261 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5262 (PMOVSXWQrm addr:$src)>;
5264 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5265 (PMOVZXBDrm addr:$src)>;
5266 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5267 (PMOVZXWQrm addr:$src)>;
5270 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5271 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5272 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5273 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5275 // Expecting a i16 load any extended to i32 value.
5276 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5277 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5278 [(set VR128:$dst, (IntId (bitconvert
5279 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5283 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5285 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5286 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5287 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5289 // Expecting a i16 load any extended to i32 value.
5290 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5291 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5292 [(set VR256:$dst, (IntId (bitconvert
5293 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5297 let Predicates = [HasAVX] in {
5298 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5300 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5303 let Predicates = [HasAVX2] in {
5304 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5305 int_x86_avx2_pmovsxbq>, VEX;
5306 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5307 int_x86_avx2_pmovzxbq>, VEX;
5309 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5310 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5312 let Predicates = [HasAVX] in {
5313 // Common patterns involving scalar load
5314 def : Pat<(int_x86_sse41_pmovsxbq
5315 (bitconvert (v4i32 (X86vzmovl
5316 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5317 (VPMOVSXBQrm addr:$src)>;
5319 def : Pat<(int_x86_sse41_pmovzxbq
5320 (bitconvert (v4i32 (X86vzmovl
5321 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5322 (VPMOVZXBQrm addr:$src)>;
5325 let Predicates = [HasSSE41] in {
5326 // Common patterns involving scalar load
5327 def : Pat<(int_x86_sse41_pmovsxbq
5328 (bitconvert (v4i32 (X86vzmovl
5329 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5330 (PMOVSXBQrm addr:$src)>;
5332 def : Pat<(int_x86_sse41_pmovzxbq
5333 (bitconvert (v4i32 (X86vzmovl
5334 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5335 (PMOVZXBQrm addr:$src)>;
5338 //===----------------------------------------------------------------------===//
5339 // SSE4.1 - Extract Instructions
5340 //===----------------------------------------------------------------------===//
5342 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5343 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5344 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5345 (ins VR128:$src1, i32i8imm:$src2),
5346 !strconcat(OpcodeStr,
5347 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5348 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5350 let neverHasSideEffects = 1, mayStore = 1 in
5351 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5352 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5353 !strconcat(OpcodeStr,
5354 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5357 // There's an AssertZext in the way of writing the store pattern
5358 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5361 let Predicates = [HasAVX] in {
5362 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5363 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5364 (ins VR128:$src1, i32i8imm:$src2),
5365 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5368 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5371 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5372 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5373 let neverHasSideEffects = 1, mayStore = 1 in
5374 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5375 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5376 !strconcat(OpcodeStr,
5377 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5380 // There's an AssertZext in the way of writing the store pattern
5381 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5384 let Predicates = [HasAVX] in
5385 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5387 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5390 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5391 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5392 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5393 (ins VR128:$src1, i32i8imm:$src2),
5394 !strconcat(OpcodeStr,
5395 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5397 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5398 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5399 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5400 !strconcat(OpcodeStr,
5401 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5402 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5403 addr:$dst)]>, OpSize;
5406 let Predicates = [HasAVX] in
5407 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5409 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5411 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5412 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5413 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5414 (ins VR128:$src1, i32i8imm:$src2),
5415 !strconcat(OpcodeStr,
5416 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5418 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5419 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5420 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5421 !strconcat(OpcodeStr,
5422 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5423 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5424 addr:$dst)]>, OpSize, REX_W;
5427 let Predicates = [HasAVX] in
5428 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5430 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5432 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5434 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5435 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5436 (ins VR128:$src1, i32i8imm:$src2),
5437 !strconcat(OpcodeStr,
5438 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5440 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5442 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5443 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5444 !strconcat(OpcodeStr,
5445 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5446 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5447 addr:$dst)]>, OpSize;
5450 let ExeDomain = SSEPackedSingle in {
5451 let Predicates = [HasAVX] in {
5452 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5453 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5454 (ins VR128:$src1, i32i8imm:$src2),
5455 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5458 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5461 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5462 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5465 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5467 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5470 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5471 Requires<[HasSSE41]>;
5473 //===----------------------------------------------------------------------===//
5474 // SSE4.1 - Insert Instructions
5475 //===----------------------------------------------------------------------===//
5477 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5478 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5479 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5481 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5483 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5485 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5486 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5487 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5489 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5491 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5493 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5494 imm:$src3))]>, OpSize;
5497 let Predicates = [HasAVX] in
5498 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5499 let Constraints = "$src1 = $dst" in
5500 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5502 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5503 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5504 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5506 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5508 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5510 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5512 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5513 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5515 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5517 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5519 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5520 imm:$src3)))]>, OpSize;
5523 let Predicates = [HasAVX] in
5524 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5525 let Constraints = "$src1 = $dst" in
5526 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5528 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5529 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5530 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5532 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5534 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5536 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5538 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5539 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5541 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5543 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5545 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5546 imm:$src3)))]>, OpSize;
5549 let Predicates = [HasAVX] in
5550 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5551 let Constraints = "$src1 = $dst" in
5552 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5554 // insertps has a few different modes, there's the first two here below which
5555 // are optimized inserts that won't zero arbitrary elements in the destination
5556 // vector. The next one matches the intrinsic and could zero arbitrary elements
5557 // in the target vector.
5558 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5559 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5560 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5562 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5564 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5566 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5568 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5569 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5571 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5573 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5575 (X86insrtps VR128:$src1,
5576 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5577 imm:$src3))]>, OpSize;
5580 let ExeDomain = SSEPackedSingle in {
5581 let Predicates = [HasAVX] in
5582 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5583 let Constraints = "$src1 = $dst" in
5584 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5587 //===----------------------------------------------------------------------===//
5588 // SSE4.1 - Round Instructions
5589 //===----------------------------------------------------------------------===//
5591 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5592 X86MemOperand x86memop, RegisterClass RC,
5593 PatFrag mem_frag32, PatFrag mem_frag64,
5594 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5595 let ExeDomain = SSEPackedSingle in {
5596 // Intrinsic operation, reg.
5597 // Vector intrinsic operation, reg
5598 def PSr : SS4AIi8<opcps, MRMSrcReg,
5599 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5600 !strconcat(OpcodeStr,
5601 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5602 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5605 // Vector intrinsic operation, mem
5606 def PSm : SS4AIi8<opcps, MRMSrcMem,
5607 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5608 !strconcat(OpcodeStr,
5609 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5611 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5613 } // ExeDomain = SSEPackedSingle
5615 let ExeDomain = SSEPackedDouble in {
5616 // Vector intrinsic operation, reg
5617 def PDr : SS4AIi8<opcpd, MRMSrcReg,
5618 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5619 !strconcat(OpcodeStr,
5620 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5621 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
5624 // Vector intrinsic operation, mem
5625 def PDm : SS4AIi8<opcpd, MRMSrcMem,
5626 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5627 !strconcat(OpcodeStr,
5628 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5630 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
5632 } // ExeDomain = SSEPackedDouble
5635 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
5638 Intrinsic F64Int, bit Is2Addr = 1> {
5639 let ExeDomain = GenericDomain in {
5641 def SSr : SS4AIi8<opcss, MRMSrcReg,
5642 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
5644 !strconcat(OpcodeStr,
5645 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5646 !strconcat(OpcodeStr,
5647 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5650 // Intrinsic operation, reg.
5651 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
5652 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5654 !strconcat(OpcodeStr,
5655 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5656 !strconcat(OpcodeStr,
5657 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5658 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5661 // Intrinsic operation, mem.
5662 def SSm : SS4AIi8<opcss, MRMSrcMem,
5663 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5665 !strconcat(OpcodeStr,
5666 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5667 !strconcat(OpcodeStr,
5668 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5670 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
5674 def SDr : SS4AIi8<opcsd, MRMSrcReg,
5675 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
5677 !strconcat(OpcodeStr,
5678 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5679 !strconcat(OpcodeStr,
5680 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5683 // Intrinsic operation, reg.
5684 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
5685 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5687 !strconcat(OpcodeStr,
5688 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5689 !strconcat(OpcodeStr,
5690 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5691 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5694 // Intrinsic operation, mem.
5695 def SDm : SS4AIi8<opcsd, MRMSrcMem,
5696 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5698 !strconcat(OpcodeStr,
5699 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5700 !strconcat(OpcodeStr,
5701 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5703 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
5705 } // ExeDomain = GenericDomain
5708 // FP round - roundss, roundps, roundsd, roundpd
5709 let Predicates = [HasAVX] in {
5711 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
5712 memopv4f32, memopv2f64,
5713 int_x86_sse41_round_ps,
5714 int_x86_sse41_round_pd>, VEX;
5715 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
5716 memopv8f32, memopv4f64,
5717 int_x86_avx_round_ps_256,
5718 int_x86_avx_round_pd_256>, VEX;
5719 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
5720 int_x86_sse41_round_ss,
5721 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
5723 def : Pat<(ffloor FR32:$src),
5724 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
5725 def : Pat<(f64 (ffloor FR64:$src)),
5726 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
5727 def : Pat<(f32 (fnearbyint FR32:$src)),
5728 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
5729 def : Pat<(f64 (fnearbyint FR64:$src)),
5730 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
5731 def : Pat<(f32 (fceil FR32:$src)),
5732 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
5733 def : Pat<(f64 (fceil FR64:$src)),
5734 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
5735 def : Pat<(f32 (frint FR32:$src)),
5736 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
5737 def : Pat<(f64 (frint FR64:$src)),
5738 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
5739 def : Pat<(f32 (ftrunc FR32:$src)),
5740 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
5741 def : Pat<(f64 (ftrunc FR64:$src)),
5742 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
5745 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
5746 memopv4f32, memopv2f64,
5747 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
5748 let Constraints = "$src1 = $dst" in
5749 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
5750 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
5752 def : Pat<(ffloor FR32:$src),
5753 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
5754 def : Pat<(f64 (ffloor FR64:$src)),
5755 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
5756 def : Pat<(f32 (fnearbyint FR32:$src)),
5757 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
5758 def : Pat<(f64 (fnearbyint FR64:$src)),
5759 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
5760 def : Pat<(f32 (fceil FR32:$src)),
5761 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
5762 def : Pat<(f64 (fceil FR64:$src)),
5763 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
5764 def : Pat<(f32 (frint FR32:$src)),
5765 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
5766 def : Pat<(f64 (frint FR64:$src)),
5767 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
5768 def : Pat<(f32 (ftrunc FR32:$src)),
5769 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
5770 def : Pat<(f64 (ftrunc FR64:$src)),
5771 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
5773 //===----------------------------------------------------------------------===//
5774 // SSE4.1 - Packed Bit Test
5775 //===----------------------------------------------------------------------===//
5777 // ptest instruction we'll lower to this in X86ISelLowering primarily from
5778 // the intel intrinsic that corresponds to this.
5779 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5780 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5781 "vptest\t{$src2, $src1|$src1, $src2}",
5782 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5784 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5785 "vptest\t{$src2, $src1|$src1, $src2}",
5786 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5789 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
5790 "vptest\t{$src2, $src1|$src1, $src2}",
5791 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
5793 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
5794 "vptest\t{$src2, $src1|$src1, $src2}",
5795 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
5799 let Defs = [EFLAGS] in {
5800 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5801 "ptest\t{$src2, $src1|$src1, $src2}",
5802 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5804 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5805 "ptest\t{$src2, $src1|$src1, $src2}",
5806 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5810 // The bit test instructions below are AVX only
5811 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
5812 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
5813 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
5814 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5815 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
5816 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
5817 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5818 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
5822 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5823 let ExeDomain = SSEPackedSingle in {
5824 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
5825 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
5827 let ExeDomain = SSEPackedDouble in {
5828 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
5829 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
5833 //===----------------------------------------------------------------------===//
5834 // SSE4.1 - Misc Instructions
5835 //===----------------------------------------------------------------------===//
5837 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
5838 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
5839 "popcnt{w}\t{$src, $dst|$dst, $src}",
5840 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
5842 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
5843 "popcnt{w}\t{$src, $dst|$dst, $src}",
5844 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
5845 (implicit EFLAGS)]>, OpSize, XS;
5847 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
5848 "popcnt{l}\t{$src, $dst|$dst, $src}",
5849 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
5851 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
5852 "popcnt{l}\t{$src, $dst|$dst, $src}",
5853 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
5854 (implicit EFLAGS)]>, XS;
5856 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
5857 "popcnt{q}\t{$src, $dst|$dst, $src}",
5858 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
5860 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
5861 "popcnt{q}\t{$src, $dst|$dst, $src}",
5862 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
5863 (implicit EFLAGS)]>, XS;
5868 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
5869 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
5870 Intrinsic IntId128> {
5871 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5873 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5874 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
5875 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5877 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5880 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5883 let Predicates = [HasAVX] in
5884 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
5885 int_x86_sse41_phminposuw>, VEX;
5886 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
5887 int_x86_sse41_phminposuw>;
5889 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
5890 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
5891 Intrinsic IntId128, bit Is2Addr = 1> {
5892 let isCommutable = 1 in
5893 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5894 (ins VR128:$src1, VR128:$src2),
5896 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5897 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5898 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
5899 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5900 (ins VR128:$src1, i128mem:$src2),
5902 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5903 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5905 (IntId128 VR128:$src1,
5906 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5909 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
5910 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5911 Intrinsic IntId256> {
5912 let isCommutable = 1 in
5913 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
5914 (ins VR256:$src1, VR256:$src2),
5915 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5916 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
5917 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
5918 (ins VR256:$src1, i256mem:$src2),
5919 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5921 (IntId256 VR256:$src1,
5922 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5925 let Predicates = [HasAVX] in {
5926 let isCommutable = 0 in
5927 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
5929 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
5931 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
5933 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
5935 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
5937 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
5939 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
5941 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
5943 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
5945 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
5949 let Predicates = [HasAVX2] in {
5950 let isCommutable = 0 in
5951 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
5952 int_x86_avx2_packusdw>, VEX_4V;
5953 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
5954 int_x86_avx2_pmins_b>, VEX_4V;
5955 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
5956 int_x86_avx2_pmins_d>, VEX_4V;
5957 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
5958 int_x86_avx2_pminu_d>, VEX_4V;
5959 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
5960 int_x86_avx2_pminu_w>, VEX_4V;
5961 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
5962 int_x86_avx2_pmaxs_b>, VEX_4V;
5963 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
5964 int_x86_avx2_pmaxs_d>, VEX_4V;
5965 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
5966 int_x86_avx2_pmaxu_d>, VEX_4V;
5967 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
5968 int_x86_avx2_pmaxu_w>, VEX_4V;
5969 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
5970 int_x86_avx2_pmul_dq>, VEX_4V;
5973 let Constraints = "$src1 = $dst" in {
5974 let isCommutable = 0 in
5975 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
5976 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
5977 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
5978 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
5979 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
5980 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
5981 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
5982 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
5983 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
5984 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
5987 /// SS48I_binop_rm - Simple SSE41 binary operator.
5988 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5989 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5990 X86MemOperand x86memop, bit Is2Addr = 1> {
5991 let isCommutable = 1 in
5992 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
5993 (ins RC:$src1, RC:$src2),
5995 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5996 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5997 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
5998 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
5999 (ins RC:$src1, x86memop:$src2),
6001 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6002 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6004 (OpVT (OpNode RC:$src1,
6005 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6008 let Predicates = [HasAVX] in {
6009 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6010 memopv2i64, i128mem, 0>, VEX_4V;
6011 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6012 memopv2i64, i128mem, 0>, VEX_4V;
6014 let Predicates = [HasAVX2] in {
6015 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6016 memopv4i64, i256mem, 0>, VEX_4V;
6017 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6018 memopv4i64, i256mem, 0>, VEX_4V;
6021 let Constraints = "$src1 = $dst" in {
6022 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6023 memopv2i64, i128mem>;
6024 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6025 memopv2i64, i128mem>;
6028 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6029 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6030 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6031 X86MemOperand x86memop, bit Is2Addr = 1> {
6032 let isCommutable = 1 in
6033 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6034 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6036 !strconcat(OpcodeStr,
6037 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6038 !strconcat(OpcodeStr,
6039 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6040 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6042 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6043 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6045 !strconcat(OpcodeStr,
6046 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6047 !strconcat(OpcodeStr,
6048 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6051 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6055 let Predicates = [HasAVX] in {
6056 let isCommutable = 0 in {
6057 let ExeDomain = SSEPackedSingle in {
6058 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6059 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6060 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6061 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6063 let ExeDomain = SSEPackedDouble in {
6064 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6065 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6066 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6067 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6069 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6070 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6071 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6072 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6074 let ExeDomain = SSEPackedSingle in
6075 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6076 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6077 let ExeDomain = SSEPackedDouble in
6078 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6079 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6080 let ExeDomain = SSEPackedSingle in
6081 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6082 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6085 let Predicates = [HasAVX2] in {
6086 let isCommutable = 0 in {
6087 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6088 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6089 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6090 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6094 let Constraints = "$src1 = $dst" in {
6095 let isCommutable = 0 in {
6096 let ExeDomain = SSEPackedSingle in
6097 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6098 VR128, memopv4f32, i128mem>;
6099 let ExeDomain = SSEPackedDouble in
6100 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6101 VR128, memopv2f64, i128mem>;
6102 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6103 VR128, memopv2i64, i128mem>;
6104 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6105 VR128, memopv2i64, i128mem>;
6107 let ExeDomain = SSEPackedSingle in
6108 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6109 VR128, memopv4f32, i128mem>;
6110 let ExeDomain = SSEPackedDouble in
6111 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6112 VR128, memopv2f64, i128mem>;
6115 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6116 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6117 RegisterClass RC, X86MemOperand x86memop,
6118 PatFrag mem_frag, Intrinsic IntId> {
6119 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6120 (ins RC:$src1, RC:$src2, RC:$src3),
6121 !strconcat(OpcodeStr,
6122 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6123 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6124 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6126 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6127 (ins RC:$src1, x86memop:$src2, RC:$src3),
6128 !strconcat(OpcodeStr,
6129 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6131 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6133 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6136 let Predicates = [HasAVX] in {
6137 let ExeDomain = SSEPackedDouble in {
6138 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6139 memopv2f64, int_x86_sse41_blendvpd>;
6140 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6141 memopv4f64, int_x86_avx_blendv_pd_256>;
6142 } // ExeDomain = SSEPackedDouble
6143 let ExeDomain = SSEPackedSingle in {
6144 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6145 memopv4f32, int_x86_sse41_blendvps>;
6146 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6147 memopv8f32, int_x86_avx_blendv_ps_256>;
6148 } // ExeDomain = SSEPackedSingle
6149 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6150 memopv2i64, int_x86_sse41_pblendvb>;
6153 let Predicates = [HasAVX2] in {
6154 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6155 memopv4i64, int_x86_avx2_pblendvb>;
6158 let Predicates = [HasAVX] in {
6159 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6160 (v16i8 VR128:$src2))),
6161 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6162 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6163 (v4i32 VR128:$src2))),
6164 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6165 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6166 (v4f32 VR128:$src2))),
6167 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6168 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6169 (v2i64 VR128:$src2))),
6170 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6171 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6172 (v2f64 VR128:$src2))),
6173 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6174 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6175 (v8i32 VR256:$src2))),
6176 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6177 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6178 (v8f32 VR256:$src2))),
6179 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6180 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6181 (v4i64 VR256:$src2))),
6182 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6183 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6184 (v4f64 VR256:$src2))),
6185 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6188 let Predicates = [HasAVX2] in {
6189 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6190 (v32i8 VR256:$src2))),
6191 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6194 /// SS41I_ternary_int - SSE 4.1 ternary operator
6195 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6196 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6198 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6199 (ins VR128:$src1, VR128:$src2),
6200 !strconcat(OpcodeStr,
6201 "\t{$src2, $dst|$dst, $src2}"),
6202 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6205 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6206 (ins VR128:$src1, i128mem:$src2),
6207 !strconcat(OpcodeStr,
6208 "\t{$src2, $dst|$dst, $src2}"),
6211 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6215 let ExeDomain = SSEPackedDouble in
6216 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6217 int_x86_sse41_blendvpd>;
6218 let ExeDomain = SSEPackedSingle in
6219 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6220 int_x86_sse41_blendvps>;
6221 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6222 int_x86_sse41_pblendvb>;
6224 let Predicates = [HasSSE41] in {
6225 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6226 (v16i8 VR128:$src2))),
6227 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6228 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6229 (v4i32 VR128:$src2))),
6230 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6231 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6232 (v4f32 VR128:$src2))),
6233 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6234 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6235 (v2i64 VR128:$src2))),
6236 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6237 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6238 (v2f64 VR128:$src2))),
6239 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6242 let Predicates = [HasAVX] in
6243 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6244 "vmovntdqa\t{$src, $dst|$dst, $src}",
6245 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6247 let Predicates = [HasAVX2] in
6248 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6249 "vmovntdqa\t{$src, $dst|$dst, $src}",
6250 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6252 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6253 "movntdqa\t{$src, $dst|$dst, $src}",
6254 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6257 //===----------------------------------------------------------------------===//
6258 // SSE4.2 - Compare Instructions
6259 //===----------------------------------------------------------------------===//
6261 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6262 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6263 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6264 X86MemOperand x86memop, bit Is2Addr = 1> {
6265 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6266 (ins RC:$src1, RC:$src2),
6268 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6269 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6270 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6272 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6273 (ins RC:$src1, x86memop:$src2),
6275 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6276 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6278 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6281 let Predicates = [HasAVX] in
6282 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6283 memopv2i64, i128mem, 0>, VEX_4V;
6285 let Predicates = [HasAVX2] in
6286 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6287 memopv4i64, i256mem, 0>, VEX_4V;
6289 let Constraints = "$src1 = $dst" in
6290 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6291 memopv2i64, i128mem>;
6293 //===----------------------------------------------------------------------===//
6294 // SSE4.2 - String/text Processing Instructions
6295 //===----------------------------------------------------------------------===//
6297 // Packed Compare Implicit Length Strings, Return Mask
6298 multiclass pseudo_pcmpistrm<string asm> {
6299 def REG : PseudoI<(outs VR128:$dst),
6300 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6301 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6303 def MEM : PseudoI<(outs VR128:$dst),
6304 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6305 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6306 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6309 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6310 let AddedComplexity = 1 in
6311 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6312 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6315 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6316 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6317 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6318 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6320 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6321 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6322 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6325 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6326 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6327 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6328 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6330 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6331 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6332 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6335 // Packed Compare Explicit Length Strings, Return Mask
6336 multiclass pseudo_pcmpestrm<string asm> {
6337 def REG : PseudoI<(outs VR128:$dst),
6338 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6339 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6340 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6341 def MEM : PseudoI<(outs VR128:$dst),
6342 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6343 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6344 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6347 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6348 let AddedComplexity = 1 in
6349 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6350 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6353 let Predicates = [HasAVX],
6354 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6355 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6356 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6357 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6359 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6360 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6361 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6364 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6365 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6366 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6367 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6369 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6370 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6371 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6374 // Packed Compare Implicit Length Strings, Return Index
6375 let Defs = [ECX, EFLAGS] in {
6376 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6377 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6378 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6379 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6380 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6381 (implicit EFLAGS)]>, OpSize;
6382 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6383 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6384 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6385 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6386 (implicit EFLAGS)]>, OpSize;
6390 let Predicates = [HasAVX] in {
6391 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6393 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6395 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6397 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6399 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6401 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6405 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6406 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6407 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6408 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6409 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6410 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6412 // Packed Compare Explicit Length Strings, Return Index
6413 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6414 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6415 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6416 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6417 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6418 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6419 (implicit EFLAGS)]>, OpSize;
6420 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6421 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6422 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6424 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6425 (implicit EFLAGS)]>, OpSize;
6429 let Predicates = [HasAVX] in {
6430 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6432 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6434 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6436 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6438 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6440 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6444 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6445 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6446 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6447 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6448 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6449 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6451 //===----------------------------------------------------------------------===//
6452 // SSE4.2 - CRC Instructions
6453 //===----------------------------------------------------------------------===//
6455 // No CRC instructions have AVX equivalents
6457 // crc intrinsic instruction
6458 // This set of instructions are only rm, the only difference is the size
6460 let Constraints = "$src1 = $dst" in {
6461 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6462 (ins GR32:$src1, i8mem:$src2),
6463 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6465 (int_x86_sse42_crc32_32_8 GR32:$src1,
6466 (load addr:$src2)))]>;
6467 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6468 (ins GR32:$src1, GR8:$src2),
6469 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6471 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6472 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6473 (ins GR32:$src1, i16mem:$src2),
6474 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6476 (int_x86_sse42_crc32_32_16 GR32:$src1,
6477 (load addr:$src2)))]>,
6479 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6480 (ins GR32:$src1, GR16:$src2),
6481 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6483 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6485 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6486 (ins GR32:$src1, i32mem:$src2),
6487 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6489 (int_x86_sse42_crc32_32_32 GR32:$src1,
6490 (load addr:$src2)))]>;
6491 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6492 (ins GR32:$src1, GR32:$src2),
6493 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6495 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6496 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6497 (ins GR64:$src1, i8mem:$src2),
6498 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6500 (int_x86_sse42_crc32_64_8 GR64:$src1,
6501 (load addr:$src2)))]>,
6503 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6504 (ins GR64:$src1, GR8:$src2),
6505 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6507 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6509 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6510 (ins GR64:$src1, i64mem:$src2),
6511 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6513 (int_x86_sse42_crc32_64_64 GR64:$src1,
6514 (load addr:$src2)))]>,
6516 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6517 (ins GR64:$src1, GR64:$src2),
6518 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6520 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6524 //===----------------------------------------------------------------------===//
6525 // AES-NI Instructions
6526 //===----------------------------------------------------------------------===//
6528 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6529 Intrinsic IntId128, bit Is2Addr = 1> {
6530 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6531 (ins VR128:$src1, VR128:$src2),
6533 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6534 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6535 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6537 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6538 (ins VR128:$src1, i128mem:$src2),
6540 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6541 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6543 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6546 // Perform One Round of an AES Encryption/Decryption Flow
6547 let Predicates = [HasAVX, HasAES] in {
6548 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6549 int_x86_aesni_aesenc, 0>, VEX_4V;
6550 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6551 int_x86_aesni_aesenclast, 0>, VEX_4V;
6552 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6553 int_x86_aesni_aesdec, 0>, VEX_4V;
6554 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6555 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6558 let Constraints = "$src1 = $dst" in {
6559 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6560 int_x86_aesni_aesenc>;
6561 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
6562 int_x86_aesni_aesenclast>;
6563 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
6564 int_x86_aesni_aesdec>;
6565 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
6566 int_x86_aesni_aesdeclast>;
6569 // Perform the AES InvMixColumn Transformation
6570 let Predicates = [HasAVX, HasAES] in {
6571 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6573 "vaesimc\t{$src1, $dst|$dst, $src1}",
6575 (int_x86_aesni_aesimc VR128:$src1))]>,
6577 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6578 (ins i128mem:$src1),
6579 "vaesimc\t{$src1, $dst|$dst, $src1}",
6580 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6583 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6585 "aesimc\t{$src1, $dst|$dst, $src1}",
6587 (int_x86_aesni_aesimc VR128:$src1))]>,
6589 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6590 (ins i128mem:$src1),
6591 "aesimc\t{$src1, $dst|$dst, $src1}",
6592 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6595 // AES Round Key Generation Assist
6596 let Predicates = [HasAVX, HasAES] in {
6597 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6598 (ins VR128:$src1, i8imm:$src2),
6599 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6601 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6603 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6604 (ins i128mem:$src1, i8imm:$src2),
6605 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6607 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6610 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6611 (ins VR128:$src1, i8imm:$src2),
6612 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6614 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6616 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6617 (ins i128mem:$src1, i8imm:$src2),
6618 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6620 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6623 //===----------------------------------------------------------------------===//
6624 // CLMUL Instructions
6625 //===----------------------------------------------------------------------===//
6627 // Carry-less Multiplication instructions
6628 let neverHasSideEffects = 1 in {
6629 // AVX carry-less Multiplication instructions
6630 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6631 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6632 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6636 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6637 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6638 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6641 let Constraints = "$src1 = $dst" in {
6642 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6643 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6644 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6648 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6649 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6650 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6652 } // Constraints = "$src1 = $dst"
6653 } // neverHasSideEffects = 1
6656 multiclass pclmul_alias<string asm, int immop> {
6657 def : InstAlias<!strconcat("pclmul", asm,
6658 "dq {$src, $dst|$dst, $src}"),
6659 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
6661 def : InstAlias<!strconcat("pclmul", asm,
6662 "dq {$src, $dst|$dst, $src}"),
6663 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
6665 def : InstAlias<!strconcat("vpclmul", asm,
6666 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6667 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
6669 def : InstAlias<!strconcat("vpclmul", asm,
6670 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6671 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
6673 defm : pclmul_alias<"hqhq", 0x11>;
6674 defm : pclmul_alias<"hqlq", 0x01>;
6675 defm : pclmul_alias<"lqhq", 0x10>;
6676 defm : pclmul_alias<"lqlq", 0x00>;
6678 //===----------------------------------------------------------------------===//
6680 //===----------------------------------------------------------------------===//
6682 //===----------------------------------------------------------------------===//
6683 // VBROADCAST - Load from memory and broadcast to all elements of the
6684 // destination operand
6686 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
6687 X86MemOperand x86memop, Intrinsic Int> :
6688 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
6689 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6690 [(set RC:$dst, (Int addr:$src))]>, VEX;
6692 // AVX2 adds register forms
6693 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
6695 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
6696 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6697 [(set RC:$dst, (Int VR128:$src))]>, VEX;
6699 let ExeDomain = SSEPackedSingle in {
6700 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
6701 int_x86_avx_vbroadcast_ss>;
6702 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
6703 int_x86_avx_vbroadcast_ss_256>;
6705 let ExeDomain = SSEPackedDouble in
6706 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
6707 int_x86_avx_vbroadcast_sd_256>;
6708 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
6709 int_x86_avx_vbroadcastf128_pd_256>;
6711 let ExeDomain = SSEPackedSingle in {
6712 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
6713 int_x86_avx2_vbroadcast_ss_ps>;
6714 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
6715 int_x86_avx2_vbroadcast_ss_ps_256>;
6717 let ExeDomain = SSEPackedDouble in
6718 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
6719 int_x86_avx2_vbroadcast_sd_pd_256>;
6721 let Predicates = [HasAVX2] in
6722 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
6723 int_x86_avx2_vbroadcasti128>;
6725 let Predicates = [HasAVX] in
6726 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
6727 (VBROADCASTF128 addr:$src)>;
6730 //===----------------------------------------------------------------------===//
6731 // VINSERTF128 - Insert packed floating-point values
6733 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
6734 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
6735 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
6736 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6739 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
6740 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
6741 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6745 let Predicates = [HasAVX] in {
6746 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
6747 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6748 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
6749 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6750 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
6751 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6754 //===----------------------------------------------------------------------===//
6755 // VEXTRACTF128 - Extract packed floating-point values
6757 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
6758 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
6759 (ins VR256:$src1, i8imm:$src2),
6760 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6763 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
6764 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
6765 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6769 let Predicates = [HasAVX] in {
6770 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
6771 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6772 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
6773 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6774 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
6775 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6778 //===----------------------------------------------------------------------===//
6779 // VMASKMOV - Conditional SIMD Packed Loads and Stores
6781 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
6782 Intrinsic IntLd, Intrinsic IntLd256,
6783 Intrinsic IntSt, Intrinsic IntSt256> {
6784 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
6785 (ins VR128:$src1, f128mem:$src2),
6786 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6787 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
6789 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
6790 (ins VR256:$src1, f256mem:$src2),
6791 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6792 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
6794 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
6795 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
6796 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6797 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
6798 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
6799 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
6800 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6801 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
6804 let ExeDomain = SSEPackedSingle in
6805 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
6806 int_x86_avx_maskload_ps,
6807 int_x86_avx_maskload_ps_256,
6808 int_x86_avx_maskstore_ps,
6809 int_x86_avx_maskstore_ps_256>;
6810 let ExeDomain = SSEPackedDouble in
6811 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
6812 int_x86_avx_maskload_pd,
6813 int_x86_avx_maskload_pd_256,
6814 int_x86_avx_maskstore_pd,
6815 int_x86_avx_maskstore_pd_256>;
6817 //===----------------------------------------------------------------------===//
6818 // VPERMIL - Permute Single and Double Floating-Point Values
6820 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
6821 RegisterClass RC, X86MemOperand x86memop_f,
6822 X86MemOperand x86memop_i, PatFrag i_frag,
6823 Intrinsic IntVar, ValueType vt> {
6824 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
6825 (ins RC:$src1, RC:$src2),
6826 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6827 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
6828 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
6829 (ins RC:$src1, x86memop_i:$src2),
6830 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6831 [(set RC:$dst, (IntVar RC:$src1,
6832 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
6834 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
6835 (ins RC:$src1, i8imm:$src2),
6836 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6837 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
6838 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
6839 (ins x86memop_f:$src1, i8imm:$src2),
6840 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6842 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
6845 let ExeDomain = SSEPackedSingle in {
6846 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
6847 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
6848 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
6849 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
6851 let ExeDomain = SSEPackedDouble in {
6852 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
6853 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
6854 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
6855 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
6858 let Predicates = [HasAVX] in {
6859 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
6860 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6861 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
6862 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6863 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
6865 (VPERMILPSYmi addr:$src1, imm:$imm)>;
6866 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
6867 (VPERMILPDYmi addr:$src1, imm:$imm)>;
6869 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
6870 (VPERMILPDri VR128:$src1, imm:$imm)>;
6871 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
6872 (VPERMILPDmi addr:$src1, imm:$imm)>;
6875 //===----------------------------------------------------------------------===//
6876 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
6878 let ExeDomain = SSEPackedSingle in {
6879 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
6880 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
6881 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6882 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
6883 (i8 imm:$src3))))]>, VEX_4V;
6884 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
6885 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
6886 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6887 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
6888 (i8 imm:$src3)))]>, VEX_4V;
6891 let Predicates = [HasAVX] in {
6892 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6893 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6894 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6895 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6896 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6897 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6898 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6899 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6900 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6901 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6903 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
6904 (memopv8f32 addr:$src2), (i8 imm:$imm))),
6905 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6906 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
6907 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
6908 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6909 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
6910 (memopv4i64 addr:$src2), (i8 imm:$imm))),
6911 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6912 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
6913 (memopv4f64 addr:$src2), (i8 imm:$imm))),
6914 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6915 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
6916 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
6917 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6918 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
6919 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
6920 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6923 //===----------------------------------------------------------------------===//
6924 // VZERO - Zero YMM registers
6926 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
6927 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
6928 // Zero All YMM registers
6929 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
6930 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
6932 // Zero Upper bits of YMM registers
6933 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
6934 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
6937 //===----------------------------------------------------------------------===//
6938 // Half precision conversion instructions
6939 //===----------------------------------------------------------------------===//
6940 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
6941 let Predicates = [HasAVX, HasF16C] in {
6942 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
6943 "vcvtph2ps\t{$src, $dst|$dst, $src}",
6944 [(set RC:$dst, (Int VR128:$src))]>,
6946 let neverHasSideEffects = 1, mayLoad = 1 in
6947 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
6948 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
6952 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
6953 let Predicates = [HasAVX, HasF16C] in {
6954 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
6955 (ins RC:$src1, i32i8imm:$src2),
6956 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6957 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
6959 let neverHasSideEffects = 1, mayLoad = 1 in
6960 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
6961 (ins RC:$src1, i32i8imm:$src2),
6962 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
6967 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
6968 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
6969 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
6970 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
6972 //===----------------------------------------------------------------------===//
6973 // AVX2 Instructions
6974 //===----------------------------------------------------------------------===//
6976 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
6977 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
6978 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6979 X86MemOperand x86memop> {
6980 let isCommutable = 1 in
6981 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
6982 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6983 !strconcat(OpcodeStr,
6984 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6985 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6987 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
6988 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6989 !strconcat(OpcodeStr,
6990 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6993 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6997 let isCommutable = 0 in {
6998 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
6999 VR128, memopv2i64, i128mem>;
7000 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7001 VR256, memopv4i64, i256mem>;
7004 //===----------------------------------------------------------------------===//
7005 // VPBROADCAST - Load from memory and broadcast to all elements of the
7006 // destination operand
7008 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7009 X86MemOperand x86memop, PatFrag ld_frag,
7010 Intrinsic Int128, Intrinsic Int256> {
7011 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7012 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7013 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7014 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7015 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7017 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7018 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7019 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7020 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7021 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7022 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7024 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7027 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7028 int_x86_avx2_pbroadcastb_128,
7029 int_x86_avx2_pbroadcastb_256>;
7030 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7031 int_x86_avx2_pbroadcastw_128,
7032 int_x86_avx2_pbroadcastw_256>;
7033 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7034 int_x86_avx2_pbroadcastd_128,
7035 int_x86_avx2_pbroadcastd_256>;
7036 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7037 int_x86_avx2_pbroadcastq_128,
7038 int_x86_avx2_pbroadcastq_256>;
7040 let Predicates = [HasAVX2] in {
7041 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7042 (VPBROADCASTBrm addr:$src)>;
7043 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7044 (VPBROADCASTBYrm addr:$src)>;
7045 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7046 (VPBROADCASTWrm addr:$src)>;
7047 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7048 (VPBROADCASTWYrm addr:$src)>;
7049 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7050 (VPBROADCASTDrm addr:$src)>;
7051 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7052 (VPBROADCASTDYrm addr:$src)>;
7053 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7054 (VPBROADCASTQrm addr:$src)>;
7055 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7056 (VPBROADCASTQYrm addr:$src)>;
7059 // AVX1 broadcast patterns
7060 let Predicates = [HasAVX] in {
7061 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7062 (VBROADCASTSSYrm addr:$src)>;
7063 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7064 (VBROADCASTSDrm addr:$src)>;
7065 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7066 (VBROADCASTSSYrm addr:$src)>;
7067 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7068 (VBROADCASTSDrm addr:$src)>;
7070 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7071 (VBROADCASTSSrm addr:$src)>;
7072 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7073 (VBROADCASTSSrm addr:$src)>;
7076 //===----------------------------------------------------------------------===//
7077 // VPERM - Permute instructions
7080 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7082 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7083 (ins VR256:$src1, VR256:$src2),
7084 !strconcat(OpcodeStr,
7085 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7086 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7087 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7088 (ins VR256:$src1, i256mem:$src2),
7089 !strconcat(OpcodeStr,
7090 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7091 [(set VR256:$dst, (Int VR256:$src1,
7092 (bitconvert (mem_frag addr:$src2))))]>,
7096 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7097 let ExeDomain = SSEPackedSingle in
7098 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7100 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7102 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7103 (ins VR256:$src1, i8imm:$src2),
7104 !strconcat(OpcodeStr,
7105 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7106 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7107 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7108 (ins i256mem:$src1, i8imm:$src2),
7109 !strconcat(OpcodeStr,
7110 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7111 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7115 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7117 let ExeDomain = SSEPackedDouble in
7118 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7121 //===----------------------------------------------------------------------===//
7122 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7124 let AddedComplexity = 1 in {
7125 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7126 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7127 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7128 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7129 (i8 imm:$src3))))]>, VEX_4V;
7130 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7131 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7132 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7133 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7134 (i8 imm:$src3)))]>, VEX_4V;
7137 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7138 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7139 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7140 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7141 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7142 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7143 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7145 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7147 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7148 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7149 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7150 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7151 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7153 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7157 //===----------------------------------------------------------------------===//
7158 // VINSERTI128 - Insert packed integer values
7160 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7161 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7162 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7164 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7166 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7167 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7168 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7170 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7171 imm:$src3))]>, VEX_4V;
7173 let Predicates = [HasAVX2] in {
7174 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7176 (VINSERTI128rr VR256:$src1, VR128:$src2,
7177 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7178 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7180 (VINSERTI128rr VR256:$src1, VR128:$src2,
7181 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7182 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7184 (VINSERTI128rr VR256:$src1, VR128:$src2,
7185 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7186 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7188 (VINSERTI128rr VR256:$src1, VR128:$src2,
7189 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7193 let Predicates = [HasAVX] in {
7194 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7196 (VINSERTF128rr VR256:$src1, VR128:$src2,
7197 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7198 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7200 (VINSERTF128rr VR256:$src1, VR128:$src2,
7201 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7202 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7204 (VINSERTF128rr VR256:$src1, VR128:$src2,
7205 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7206 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7208 (VINSERTF128rr VR256:$src1, VR128:$src2,
7209 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7210 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7212 (VINSERTF128rr VR256:$src1, VR128:$src2,
7213 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7214 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7216 (VINSERTF128rr VR256:$src1, VR128:$src2,
7217 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7220 //===----------------------------------------------------------------------===//
7221 // VEXTRACTI128 - Extract packed integer values
7223 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7224 (ins VR256:$src1, i8imm:$src2),
7225 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7227 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7229 let neverHasSideEffects = 1, mayStore = 1 in
7230 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7231 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7232 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7234 let Predicates = [HasAVX2] in {
7235 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7236 (v2i64 (VEXTRACTI128rr
7237 (v4i64 VR256:$src1),
7238 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7239 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7240 (v4i32 (VEXTRACTI128rr
7241 (v8i32 VR256:$src1),
7242 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7243 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7244 (v8i16 (VEXTRACTI128rr
7245 (v16i16 VR256:$src1),
7246 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7247 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7248 (v16i8 (VEXTRACTI128rr
7249 (v32i8 VR256:$src1),
7250 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7254 let Predicates = [HasAVX] in {
7255 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7256 (v4f32 (VEXTRACTF128rr
7257 (v8f32 VR256:$src1),
7258 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7259 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7260 (v2f64 (VEXTRACTF128rr
7261 (v4f64 VR256:$src1),
7262 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7263 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7264 (v2i64 (VEXTRACTF128rr
7265 (v4i64 VR256:$src1),
7266 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7267 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7268 (v4i32 (VEXTRACTF128rr
7269 (v8i32 VR256:$src1),
7270 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7271 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7272 (v8i16 (VEXTRACTF128rr
7273 (v16i16 VR256:$src1),
7274 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7275 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7276 (v16i8 (VEXTRACTF128rr
7277 (v32i8 VR256:$src1),
7278 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7281 //===----------------------------------------------------------------------===//
7282 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7284 multiclass avx2_pmovmask<string OpcodeStr,
7285 Intrinsic IntLd128, Intrinsic IntLd256,
7286 Intrinsic IntSt128, Intrinsic IntSt256> {
7287 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7288 (ins VR128:$src1, i128mem:$src2),
7289 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7290 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7291 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7292 (ins VR256:$src1, i256mem:$src2),
7293 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7294 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7295 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7296 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7297 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7298 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7299 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7300 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7301 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7302 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7305 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7306 int_x86_avx2_maskload_d,
7307 int_x86_avx2_maskload_d_256,
7308 int_x86_avx2_maskstore_d,
7309 int_x86_avx2_maskstore_d_256>;
7310 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7311 int_x86_avx2_maskload_q,
7312 int_x86_avx2_maskload_q_256,
7313 int_x86_avx2_maskstore_q,
7314 int_x86_avx2_maskstore_q_256>, VEX_W;
7317 //===----------------------------------------------------------------------===//
7318 // Variable Bit Shifts
7320 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7321 ValueType vt128, ValueType vt256> {
7322 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7323 (ins VR128:$src1, VR128:$src2),
7324 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7326 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7328 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7329 (ins VR128:$src1, i128mem:$src2),
7330 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7332 (vt128 (OpNode VR128:$src1,
7333 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7335 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7336 (ins VR256:$src1, VR256:$src2),
7337 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7339 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7341 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7342 (ins VR256:$src1, i256mem:$src2),
7343 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7345 (vt256 (OpNode VR256:$src1,
7346 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7350 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7351 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7352 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7353 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7354 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;