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)))], 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)))], d>;
79 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
80 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
81 string OpcodeStr, X86MemOperand x86memop,
82 list<dag> pat_rr, list<dag> pat_rm,
84 let isCommutable = 1 in
85 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
87 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
88 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
90 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
92 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
93 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
97 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
98 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
99 string asm, string SSEVer, string FPSizeStr,
100 X86MemOperand x86memop, PatFrag mem_frag,
101 Domain d, bit Is2Addr = 1> {
102 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
104 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
105 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
106 [(set RC:$dst, (!cast<Intrinsic>(
107 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
108 RC:$src1, RC:$src2))], d>;
109 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
111 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
112 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
113 [(set RC:$dst, (!cast<Intrinsic>(
114 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
115 RC:$src1, (mem_frag addr:$src2)))], d>;
118 //===----------------------------------------------------------------------===//
119 // Non-instruction patterns
120 //===----------------------------------------------------------------------===//
122 // A vector extract of the first f32 position is a subregister copy
123 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
124 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
126 // A 128-bit subvector extract from the first 256-bit vector position
127 // is a subregister copy that needs no instruction.
128 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
129 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
130 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
131 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
133 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
134 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
135 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
136 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
138 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
139 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
140 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
141 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
143 // A 128-bit subvector insert to the first 256-bit vector position
144 // is a subregister copy that needs no instruction.
145 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
146 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
147 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
148 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
149 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
150 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
151 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
152 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
153 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
154 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
155 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
156 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
158 // Implicitly promote a 32-bit scalar to a vector.
159 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
160 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
161 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
162 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
163 // Implicitly promote a 64-bit scalar to a vector.
164 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
165 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
166 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
167 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
169 // Bitcasts between 128-bit vector types. Return the original type since
170 // no instruction is needed for the conversion
171 let Predicates = [HasXMMInt] in {
172 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
173 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
174 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
175 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
176 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
177 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
178 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
179 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
180 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
181 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
182 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
183 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
184 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
185 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
186 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
187 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
188 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
189 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
190 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
191 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
192 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
193 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
194 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
195 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
196 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
197 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
198 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
199 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
200 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
201 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
204 // Bitcasts between 256-bit vector types. Return the original type since
205 // no instruction is needed for the conversion
206 let Predicates = [HasAVX] in {
207 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
208 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
209 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
210 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
211 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
212 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
213 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
214 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
215 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
216 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
217 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
218 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
219 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
220 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
221 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
222 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
223 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
224 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
225 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
226 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
227 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
228 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
229 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
230 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
231 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
232 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
233 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
234 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
235 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
236 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
239 //===----------------------------------------------------------------------===//
240 // AVX & SSE - Zero/One Vectors
241 //===----------------------------------------------------------------------===//
243 // Alias instructions that map zero vector to pxor / xorp* for sse.
244 // We set canFoldAsLoad because this can be converted to a constant-pool
245 // load of an all-zeros value if folding it would be beneficial.
246 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
247 // JIT implementation, it does not expand the instructions below like
248 // X86MCInstLower does.
249 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
250 isCodeGenOnly = 1 in {
251 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
252 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
253 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
254 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
255 let ExeDomain = SSEPackedInt in
256 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
257 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
260 // The same as done above but for AVX. The 128-bit versions are the
261 // same, but re-encoded. The 256-bit does not support PI version, and
262 // doesn't need it because on sandy bridge the register is set to zero
263 // at the rename stage without using any execution unit, so SET0PSY
264 // and SET0PDY can be used for vector int instructions without penalty
265 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
266 // JIT implementatioan, it does not expand the instructions below like
267 // X86MCInstLower does.
268 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
269 isCodeGenOnly = 1, Predicates = [HasAVX] in {
270 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
271 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
272 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
273 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
274 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
275 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
276 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
277 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
278 let ExeDomain = SSEPackedInt in
279 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
280 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
283 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
284 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
285 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
287 // AVX has no support for 256-bit integer instructions, but since the 128-bit
288 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
289 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
290 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
291 (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
293 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
294 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
295 (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
297 //===----------------------------------------------------------------------===//
298 // SSE 1 & 2 - Move FP Scalar Instructions
300 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
301 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
302 // is used instead. Register-to-register movss/movsd is not modeled as an
303 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
304 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
305 //===----------------------------------------------------------------------===//
307 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
308 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
309 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
311 // Loading from memory automatically zeroing upper bits.
312 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
313 PatFrag mem_pat, string OpcodeStr> :
314 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
315 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
316 [(set RC:$dst, (mem_pat addr:$src))]>;
319 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
320 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
321 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
322 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
324 let canFoldAsLoad = 1, isReMaterializable = 1 in {
325 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
326 let AddedComplexity = 20 in
327 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
330 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
331 "movss\t{$src, $dst|$dst, $src}",
332 [(store FR32:$src, addr:$dst)]>, XS, VEX;
333 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
334 "movsd\t{$src, $dst|$dst, $src}",
335 [(store FR64:$src, addr:$dst)]>, XD, VEX;
338 let Constraints = "$src1 = $dst" in {
339 def MOVSSrr : sse12_move_rr<FR32, v4f32,
340 "movss\t{$src2, $dst|$dst, $src2}">, XS;
341 def MOVSDrr : sse12_move_rr<FR64, v2f64,
342 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
345 let canFoldAsLoad = 1, isReMaterializable = 1 in {
346 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
348 let AddedComplexity = 20 in
349 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
352 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
353 "movss\t{$src, $dst|$dst, $src}",
354 [(store FR32:$src, addr:$dst)]>;
355 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
356 "movsd\t{$src, $dst|$dst, $src}",
357 [(store FR64:$src, addr:$dst)]>;
360 let Predicates = [HasSSE1] in {
361 let AddedComplexity = 15 in {
362 // Extract the low 32-bit value from one vector and insert it into another.
363 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
364 (MOVSSrr (v4f32 VR128:$src1),
365 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
366 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
367 (MOVSSrr (v4i32 VR128:$src1),
368 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
370 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
371 // MOVSS to the lower bits.
372 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
373 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
374 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
375 (MOVSSrr (v4f32 (V_SET0PS)),
376 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
377 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
378 (MOVSSrr (v4i32 (V_SET0PI)),
379 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
382 let AddedComplexity = 20 in {
383 // MOVSSrm zeros the high parts of the register; represent this
384 // with SUBREG_TO_REG.
385 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
386 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
387 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
388 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
389 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
390 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
393 // Extract and store.
394 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
397 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
399 // Shuffle with MOVSS
400 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
401 (MOVSSrr VR128:$src1, FR32:$src2)>;
402 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
403 (MOVSSrr (v4i32 VR128:$src1),
404 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
405 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
406 (MOVSSrr (v4f32 VR128:$src1),
407 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
410 let Predicates = [HasSSE2] in {
411 let AddedComplexity = 15 in {
412 // Extract the low 64-bit value from one vector and insert it into another.
413 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
414 (MOVSDrr (v2f64 VR128:$src1),
415 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
416 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
417 (MOVSDrr (v2i64 VR128:$src1),
418 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
420 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
421 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
422 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
423 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
424 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
426 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
427 // MOVSD to the lower bits.
428 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
429 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
432 let AddedComplexity = 20 in {
433 // MOVSDrm zeros the high parts of the register; represent this
434 // with SUBREG_TO_REG.
435 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
436 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
437 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
438 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
439 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
440 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
441 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
442 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
443 def : Pat<(v2f64 (X86vzload addr:$src)),
444 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
447 // Extract and store.
448 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
451 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
453 // Shuffle with MOVSD
454 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
455 (MOVSDrr VR128:$src1, FR64:$src2)>;
456 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
457 (MOVSDrr (v2i64 VR128:$src1),
458 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
459 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
460 (MOVSDrr (v2f64 VR128:$src1),
461 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
462 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
463 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
464 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
465 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
467 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
468 // is during lowering, where it's not possible to recognize the fold cause
469 // it has two uses through a bitcast. One use disappears at isel time and the
470 // fold opportunity reappears.
471 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
472 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
473 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
474 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
477 let Predicates = [HasAVX] in {
478 let AddedComplexity = 15 in {
479 // Extract the low 32-bit value from one vector and insert it into another.
480 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
481 (VMOVSSrr (v4f32 VR128:$src1),
482 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
483 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
484 (VMOVSSrr (v4i32 VR128:$src1),
485 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
487 // Extract the low 64-bit value from one vector and insert it into another.
488 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
489 (VMOVSDrr (v2f64 VR128:$src1),
490 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
491 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
492 (VMOVSDrr (v2i64 VR128:$src1),
493 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
495 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
496 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
497 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
498 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
499 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
501 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
502 // MOVS{S,D} to the lower bits.
503 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
504 (VMOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
505 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
506 (VMOVSSrr (v4f32 (V_SET0PS)),
507 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
508 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
509 (VMOVSSrr (v4i32 (V_SET0PI)),
510 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
511 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
512 (VMOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
515 let AddedComplexity = 20 in {
516 // MOVSSrm zeros the high parts of the register; represent this
517 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
518 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
519 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
520 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
521 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
522 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
523 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
525 // MOVSDrm zeros the high parts of the register; represent this
526 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
527 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
528 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
529 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
530 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
531 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
532 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
533 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
534 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
535 def : Pat<(v2f64 (X86vzload addr:$src)),
536 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
538 // Represent the same patterns above but in the form they appear for
540 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
541 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
542 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
543 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
544 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
545 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
548 // Extract and store.
549 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
552 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
553 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
556 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
558 // Shuffle with VMOVSS
559 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
560 (VMOVSSrr VR128:$src1, FR32:$src2)>;
561 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
562 (VMOVSSrr (v4i32 VR128:$src1),
563 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
564 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
565 (VMOVSSrr (v4f32 VR128:$src1),
566 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
568 // Shuffle with VMOVSD
569 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
570 (VMOVSDrr VR128:$src1, FR64:$src2)>;
571 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
572 (VMOVSDrr (v2i64 VR128:$src1),
573 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
574 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
575 (VMOVSDrr (v2f64 VR128:$src1),
576 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
577 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
578 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
580 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
581 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
584 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
585 // is during lowering, where it's not possible to recognize the fold cause
586 // it has two uses through a bitcast. One use disappears at isel time and the
587 // fold opportunity reappears.
588 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
589 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
591 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
592 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
596 //===----------------------------------------------------------------------===//
597 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
598 //===----------------------------------------------------------------------===//
600 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
601 X86MemOperand x86memop, PatFrag ld_frag,
602 string asm, Domain d,
603 bit IsReMaterializable = 1> {
604 let neverHasSideEffects = 1 in
605 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
606 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
607 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
608 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
609 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
610 [(set RC:$dst, (ld_frag addr:$src))], d>;
613 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
614 "movaps", SSEPackedSingle>, TB, VEX;
615 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
616 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
617 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
618 "movups", SSEPackedSingle>, TB, VEX;
619 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
620 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
622 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
623 "movaps", SSEPackedSingle>, TB, VEX;
624 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
625 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
626 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
627 "movups", SSEPackedSingle>, TB, VEX;
628 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
629 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
630 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
631 "movaps", SSEPackedSingle>, TB;
632 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
633 "movapd", SSEPackedDouble>, TB, OpSize;
634 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
635 "movups", SSEPackedSingle>, TB;
636 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
637 "movupd", SSEPackedDouble, 0>, TB, OpSize;
639 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
640 "movaps\t{$src, $dst|$dst, $src}",
641 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
642 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
643 "movapd\t{$src, $dst|$dst, $src}",
644 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
645 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
646 "movups\t{$src, $dst|$dst, $src}",
647 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
648 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
649 "movupd\t{$src, $dst|$dst, $src}",
650 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
651 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
652 "movaps\t{$src, $dst|$dst, $src}",
653 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
654 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
655 "movapd\t{$src, $dst|$dst, $src}",
656 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
657 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
658 "movups\t{$src, $dst|$dst, $src}",
659 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
660 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
661 "movupd\t{$src, $dst|$dst, $src}",
662 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
664 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
665 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
666 (VMOVUPSYmr addr:$dst, VR256:$src)>;
668 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
669 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
670 (VMOVUPDYmr addr:$dst, VR256:$src)>;
672 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
673 "movaps\t{$src, $dst|$dst, $src}",
674 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
675 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
676 "movapd\t{$src, $dst|$dst, $src}",
677 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
678 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
679 "movups\t{$src, $dst|$dst, $src}",
680 [(store (v4f32 VR128:$src), addr:$dst)]>;
681 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
682 "movupd\t{$src, $dst|$dst, $src}",
683 [(store (v2f64 VR128:$src), addr:$dst)]>;
685 let Predicates = [HasAVX] in {
686 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
687 (VMOVUPSmr addr:$dst, VR128:$src)>;
688 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
689 (VMOVUPDmr addr:$dst, VR128:$src)>;
692 let Predicates = [HasSSE1] in
693 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
694 (MOVUPSmr addr:$dst, VR128:$src)>;
695 let Predicates = [HasSSE2] in
696 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
697 (MOVUPDmr addr:$dst, VR128:$src)>;
699 // Use movaps / movups for SSE integer load / store (one byte shorter).
700 // The instructions selected below are then converted to MOVDQA/MOVDQU
701 // during the SSE domain pass.
702 let Predicates = [HasSSE1] in {
703 def : Pat<(alignedloadv4i32 addr:$src),
704 (MOVAPSrm addr:$src)>;
705 def : Pat<(loadv4i32 addr:$src),
706 (MOVUPSrm addr:$src)>;
707 def : Pat<(alignedloadv2i64 addr:$src),
708 (MOVAPSrm addr:$src)>;
709 def : Pat<(loadv2i64 addr:$src),
710 (MOVUPSrm addr:$src)>;
712 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
713 (MOVAPSmr addr:$dst, VR128:$src)>;
714 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
715 (MOVAPSmr addr:$dst, VR128:$src)>;
716 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
717 (MOVAPSmr addr:$dst, VR128:$src)>;
718 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
719 (MOVAPSmr addr:$dst, VR128:$src)>;
720 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
721 (MOVUPSmr addr:$dst, VR128:$src)>;
722 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
723 (MOVUPSmr addr:$dst, VR128:$src)>;
724 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
725 (MOVUPSmr addr:$dst, VR128:$src)>;
726 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
727 (MOVUPSmr addr:$dst, VR128:$src)>;
730 // Use vmovaps/vmovups for AVX integer load/store.
731 let Predicates = [HasAVX] in {
732 // 128-bit load/store
733 def : Pat<(alignedloadv4i32 addr:$src),
734 (VMOVAPSrm addr:$src)>;
735 def : Pat<(loadv4i32 addr:$src),
736 (VMOVUPSrm addr:$src)>;
737 def : Pat<(alignedloadv2i64 addr:$src),
738 (VMOVAPSrm addr:$src)>;
739 def : Pat<(loadv2i64 addr:$src),
740 (VMOVUPSrm addr:$src)>;
742 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
743 (VMOVAPSmr addr:$dst, VR128:$src)>;
744 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
745 (VMOVAPSmr addr:$dst, VR128:$src)>;
746 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
747 (VMOVAPSmr addr:$dst, VR128:$src)>;
748 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
749 (VMOVAPSmr addr:$dst, VR128:$src)>;
750 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
751 (VMOVUPSmr addr:$dst, VR128:$src)>;
752 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
753 (VMOVUPSmr addr:$dst, VR128:$src)>;
754 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
755 (VMOVUPSmr addr:$dst, VR128:$src)>;
756 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
757 (VMOVUPSmr addr:$dst, VR128:$src)>;
759 // 256-bit load/store
760 def : Pat<(alignedloadv4i64 addr:$src),
761 (VMOVAPSYrm addr:$src)>;
762 def : Pat<(loadv4i64 addr:$src),
763 (VMOVUPSYrm addr:$src)>;
764 def : Pat<(alignedloadv8i32 addr:$src),
765 (VMOVAPSYrm addr:$src)>;
766 def : Pat<(loadv8i32 addr:$src),
767 (VMOVUPSYrm addr:$src)>;
768 def : Pat<(alignedstore (v4i64 VR256:$src), addr:$dst),
769 (VMOVAPSYmr addr:$dst, VR256:$src)>;
770 def : Pat<(alignedstore (v8i32 VR256:$src), addr:$dst),
771 (VMOVAPSYmr addr:$dst, VR256:$src)>;
772 def : Pat<(alignedstore (v16i16 VR256:$src), addr:$dst),
773 (VMOVAPSYmr addr:$dst, VR256:$src)>;
774 def : Pat<(alignedstore (v32i8 VR256:$src), addr:$dst),
775 (VMOVAPSYmr addr:$dst, VR256:$src)>;
776 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
777 (VMOVUPSYmr addr:$dst, VR256:$src)>;
778 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
779 (VMOVUPSYmr addr:$dst, VR256:$src)>;
780 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
781 (VMOVUPSYmr addr:$dst, VR256:$src)>;
782 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
783 (VMOVUPSYmr addr:$dst, VR256:$src)>;
786 //===----------------------------------------------------------------------===//
787 // SSE 1 & 2 - Move Low packed FP Instructions
788 //===----------------------------------------------------------------------===//
790 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
791 PatFrag mov_frag, string base_opc,
793 def PSrm : PI<opc, MRMSrcMem,
794 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
795 !strconcat(base_opc, "s", asm_opr),
798 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
799 SSEPackedSingle>, TB;
801 def PDrm : PI<opc, MRMSrcMem,
802 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
803 !strconcat(base_opc, "d", asm_opr),
804 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
805 (scalar_to_vector (loadf64 addr:$src2)))))],
806 SSEPackedDouble>, TB, OpSize;
809 let AddedComplexity = 20 in {
810 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
811 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
813 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
814 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
815 "\t{$src2, $dst|$dst, $src2}">;
818 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
819 "movlps\t{$src, $dst|$dst, $src}",
820 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
821 (iPTR 0))), addr:$dst)]>, VEX;
822 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
823 "movlpd\t{$src, $dst|$dst, $src}",
824 [(store (f64 (vector_extract (v2f64 VR128:$src),
825 (iPTR 0))), addr:$dst)]>, VEX;
826 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
827 "movlps\t{$src, $dst|$dst, $src}",
828 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
829 (iPTR 0))), addr:$dst)]>;
830 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
831 "movlpd\t{$src, $dst|$dst, $src}",
832 [(store (f64 (vector_extract (v2f64 VR128:$src),
833 (iPTR 0))), addr:$dst)]>;
835 let Predicates = [HasAVX] in {
836 let AddedComplexity = 20 in {
837 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
838 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
839 (VMOVLPSrm VR128:$src1, addr:$src2)>;
840 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
841 (VMOVLPSrm VR128:$src1, addr:$src2)>;
842 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
843 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
844 (VMOVLPDrm VR128:$src1, addr:$src2)>;
845 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
846 (VMOVLPDrm VR128:$src1, addr:$src2)>;
849 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
850 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
851 (VMOVLPSmr addr:$src1, VR128:$src2)>;
852 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
853 VR128:$src2)), addr:$src1),
854 (VMOVLPSmr addr:$src1, VR128:$src2)>;
856 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
857 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
858 (VMOVLPDmr addr:$src1, VR128:$src2)>;
859 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
860 (VMOVLPDmr addr:$src1, VR128:$src2)>;
862 // Shuffle with VMOVLPS
863 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
864 (VMOVLPSrm VR128:$src1, addr:$src2)>;
865 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
866 (VMOVLPSrm VR128:$src1, addr:$src2)>;
867 def : Pat<(X86Movlps VR128:$src1,
868 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
869 (VMOVLPSrm VR128:$src1, addr:$src2)>;
871 // Shuffle with VMOVLPD
872 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
873 (VMOVLPDrm VR128:$src1, addr:$src2)>;
874 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
875 (VMOVLPDrm VR128:$src1, addr:$src2)>;
876 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
877 (scalar_to_vector (loadf64 addr:$src2)))),
878 (VMOVLPDrm VR128:$src1, addr:$src2)>;
881 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
883 (VMOVLPSmr addr:$src1, VR128:$src2)>;
884 def : Pat<(store (v4i32 (X86Movlps
885 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
886 (VMOVLPSmr addr:$src1, VR128:$src2)>;
887 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
889 (VMOVLPDmr addr:$src1, VR128:$src2)>;
890 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
892 (VMOVLPDmr addr:$src1, VR128:$src2)>;
895 let Predicates = [HasSSE1] in {
896 let AddedComplexity = 20 in {
897 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
898 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
899 (MOVLPSrm VR128:$src1, addr:$src2)>;
900 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
901 (MOVLPSrm VR128:$src1, addr:$src2)>;
904 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
905 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
906 (MOVLPSmr addr:$src1, VR128:$src2)>;
907 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
908 VR128:$src2)), addr:$src1),
909 (MOVLPSmr addr:$src1, VR128:$src2)>;
911 // Shuffle with MOVLPS
912 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
913 (MOVLPSrm VR128:$src1, addr:$src2)>;
914 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
915 (MOVLPSrm VR128:$src1, addr:$src2)>;
916 def : Pat<(X86Movlps VR128:$src1,
917 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
918 (MOVLPSrm VR128:$src1, addr:$src2)>;
921 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
923 (MOVLPSmr addr:$src1, VR128:$src2)>;
924 def : Pat<(store (v4i32 (X86Movlps
925 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
927 (MOVLPSmr addr:$src1, VR128:$src2)>;
930 let Predicates = [HasSSE2] in {
931 let AddedComplexity = 20 in {
932 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
933 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
934 (MOVLPDrm VR128:$src1, addr:$src2)>;
935 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
936 (MOVLPDrm VR128:$src1, addr:$src2)>;
939 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
940 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
941 (MOVLPDmr addr:$src1, VR128:$src2)>;
942 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
943 (MOVLPDmr addr:$src1, VR128:$src2)>;
945 // Shuffle with MOVLPD
946 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
947 (MOVLPDrm VR128:$src1, addr:$src2)>;
948 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
949 (MOVLPDrm VR128:$src1, addr:$src2)>;
950 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
951 (scalar_to_vector (loadf64 addr:$src2)))),
952 (MOVLPDrm VR128:$src1, addr:$src2)>;
955 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
957 (MOVLPDmr addr:$src1, VR128:$src2)>;
958 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
960 (MOVLPDmr addr:$src1, VR128:$src2)>;
963 //===----------------------------------------------------------------------===//
964 // SSE 1 & 2 - Move Hi packed FP Instructions
965 //===----------------------------------------------------------------------===//
967 let AddedComplexity = 20 in {
968 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
969 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
971 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
972 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
973 "\t{$src2, $dst|$dst, $src2}">;
976 // v2f64 extract element 1 is always custom lowered to unpack high to low
977 // and extract element 0 so the non-store version isn't too horrible.
978 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
979 "movhps\t{$src, $dst|$dst, $src}",
980 [(store (f64 (vector_extract
981 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
982 (undef)), (iPTR 0))), addr:$dst)]>,
984 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
985 "movhpd\t{$src, $dst|$dst, $src}",
986 [(store (f64 (vector_extract
987 (v2f64 (unpckh VR128:$src, (undef))),
988 (iPTR 0))), addr:$dst)]>,
990 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
991 "movhps\t{$src, $dst|$dst, $src}",
992 [(store (f64 (vector_extract
993 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
994 (undef)), (iPTR 0))), addr:$dst)]>;
995 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
996 "movhpd\t{$src, $dst|$dst, $src}",
997 [(store (f64 (vector_extract
998 (v2f64 (unpckh VR128:$src, (undef))),
999 (iPTR 0))), addr:$dst)]>;
1001 let Predicates = [HasAVX] in {
1003 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1004 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1005 def : Pat<(X86Movlhps VR128:$src1,
1006 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1007 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1008 def : Pat<(X86Movlhps VR128:$src1,
1009 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1010 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1012 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
1013 // is during lowering, where it's not possible to recognize the load fold cause
1014 // it has two uses through a bitcast. One use disappears at isel time and the
1015 // fold opportunity reappears.
1016 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
1017 (scalar_to_vector (loadf64 addr:$src2)))),
1018 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1020 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1021 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1022 (scalar_to_vector (loadf64 addr:$src2)))),
1023 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1026 def : Pat<(store (f64 (vector_extract
1027 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1028 (VMOVHPSmr addr:$dst, VR128:$src)>;
1029 def : Pat<(store (f64 (vector_extract
1030 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1031 (VMOVHPDmr addr:$dst, VR128:$src)>;
1034 let Predicates = [HasSSE1] in {
1036 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1037 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1038 def : Pat<(X86Movlhps VR128:$src1,
1039 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1040 (MOVHPSrm VR128:$src1, addr:$src2)>;
1041 def : Pat<(X86Movlhps VR128:$src1,
1042 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1043 (MOVHPSrm VR128:$src1, addr:$src2)>;
1046 def : Pat<(store (f64 (vector_extract
1047 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1048 (MOVHPSmr addr:$dst, VR128:$src)>;
1051 let Predicates = [HasSSE2] in {
1052 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
1053 // is during lowering, where it's not possible to recognize the load fold cause
1054 // it has two uses through a bitcast. One use disappears at isel time and the
1055 // fold opportunity reappears.
1056 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
1057 (scalar_to_vector (loadf64 addr:$src2)))),
1058 (MOVHPDrm VR128:$src1, addr:$src2)>;
1060 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1061 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1062 (scalar_to_vector (loadf64 addr:$src2)))),
1063 (MOVHPDrm VR128:$src1, addr:$src2)>;
1066 def : Pat<(store (f64 (vector_extract
1067 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
1068 (MOVHPDmr addr:$dst, VR128:$src)>;
1071 //===----------------------------------------------------------------------===//
1072 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1073 //===----------------------------------------------------------------------===//
1075 let AddedComplexity = 20 in {
1076 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1077 (ins VR128:$src1, VR128:$src2),
1078 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1080 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1082 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1083 (ins VR128:$src1, VR128:$src2),
1084 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1086 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1089 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1090 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1091 (ins VR128:$src1, VR128:$src2),
1092 "movlhps\t{$src2, $dst|$dst, $src2}",
1094 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1095 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1096 (ins VR128:$src1, VR128:$src2),
1097 "movhlps\t{$src2, $dst|$dst, $src2}",
1099 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1102 let Predicates = [HasAVX] in {
1104 let AddedComplexity = 20 in {
1105 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1106 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1107 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1108 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1110 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1111 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1112 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1114 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1115 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1116 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1117 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1118 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1119 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1122 let AddedComplexity = 20 in {
1123 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1124 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1125 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1127 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1128 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1129 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1130 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1131 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1134 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1135 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1136 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1137 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1140 let Predicates = [HasSSE1] in {
1142 let AddedComplexity = 20 in {
1143 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1144 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1145 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1146 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1148 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1149 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1150 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1152 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1153 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1154 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1155 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1156 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1157 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1160 let AddedComplexity = 20 in {
1161 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1162 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1163 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1165 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1166 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1167 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1168 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1169 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1172 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1173 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1174 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1175 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1178 //===----------------------------------------------------------------------===//
1179 // SSE 1 & 2 - Conversion Instructions
1180 //===----------------------------------------------------------------------===//
1182 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1183 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1185 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1186 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1187 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1188 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1191 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1192 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1193 string asm, Domain d> {
1194 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1195 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1196 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1197 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1200 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1201 X86MemOperand x86memop, string asm> {
1202 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1203 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1204 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1205 (ins DstRC:$src1, x86memop:$src),
1206 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1209 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1210 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
1211 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1212 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1214 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1215 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
1216 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1217 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1220 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1221 // register, but the same isn't true when only using memory operands,
1222 // provide other assembly "l" and "q" forms to address this explicitly
1223 // where appropriate to do so.
1224 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1226 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1228 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1230 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1232 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1235 let Predicates = [HasAVX] in {
1236 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1237 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1238 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1239 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1240 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1241 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1242 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1243 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1245 def : Pat<(f32 (sint_to_fp GR32:$src)),
1246 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1247 def : Pat<(f32 (sint_to_fp GR64:$src)),
1248 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1249 def : Pat<(f64 (sint_to_fp GR32:$src)),
1250 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1251 def : Pat<(f64 (sint_to_fp GR64:$src)),
1252 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1255 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1256 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1257 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1258 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1259 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1260 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1261 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1262 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1263 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1264 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1265 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1266 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1267 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1268 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1269 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1270 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1272 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1273 // and/or XMM operand(s).
1275 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1276 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1278 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1279 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1280 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1281 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1282 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1283 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1286 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1287 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1288 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1289 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1291 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1292 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1293 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1294 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1295 (ins DstRC:$src1, x86memop:$src2),
1297 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1298 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1299 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1302 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1303 f128mem, load, "cvtsd2si">, XD, VEX;
1304 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1305 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1308 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1309 // Get rid of this hack or rename the intrinsics, there are several
1310 // intructions that only match with the intrinsic form, why create duplicates
1311 // to let them be recognized by the assembler?
1312 let Pattern = []<dag> in {
1313 defm VCVTSD2SI : sse12_cvt_s<0x2D, FR64, GR32, undef, f64mem, load,
1314 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
1315 defm VCVTSD2SI64 : sse12_cvt_s<0x2D, FR64, GR64, undef, f64mem, load,
1316 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
1318 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1319 f128mem, load, "cvtsd2si{l}">, XD;
1320 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1321 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1324 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1325 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1326 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1327 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1329 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1330 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1331 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1332 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1335 let Constraints = "$src1 = $dst" in {
1336 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1337 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1339 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1340 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1341 "cvtsi2ss{q}">, XS, REX_W;
1342 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1343 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1345 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1346 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1347 "cvtsi2sd">, XD, REX_W;
1352 // Aliases for intrinsics
1353 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1354 f32mem, load, "cvttss2si">, XS, VEX;
1355 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1356 int_x86_sse_cvttss2si64, f32mem, load,
1357 "cvttss2si">, XS, VEX, VEX_W;
1358 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1359 f128mem, load, "cvttsd2si">, XD, VEX;
1360 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1361 int_x86_sse2_cvttsd2si64, f128mem, load,
1362 "cvttsd2si">, XD, VEX, VEX_W;
1363 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1364 f32mem, load, "cvttss2si">, XS;
1365 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1366 int_x86_sse_cvttss2si64, f32mem, load,
1367 "cvttss2si{q}">, XS, REX_W;
1368 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1369 f128mem, load, "cvttsd2si">, XD;
1370 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1371 int_x86_sse2_cvttsd2si64, f128mem, load,
1372 "cvttsd2si{q}">, XD, REX_W;
1374 let Pattern = []<dag> in {
1375 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1376 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
1377 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1378 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1380 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1381 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1382 SSEPackedSingle>, TB, VEX;
1383 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1384 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1385 SSEPackedSingle>, TB, VEX;
1388 let Pattern = []<dag> in {
1389 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1390 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1391 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1392 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1393 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1394 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1395 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1398 let Predicates = [HasSSE1] in {
1399 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1400 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1401 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1402 (CVTSS2SIrm addr:$src)>;
1403 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1404 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1405 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1406 (CVTSS2SI64rm addr:$src)>;
1409 let Predicates = [HasAVX] in {
1410 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1411 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1412 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1413 (VCVTSS2SIrm addr:$src)>;
1414 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1415 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1416 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1417 (VCVTSS2SI64rm addr:$src)>;
1422 // Convert scalar double to scalar single
1423 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1424 (ins FR64:$src1, FR64:$src2),
1425 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1427 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1428 (ins FR64:$src1, f64mem:$src2),
1429 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1430 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
1432 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1435 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1436 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1437 [(set FR32:$dst, (fround FR64:$src))]>;
1438 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1439 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1440 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1441 Requires<[HasSSE2, OptForSize]>;
1443 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1444 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1446 let Constraints = "$src1 = $dst" in
1447 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1448 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1450 // Convert scalar single to scalar double
1451 // SSE2 instructions with XS prefix
1452 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1453 (ins FR32:$src1, FR32:$src2),
1454 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1455 []>, XS, Requires<[HasAVX]>, VEX_4V;
1456 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1457 (ins FR32:$src1, f32mem:$src2),
1458 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1459 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
1461 let Predicates = [HasAVX] in {
1462 def : Pat<(f64 (fextend FR32:$src)),
1463 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1464 def : Pat<(fextend (loadf32 addr:$src)),
1465 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1466 def : Pat<(extloadf32 addr:$src),
1467 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1470 def : Pat<(extloadf32 addr:$src),
1471 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1472 Requires<[HasAVX, OptForSpeed]>;
1474 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1475 "cvtss2sd\t{$src, $dst|$dst, $src}",
1476 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1477 Requires<[HasSSE2]>;
1478 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1479 "cvtss2sd\t{$src, $dst|$dst, $src}",
1480 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1481 Requires<[HasSSE2, OptForSize]>;
1483 def : Pat<(extloadf32 addr:$src),
1484 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1486 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1487 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1488 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1489 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1490 VR128:$src2))]>, XS, VEX_4V,
1492 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1493 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1494 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1495 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1496 (load addr:$src2)))]>, XS, VEX_4V,
1498 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1499 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1500 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1501 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1502 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1503 VR128:$src2))]>, XS,
1504 Requires<[HasSSE2]>;
1505 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1506 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1507 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1508 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1509 (load addr:$src2)))]>, XS,
1510 Requires<[HasSSE2]>;
1513 // Convert doubleword to packed single/double fp
1514 // SSE2 instructions without OpSize prefix
1515 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1516 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1517 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1518 TB, VEX, Requires<[HasAVX]>;
1519 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1520 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1521 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1522 (bitconvert (memopv2i64 addr:$src))))]>,
1523 TB, VEX, Requires<[HasAVX]>;
1524 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1525 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1526 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1527 TB, Requires<[HasSSE2]>;
1528 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1529 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1530 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1531 (bitconvert (memopv2i64 addr:$src))))]>,
1532 TB, Requires<[HasSSE2]>;
1534 // FIXME: why the non-intrinsic version is described as SSE3?
1535 // SSE2 instructions with XS prefix
1536 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1537 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1538 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1539 XS, VEX, Requires<[HasAVX]>;
1540 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1541 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1542 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1543 (bitconvert (memopv2i64 addr:$src))))]>,
1544 XS, VEX, Requires<[HasAVX]>;
1545 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1546 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1547 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1548 XS, Requires<[HasSSE2]>;
1549 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1550 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1551 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1552 (bitconvert (memopv2i64 addr:$src))))]>,
1553 XS, Requires<[HasSSE2]>;
1556 // Convert packed single/double fp to doubleword
1557 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1558 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1559 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1560 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1561 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1562 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1563 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1564 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1565 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1566 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1567 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1568 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1570 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1571 "cvtps2dq\t{$src, $dst|$dst, $src}",
1572 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1574 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1576 "cvtps2dq\t{$src, $dst|$dst, $src}",
1577 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1578 (memop addr:$src)))]>, VEX;
1579 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1580 "cvtps2dq\t{$src, $dst|$dst, $src}",
1581 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1582 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1583 "cvtps2dq\t{$src, $dst|$dst, $src}",
1584 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1585 (memop addr:$src)))]>;
1587 // SSE2 packed instructions with XD prefix
1588 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1589 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1590 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1591 XD, VEX, Requires<[HasAVX]>;
1592 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1593 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1594 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1595 (memop addr:$src)))]>,
1596 XD, VEX, Requires<[HasAVX]>;
1597 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1598 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1599 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1600 XD, Requires<[HasSSE2]>;
1601 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1602 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1603 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1604 (memop addr:$src)))]>,
1605 XD, Requires<[HasSSE2]>;
1608 // Convert with truncation packed single/double fp to doubleword
1609 // SSE2 packed instructions with XS prefix
1610 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1611 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1612 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1613 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1614 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1615 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1616 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1617 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1618 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1619 "cvttps2dq\t{$src, $dst|$dst, $src}",
1621 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1622 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1623 "cvttps2dq\t{$src, $dst|$dst, $src}",
1625 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1627 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1628 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1630 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1631 XS, VEX, Requires<[HasAVX]>;
1632 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1633 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1634 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1635 (memop addr:$src)))]>,
1636 XS, VEX, Requires<[HasAVX]>;
1638 let Predicates = [HasSSE2] in {
1639 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1640 (Int_CVTDQ2PSrr VR128:$src)>;
1641 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1642 (CVTTPS2DQrr VR128:$src)>;
1645 let Predicates = [HasAVX] in {
1646 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1647 (Int_VCVTDQ2PSrr VR128:$src)>;
1648 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1649 (VCVTTPS2DQrr VR128:$src)>;
1650 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1651 (VCVTDQ2PSYrr VR256:$src)>;
1652 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1653 (VCVTTPS2DQYrr VR256:$src)>;
1656 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
1658 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1659 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
1661 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
1663 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1664 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1665 (memop addr:$src)))]>, VEX;
1666 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1667 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1668 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1669 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1670 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1671 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1672 (memop addr:$src)))]>;
1674 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1675 // register, but the same isn't true when using memory operands instead.
1676 // Provide other assembly rr and rm forms to address this explicitly.
1677 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1678 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1679 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1680 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1683 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1684 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1685 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1686 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1689 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1690 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1691 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1692 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1694 // Convert packed single to packed double
1695 let Predicates = [HasAVX] in {
1696 // SSE2 instructions without OpSize prefix
1697 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1698 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1699 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1700 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1701 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1702 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1703 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1704 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1706 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1707 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1708 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1709 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1711 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1712 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1713 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1714 TB, VEX, Requires<[HasAVX]>;
1715 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1716 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1717 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1718 (load addr:$src)))]>,
1719 TB, VEX, Requires<[HasAVX]>;
1720 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1721 "cvtps2pd\t{$src, $dst|$dst, $src}",
1722 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1723 TB, Requires<[HasSSE2]>;
1724 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1725 "cvtps2pd\t{$src, $dst|$dst, $src}",
1726 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1727 (load addr:$src)))]>,
1728 TB, Requires<[HasSSE2]>;
1730 // Convert packed double to packed single
1731 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1732 // register, but the same isn't true when using memory operands instead.
1733 // Provide other assembly rr and rm forms to address this explicitly.
1734 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1735 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1736 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1737 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1740 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1741 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1742 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1743 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1746 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1747 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1748 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1749 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1750 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1751 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1752 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1753 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1756 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1757 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1758 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1759 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1761 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1762 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1763 (memop addr:$src)))]>;
1764 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1765 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1766 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1767 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1768 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1769 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1770 (memop addr:$src)))]>;
1772 // AVX 256-bit register conversion intrinsics
1773 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1774 // whenever possible to avoid declaring two versions of each one.
1775 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1776 (VCVTDQ2PSYrr VR256:$src)>;
1777 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1778 (VCVTDQ2PSYrm addr:$src)>;
1780 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1781 (VCVTPD2PSYrr VR256:$src)>;
1782 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1783 (VCVTPD2PSYrm addr:$src)>;
1785 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1786 (VCVTPS2DQYrr VR256:$src)>;
1787 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1788 (VCVTPS2DQYrm addr:$src)>;
1790 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1791 (VCVTPS2PDYrr VR128:$src)>;
1792 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1793 (VCVTPS2PDYrm addr:$src)>;
1795 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1796 (VCVTTPD2DQYrr VR256:$src)>;
1797 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1798 (VCVTTPD2DQYrm addr:$src)>;
1800 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1801 (VCVTTPS2DQYrr VR256:$src)>;
1802 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1803 (VCVTTPS2DQYrm addr:$src)>;
1805 // Match fround and fextend for 128/256-bit conversions
1806 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1807 (VCVTPD2PSYrr VR256:$src)>;
1808 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1809 (VCVTPD2PSYrm addr:$src)>;
1811 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1812 (VCVTPS2PDYrr VR128:$src)>;
1813 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1814 (VCVTPS2PDYrm addr:$src)>;
1816 //===----------------------------------------------------------------------===//
1817 // SSE 1 & 2 - Compare Instructions
1818 //===----------------------------------------------------------------------===//
1820 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1821 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1822 string asm, string asm_alt> {
1823 let isAsmParserOnly = 1 in {
1824 def rr : SIi8<0xC2, MRMSrcReg,
1825 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1828 def rm : SIi8<0xC2, MRMSrcMem,
1829 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1833 // Accept explicit immediate argument form instead of comparison code.
1834 def rr_alt : SIi8<0xC2, MRMSrcReg,
1835 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1838 def rm_alt : SIi8<0xC2, MRMSrcMem,
1839 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1843 let neverHasSideEffects = 1 in {
1844 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1845 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1846 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1848 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1849 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1850 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1854 let Constraints = "$src1 = $dst" in {
1855 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1856 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1857 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1858 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1859 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1860 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1861 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1862 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1863 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1864 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1865 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1866 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1867 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1868 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1869 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1870 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1872 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1873 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1874 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1875 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1876 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1877 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1878 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1879 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1880 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1881 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1882 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1883 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1884 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1887 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1888 Intrinsic Int, string asm> {
1889 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1890 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1891 [(set VR128:$dst, (Int VR128:$src1,
1892 VR128:$src, imm:$cc))]>;
1893 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1894 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1895 [(set VR128:$dst, (Int VR128:$src1,
1896 (load addr:$src), imm:$cc))]>;
1899 // Aliases to match intrinsics which expect XMM operand(s).
1900 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1901 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1903 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1904 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1906 let Constraints = "$src1 = $dst" in {
1907 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1908 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1909 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1910 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1914 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1915 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1916 ValueType vt, X86MemOperand x86memop,
1917 PatFrag ld_frag, string OpcodeStr, Domain d> {
1918 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1919 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1920 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1921 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1922 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1923 [(set EFLAGS, (OpNode (vt RC:$src1),
1924 (ld_frag addr:$src2)))], d>;
1927 let Defs = [EFLAGS] in {
1928 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1929 "ucomiss", SSEPackedSingle>, TB, VEX;
1930 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1931 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
1932 let Pattern = []<dag> in {
1933 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1934 "comiss", SSEPackedSingle>, TB, VEX;
1935 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1936 "comisd", SSEPackedDouble>, TB, OpSize, VEX;
1939 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1940 load, "ucomiss", SSEPackedSingle>, TB, VEX;
1941 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1942 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
1944 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1945 load, "comiss", SSEPackedSingle>, TB, VEX;
1946 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1947 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
1948 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1949 "ucomiss", SSEPackedSingle>, TB;
1950 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1951 "ucomisd", SSEPackedDouble>, TB, OpSize;
1953 let Pattern = []<dag> in {
1954 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1955 "comiss", SSEPackedSingle>, TB;
1956 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1957 "comisd", SSEPackedDouble>, TB, OpSize;
1960 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1961 load, "ucomiss", SSEPackedSingle>, TB;
1962 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1963 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1965 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1966 "comiss", SSEPackedSingle>, TB;
1967 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1968 "comisd", SSEPackedDouble>, TB, OpSize;
1969 } // Defs = [EFLAGS]
1971 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1972 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1973 Intrinsic Int, string asm, string asm_alt,
1975 let isAsmParserOnly = 1 in {
1976 def rri : PIi8<0xC2, MRMSrcReg,
1977 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1978 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1979 def rmi : PIi8<0xC2, MRMSrcMem,
1980 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1981 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1984 // Accept explicit immediate argument form instead of comparison code.
1985 def rri_alt : PIi8<0xC2, MRMSrcReg,
1986 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1988 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1989 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1993 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1994 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1995 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1996 SSEPackedSingle>, TB, VEX_4V;
1997 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1998 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1999 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
2000 SSEPackedDouble>, TB, OpSize, VEX_4V;
2001 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2002 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
2003 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
2004 SSEPackedSingle>, TB, VEX_4V;
2005 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2006 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
2007 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
2008 SSEPackedDouble>, TB, OpSize, VEX_4V;
2009 let Constraints = "$src1 = $dst" in {
2010 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2011 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
2012 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
2013 SSEPackedSingle>, TB;
2014 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2015 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
2016 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
2017 SSEPackedDouble>, TB, OpSize;
2020 let Predicates = [HasSSE1] in {
2021 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2022 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2023 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2024 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2027 let Predicates = [HasSSE2] in {
2028 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2029 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2030 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2031 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2034 let Predicates = [HasAVX] in {
2035 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2036 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2037 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2038 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2039 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2040 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2041 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2042 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2044 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2045 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2046 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2047 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2048 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2049 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2050 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2051 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2054 //===----------------------------------------------------------------------===//
2055 // SSE 1 & 2 - Shuffle Instructions
2056 //===----------------------------------------------------------------------===//
2058 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2059 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2060 ValueType vt, string asm, PatFrag mem_frag,
2061 Domain d, bit IsConvertibleToThreeAddress = 0> {
2062 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2063 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2064 [(set RC:$dst, (vt (shufp:$src3
2065 RC:$src1, (mem_frag addr:$src2))))], d>;
2066 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2067 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2068 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2070 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2073 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2074 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2075 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2076 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2077 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2078 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2079 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2080 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2081 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2082 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2083 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2084 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2086 let Constraints = "$src1 = $dst" in {
2087 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2088 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2089 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2091 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2092 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2093 memopv2f64, SSEPackedDouble>, TB, OpSize;
2096 let Predicates = [HasSSE1] in {
2097 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2098 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2099 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2100 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2101 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2102 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2103 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2104 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2105 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2106 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2107 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2108 // fall back to this for SSE1)
2109 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2110 (SHUFPSrri VR128:$src2, VR128:$src1,
2111 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2112 // Special unary SHUFPSrri case.
2113 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2114 (SHUFPSrri VR128:$src1, VR128:$src1,
2115 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2118 let Predicates = [HasSSE2] in {
2119 // Special binary v4i32 shuffle cases with SHUFPS.
2120 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2121 (SHUFPSrri VR128:$src1, VR128:$src2,
2122 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2123 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2124 (bc_v4i32 (memopv2i64 addr:$src2)))),
2125 (SHUFPSrmi VR128:$src1, addr:$src2,
2126 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2127 // Special unary SHUFPDrri cases.
2128 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2129 (SHUFPDrri VR128:$src1, VR128:$src1,
2130 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2131 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2132 (SHUFPDrri VR128:$src1, VR128:$src1,
2133 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2134 // Special binary v2i64 shuffle cases using SHUFPDrri.
2135 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2136 (SHUFPDrri VR128:$src1, VR128:$src2,
2137 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2138 // Generic SHUFPD patterns
2139 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2140 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2141 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2142 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2143 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2144 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2145 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2148 let Predicates = [HasAVX] in {
2149 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2150 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2151 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2152 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2153 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2154 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2155 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2156 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2157 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2158 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2159 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2160 // fall back to this for SSE1)
2161 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2162 (VSHUFPSrri VR128:$src2, VR128:$src1,
2163 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2164 // Special unary SHUFPSrri case.
2165 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2166 (VSHUFPSrri VR128:$src1, VR128:$src1,
2167 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2168 // Special binary v4i32 shuffle cases with SHUFPS.
2169 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2170 (VSHUFPSrri VR128:$src1, VR128:$src2,
2171 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2172 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2173 (bc_v4i32 (memopv2i64 addr:$src2)))),
2174 (VSHUFPSrmi VR128:$src1, addr:$src2,
2175 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2176 // Special unary SHUFPDrri cases.
2177 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2178 (VSHUFPDrri VR128:$src1, VR128:$src1,
2179 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2180 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2181 (VSHUFPDrri VR128:$src1, VR128:$src1,
2182 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2183 // Special binary v2i64 shuffle cases using SHUFPDrri.
2184 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2185 (VSHUFPDrri VR128:$src1, VR128:$src2,
2186 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2188 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2189 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2190 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2191 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2192 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2193 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2194 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2197 def : Pat<(v8i32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2198 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2199 def : Pat<(v8i32 (X86Shufps VR256:$src1,
2200 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2201 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2203 def : Pat<(v8f32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2204 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2205 def : Pat<(v8f32 (X86Shufps VR256:$src1,
2206 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2207 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2209 def : Pat<(v4i64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2210 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2211 def : Pat<(v4i64 (X86Shufpd VR256:$src1,
2212 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2213 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2215 def : Pat<(v4f64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2216 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2217 def : Pat<(v4f64 (X86Shufpd VR256:$src1,
2218 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2219 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2222 //===----------------------------------------------------------------------===//
2223 // SSE 1 & 2 - Unpack Instructions
2224 //===----------------------------------------------------------------------===//
2226 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2227 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2228 PatFrag mem_frag, RegisterClass RC,
2229 X86MemOperand x86memop, string asm,
2231 def rr : PI<opc, MRMSrcReg,
2232 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2234 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2235 def rm : PI<opc, MRMSrcMem,
2236 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2238 (vt (OpNode RC:$src1,
2239 (mem_frag addr:$src2))))], d>;
2242 let AddedComplexity = 10 in {
2243 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2244 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2245 SSEPackedSingle>, TB, VEX_4V;
2246 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2247 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2248 SSEPackedDouble>, TB, OpSize, VEX_4V;
2249 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2250 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2251 SSEPackedSingle>, TB, VEX_4V;
2252 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2253 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2254 SSEPackedDouble>, TB, OpSize, VEX_4V;
2256 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2257 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2258 SSEPackedSingle>, TB, VEX_4V;
2259 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2260 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2261 SSEPackedDouble>, TB, OpSize, VEX_4V;
2262 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2263 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2264 SSEPackedSingle>, TB, VEX_4V;
2265 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2266 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2267 SSEPackedDouble>, TB, OpSize, VEX_4V;
2269 let Constraints = "$src1 = $dst" in {
2270 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2271 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2272 SSEPackedSingle>, TB;
2273 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2274 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2275 SSEPackedDouble>, TB, OpSize;
2276 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2277 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2278 SSEPackedSingle>, TB;
2279 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2280 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2281 SSEPackedDouble>, TB, OpSize;
2282 } // Constraints = "$src1 = $dst"
2283 } // AddedComplexity
2285 let Predicates = [HasSSE1] in {
2286 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
2287 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2288 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
2289 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2290 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
2291 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2292 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
2293 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2296 let Predicates = [HasSSE2] in {
2297 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
2298 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2299 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
2300 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2301 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
2302 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2303 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
2304 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2306 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
2307 // problem is during lowering, where it's not possible to recognize the load
2308 // fold cause it has two uses through a bitcast. One use disappears at isel
2309 // time and the fold opportunity reappears.
2310 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2311 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2313 let AddedComplexity = 10 in
2314 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2315 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2318 let Predicates = [HasAVX] in {
2319 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
2320 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2321 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
2322 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2323 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
2324 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2325 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
2326 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2328 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
2329 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2330 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
2331 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2332 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
2333 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2334 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, (memopv8i32 addr:$src2))),
2335 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2336 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, (memopv8f32 addr:$src2))),
2337 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2338 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
2339 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2340 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, (memopv8i32 addr:$src2))),
2341 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2342 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
2343 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2345 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
2346 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2347 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
2348 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2349 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
2350 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2351 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
2352 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2354 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
2355 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2356 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
2357 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2358 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, (memopv4i64 addr:$src2))),
2359 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2360 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
2361 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2362 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, (memopv4f64 addr:$src2))),
2363 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2364 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
2365 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2366 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, (memopv4i64 addr:$src2))),
2367 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2368 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
2369 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2371 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
2372 // problem is during lowering, where it's not possible to recognize the load
2373 // fold cause it has two uses through a bitcast. One use disappears at isel
2374 // time and the fold opportunity reappears.
2375 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2376 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2377 let AddedComplexity = 10 in
2378 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2379 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2382 //===----------------------------------------------------------------------===//
2383 // SSE 1 & 2 - Extract Floating-Point Sign mask
2384 //===----------------------------------------------------------------------===//
2386 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2387 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2389 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2390 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2391 [(set GR32:$dst, (Int RC:$src))], d>;
2392 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2393 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2396 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2397 SSEPackedSingle>, TB;
2398 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2399 SSEPackedDouble>, TB, OpSize;
2401 def : Pat<(i32 (X86fgetsign FR32:$src)),
2402 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2403 sub_ss))>, Requires<[HasSSE1]>;
2404 def : Pat<(i64 (X86fgetsign FR32:$src)),
2405 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2406 sub_ss))>, Requires<[HasSSE1]>;
2407 def : Pat<(i32 (X86fgetsign FR64:$src)),
2408 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2409 sub_sd))>, Requires<[HasSSE2]>;
2410 def : Pat<(i64 (X86fgetsign FR64:$src)),
2411 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2412 sub_sd))>, Requires<[HasSSE2]>;
2414 let Predicates = [HasAVX] in {
2415 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2416 "movmskps", SSEPackedSingle>, TB, VEX;
2417 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2418 "movmskpd", SSEPackedDouble>, TB, OpSize,
2420 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2421 "movmskps", SSEPackedSingle>, TB, VEX;
2422 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2423 "movmskpd", SSEPackedDouble>, TB, OpSize,
2426 def : Pat<(i32 (X86fgetsign FR32:$src)),
2427 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2429 def : Pat<(i64 (X86fgetsign FR32:$src)),
2430 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2432 def : Pat<(i32 (X86fgetsign FR64:$src)),
2433 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2435 def : Pat<(i64 (X86fgetsign FR64:$src)),
2436 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2440 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2441 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2442 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2443 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB, OpSize,
2445 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2446 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2447 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2448 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB, OpSize,
2452 //===----------------------------------------------------------------------===//
2453 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
2454 //===----------------------------------------------------------------------===//
2456 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
2457 // names that start with 'Fs'.
2459 // Alias instructions that map fld0 to pxor for sse.
2460 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
2461 canFoldAsLoad = 1 in {
2462 // FIXME: Set encoding to pseudo!
2463 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
2464 [(set FR32:$dst, fp32imm0)]>,
2465 Requires<[HasSSE1]>, TB, OpSize;
2466 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
2467 [(set FR64:$dst, fpimm0)]>,
2468 Requires<[HasSSE2]>, TB, OpSize;
2469 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
2470 [(set FR32:$dst, fp32imm0)]>,
2471 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
2472 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
2473 [(set FR64:$dst, fpimm0)]>,
2474 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
2477 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
2478 // bits are disregarded.
2479 let neverHasSideEffects = 1 in {
2480 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2481 "movaps\t{$src, $dst|$dst, $src}", []>;
2482 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2483 "movapd\t{$src, $dst|$dst, $src}", []>;
2486 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
2487 // bits are disregarded.
2488 let canFoldAsLoad = 1, isReMaterializable = 1 in {
2489 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
2490 "movaps\t{$src, $dst|$dst, $src}",
2491 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
2492 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
2493 "movapd\t{$src, $dst|$dst, $src}",
2494 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
2497 //===----------------------------------------------------------------------===//
2498 // SSE 1 & 2 - Logical Instructions
2499 //===----------------------------------------------------------------------===//
2501 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2503 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2505 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2506 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2508 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2509 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2511 let Constraints = "$src1 = $dst" in {
2512 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2513 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2515 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2516 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2520 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2521 let mayLoad = 0 in {
2522 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2523 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2524 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2527 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2528 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2530 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2532 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2534 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2535 // are all promoted to v2i64, and the patterns are covered by the int
2536 // version. This is needed in SSE only, because v2i64 isn't supported on
2537 // SSE1, but only on SSE2.
2538 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2539 !strconcat(OpcodeStr, "ps"), f128mem, [],
2540 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2541 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2543 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2544 !strconcat(OpcodeStr, "pd"), f128mem,
2545 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2546 (bc_v2i64 (v2f64 VR128:$src2))))],
2547 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2548 (memopv2i64 addr:$src2)))], 0>,
2550 let Constraints = "$src1 = $dst" in {
2551 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2552 !strconcat(OpcodeStr, "ps"), f128mem,
2553 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2554 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2555 (memopv2i64 addr:$src2)))]>, TB;
2557 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2558 !strconcat(OpcodeStr, "pd"), f128mem,
2559 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2560 (bc_v2i64 (v2f64 VR128:$src2))))],
2561 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2562 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2566 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2568 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2570 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2571 !strconcat(OpcodeStr, "ps"), f256mem,
2572 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2573 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2574 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2576 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2577 !strconcat(OpcodeStr, "pd"), f256mem,
2578 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2579 (bc_v4i64 (v4f64 VR256:$src2))))],
2580 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2581 (memopv4i64 addr:$src2)))], 0>,
2585 // AVX 256-bit packed logical ops forms
2586 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2587 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2588 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2589 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2591 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2592 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2593 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2594 let isCommutable = 0 in
2595 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2597 //===----------------------------------------------------------------------===//
2598 // SSE 1 & 2 - Arithmetic Instructions
2599 //===----------------------------------------------------------------------===//
2601 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2604 /// In addition, we also have a special variant of the scalar form here to
2605 /// represent the associated intrinsic operation. This form is unlike the
2606 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2607 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2609 /// These three forms can each be reg+reg or reg+mem.
2612 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2614 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2616 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2617 OpNode, FR32, f32mem, Is2Addr>, XS;
2618 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2619 OpNode, FR64, f64mem, Is2Addr>, XD;
2622 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2624 let mayLoad = 0 in {
2625 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2626 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2627 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2628 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2632 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2634 let mayLoad = 0 in {
2635 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2636 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2637 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2638 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2642 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2644 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2645 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2646 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2647 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2650 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2652 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2653 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2654 SSEPackedSingle, Is2Addr>, TB;
2656 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2657 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2658 SSEPackedDouble, Is2Addr>, TB, OpSize;
2661 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2662 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2663 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2664 SSEPackedSingle, 0>, TB;
2666 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2667 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2668 SSEPackedDouble, 0>, TB, OpSize;
2671 // Binary Arithmetic instructions
2672 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2673 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
2674 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2675 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2676 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2677 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
2678 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2679 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2681 let isCommutable = 0 in {
2682 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2683 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
2684 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2685 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2686 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2687 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
2688 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2689 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2690 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2691 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
2692 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2693 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2694 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2695 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2696 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2697 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
2698 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2699 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2700 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2701 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2704 let Constraints = "$src1 = $dst" in {
2705 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2706 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2707 basic_sse12_fp_binop_s_int<0x58, "add">;
2708 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2709 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2710 basic_sse12_fp_binop_s_int<0x59, "mul">;
2712 let isCommutable = 0 in {
2713 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2714 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2715 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2716 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2717 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2718 basic_sse12_fp_binop_s_int<0x5E, "div">;
2719 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2720 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2721 basic_sse12_fp_binop_s_int<0x5F, "max">,
2722 basic_sse12_fp_binop_p_int<0x5F, "max">;
2723 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2724 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2725 basic_sse12_fp_binop_s_int<0x5D, "min">,
2726 basic_sse12_fp_binop_p_int<0x5D, "min">;
2731 /// In addition, we also have a special variant of the scalar form here to
2732 /// represent the associated intrinsic operation. This form is unlike the
2733 /// plain scalar form, in that it takes an entire vector (instead of a
2734 /// scalar) and leaves the top elements undefined.
2736 /// And, we have a special variant form for a full-vector intrinsic form.
2738 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2739 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2740 SDNode OpNode, Intrinsic F32Int> {
2741 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2742 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2743 [(set FR32:$dst, (OpNode FR32:$src))]>;
2744 // For scalar unary operations, fold a load into the operation
2745 // only in OptForSize mode. It eliminates an instruction, but it also
2746 // eliminates a whole-register clobber (the load), so it introduces a
2747 // partial register update condition.
2748 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2749 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2750 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2751 Requires<[HasSSE1, OptForSize]>;
2752 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2753 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2754 [(set VR128:$dst, (F32Int VR128:$src))]>;
2755 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2756 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2757 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2760 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2761 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2762 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2763 !strconcat(OpcodeStr,
2764 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2765 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2766 !strconcat(OpcodeStr,
2767 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2768 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2769 (ins ssmem:$src1, VR128:$src2),
2770 !strconcat(OpcodeStr,
2771 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2774 /// sse1_fp_unop_p - SSE1 unops in packed form.
2775 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2776 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2777 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2778 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2779 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2780 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2781 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2784 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2785 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2786 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2787 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2788 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2789 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2790 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2791 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2794 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2795 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2796 Intrinsic V4F32Int> {
2797 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2798 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2799 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2800 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2801 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2802 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2805 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2806 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2807 Intrinsic V4F32Int> {
2808 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2809 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2810 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2811 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2812 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2813 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2816 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2817 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2818 SDNode OpNode, Intrinsic F64Int> {
2819 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2820 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2821 [(set FR64:$dst, (OpNode FR64:$src))]>;
2822 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2823 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2824 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2825 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2826 Requires<[HasSSE2, OptForSize]>;
2827 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2828 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2829 [(set VR128:$dst, (F64Int VR128:$src))]>;
2830 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2831 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2832 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2835 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2836 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2837 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2838 !strconcat(OpcodeStr,
2839 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2840 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2841 !strconcat(OpcodeStr,
2842 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2843 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2844 (ins VR128:$src1, sdmem:$src2),
2845 !strconcat(OpcodeStr,
2846 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2849 /// sse2_fp_unop_p - SSE2 unops in vector forms.
2850 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
2852 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2853 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2854 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
2855 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2856 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2857 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
2860 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
2861 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2862 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2863 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2864 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
2865 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2866 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2867 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
2870 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
2871 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2872 Intrinsic V2F64Int> {
2873 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2874 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2875 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
2876 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2877 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2878 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
2881 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
2882 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2883 Intrinsic V2F64Int> {
2884 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2885 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2886 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
2887 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2888 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2889 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
2892 let Predicates = [HasAVX] in {
2894 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
2895 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V;
2897 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
2898 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
2899 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2900 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2901 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
2902 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
2903 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
2904 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
2907 // Reciprocal approximations. Note that these typically require refinement
2908 // in order to obtain suitable precision.
2909 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V;
2910 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
2911 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
2912 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
2913 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
2915 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V;
2916 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
2917 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
2918 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
2919 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
2922 def : Pat<(f32 (fsqrt FR32:$src)),
2923 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2924 def : Pat<(f32 (fsqrt (load addr:$src))),
2925 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2926 Requires<[HasAVX, OptForSize]>;
2927 def : Pat<(f64 (fsqrt FR64:$src)),
2928 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
2929 def : Pat<(f64 (fsqrt (load addr:$src))),
2930 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
2931 Requires<[HasAVX, OptForSize]>;
2933 def : Pat<(f32 (X86frsqrt FR32:$src)),
2934 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2935 def : Pat<(f32 (X86frsqrt (load addr:$src))),
2936 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2937 Requires<[HasAVX, OptForSize]>;
2939 def : Pat<(f32 (X86frcp FR32:$src)),
2940 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2941 def : Pat<(f32 (X86frcp (load addr:$src))),
2942 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2943 Requires<[HasAVX, OptForSize]>;
2945 let Predicates = [HasAVX] in {
2946 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
2947 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2948 (VSQRTSSr (f32 (IMPLICIT_DEF)),
2949 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2951 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
2952 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2954 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
2955 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
2956 (VSQRTSDr (f64 (IMPLICIT_DEF)),
2957 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
2959 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
2960 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
2962 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
2963 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2964 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
2965 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2967 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
2968 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2970 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
2971 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2972 (VRCPSSr (f32 (IMPLICIT_DEF)),
2973 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2975 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
2976 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2980 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
2981 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
2982 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
2983 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
2984 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
2985 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
2987 // Reciprocal approximations. Note that these typically require refinement
2988 // in order to obtain suitable precision.
2989 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
2990 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
2991 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
2992 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
2993 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
2994 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
2996 // There is no f64 version of the reciprocal approximation instructions.
2998 //===----------------------------------------------------------------------===//
2999 // SSE 1 & 2 - Non-temporal stores
3000 //===----------------------------------------------------------------------===//
3002 let AddedComplexity = 400 in { // Prefer non-temporal versions
3003 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3004 (ins f128mem:$dst, VR128:$src),
3005 "movntps\t{$src, $dst|$dst, $src}",
3006 [(alignednontemporalstore (v4f32 VR128:$src),
3008 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3009 (ins f128mem:$dst, VR128:$src),
3010 "movntpd\t{$src, $dst|$dst, $src}",
3011 [(alignednontemporalstore (v2f64 VR128:$src),
3013 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
3014 (ins f128mem:$dst, VR128:$src),
3015 "movntdq\t{$src, $dst|$dst, $src}",
3016 [(alignednontemporalstore (v2f64 VR128:$src),
3019 let ExeDomain = SSEPackedInt in
3020 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3021 (ins f128mem:$dst, VR128:$src),
3022 "movntdq\t{$src, $dst|$dst, $src}",
3023 [(alignednontemporalstore (v4f32 VR128:$src),
3026 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3027 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3029 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3030 (ins f256mem:$dst, VR256:$src),
3031 "movntps\t{$src, $dst|$dst, $src}",
3032 [(alignednontemporalstore (v8f32 VR256:$src),
3034 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3035 (ins f256mem:$dst, VR256:$src),
3036 "movntpd\t{$src, $dst|$dst, $src}",
3037 [(alignednontemporalstore (v4f64 VR256:$src),
3039 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
3040 (ins f256mem:$dst, VR256:$src),
3041 "movntdq\t{$src, $dst|$dst, $src}",
3042 [(alignednontemporalstore (v4f64 VR256:$src),
3044 let ExeDomain = SSEPackedInt in
3045 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3046 (ins f256mem:$dst, VR256:$src),
3047 "movntdq\t{$src, $dst|$dst, $src}",
3048 [(alignednontemporalstore (v8f32 VR256:$src),
3052 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3053 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3054 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3055 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3056 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3057 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3059 let AddedComplexity = 400 in { // Prefer non-temporal versions
3060 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3061 "movntps\t{$src, $dst|$dst, $src}",
3062 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3063 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3064 "movntpd\t{$src, $dst|$dst, $src}",
3065 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3067 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3068 "movntdq\t{$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, $src}",
3074 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3076 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3077 (MOVNTDQmr addr:$dst, VR128:$src)>;
3079 // There is no AVX form for instructions below this point
3080 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$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:$dst)]>,
3087 TB, Requires<[HasSSE2]>;
3090 //===----------------------------------------------------------------------===//
3091 // SSE 1 & 2 - Prefetch and memory fence
3092 //===----------------------------------------------------------------------===//
3094 // Prefetch intrinsic.
3095 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
3096 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
3097 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
3098 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
3099 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
3100 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
3101 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
3102 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
3104 // Load, store, and memory fence
3105 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
3106 TB, Requires<[HasSSE1]>;
3107 def : Pat<(X86SFence), (SFENCE)>;
3109 //===----------------------------------------------------------------------===//
3110 // SSE 1 & 2 - Load/Store XCSR register
3111 //===----------------------------------------------------------------------===//
3113 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3114 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3115 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3116 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3118 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3119 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3120 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3121 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3123 //===---------------------------------------------------------------------===//
3124 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3125 //===---------------------------------------------------------------------===//
3127 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3129 let neverHasSideEffects = 1 in {
3130 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3131 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3132 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3133 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3135 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3136 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
3137 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3138 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
3140 let canFoldAsLoad = 1, mayLoad = 1 in {
3141 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3142 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3143 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3144 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3145 let Predicates = [HasAVX] in {
3146 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3147 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3148 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3149 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3153 let mayStore = 1 in {
3154 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3155 (ins i128mem:$dst, VR128:$src),
3156 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3157 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3158 (ins i256mem:$dst, VR256:$src),
3159 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3160 let Predicates = [HasAVX] in {
3161 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3162 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3163 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3164 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3168 let neverHasSideEffects = 1 in
3169 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3170 "movdqa\t{$src, $dst|$dst, $src}", []>;
3172 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3173 "movdqu\t{$src, $dst|$dst, $src}",
3174 []>, XS, Requires<[HasSSE2]>;
3176 let canFoldAsLoad = 1, mayLoad = 1 in {
3177 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3178 "movdqa\t{$src, $dst|$dst, $src}",
3179 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3180 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3181 "movdqu\t{$src, $dst|$dst, $src}",
3182 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3183 XS, Requires<[HasSSE2]>;
3186 let mayStore = 1 in {
3187 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3188 "movdqa\t{$src, $dst|$dst, $src}",
3189 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3190 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3191 "movdqu\t{$src, $dst|$dst, $src}",
3192 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3193 XS, Requires<[HasSSE2]>;
3196 // Intrinsic forms of MOVDQU load and store
3197 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3198 "vmovdqu\t{$src, $dst|$dst, $src}",
3199 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3200 XS, VEX, Requires<[HasAVX]>;
3202 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3203 "movdqu\t{$src, $dst|$dst, $src}",
3204 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3205 XS, Requires<[HasSSE2]>;
3207 } // ExeDomain = SSEPackedInt
3209 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3210 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3211 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3213 //===---------------------------------------------------------------------===//
3214 // SSE2 - Packed Integer Arithmetic Instructions
3215 //===---------------------------------------------------------------------===//
3217 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3219 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3220 bit IsCommutable = 0, bit Is2Addr = 1> {
3221 let isCommutable = IsCommutable in
3222 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3223 (ins VR128:$src1, VR128:$src2),
3225 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3226 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3227 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
3228 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3229 (ins VR128:$src1, i128mem:$src2),
3231 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3232 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3233 [(set VR128:$dst, (IntId VR128:$src1,
3234 (bitconvert (memopv2i64 addr:$src2))))]>;
3237 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3238 string OpcodeStr, Intrinsic IntId,
3239 Intrinsic IntId2, bit Is2Addr = 1> {
3240 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3241 (ins VR128:$src1, VR128:$src2),
3243 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3244 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3245 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
3246 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3247 (ins VR128:$src1, i128mem:$src2),
3249 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3250 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3251 [(set VR128:$dst, (IntId VR128:$src1,
3252 (bitconvert (memopv2i64 addr:$src2))))]>;
3253 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
3254 (ins VR128:$src1, i32i8imm:$src2),
3256 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3257 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3258 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
3261 /// PDI_binop_rm - Simple SSE2 binary operator.
3262 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
3263 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
3264 let isCommutable = IsCommutable in
3265 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3266 (ins VR128:$src1, VR128:$src2),
3268 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3269 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3270 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
3271 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3272 (ins VR128:$src1, i128mem:$src2),
3274 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3275 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3276 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
3277 (bitconvert (memopv2i64 addr:$src2)))))]>;
3280 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
3282 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
3283 /// to collapse (bitconvert VT to VT) into its operand.
3285 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
3286 bit IsCommutable = 0, bit Is2Addr = 1> {
3287 let isCommutable = IsCommutable in
3288 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3289 (ins VR128:$src1, VR128:$src2),
3291 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3292 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3293 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
3294 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3295 (ins VR128:$src1, i128mem:$src2),
3297 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3298 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3299 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
3302 } // ExeDomain = SSEPackedInt
3304 // 128-bit Integer Arithmetic
3306 let Predicates = [HasAVX] in {
3307 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
3308 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
3309 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
3310 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
3311 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
3312 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
3313 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
3314 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
3315 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
3318 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
3320 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
3322 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
3324 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
3326 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
3328 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
3330 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
3332 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
3334 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
3336 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
3338 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
3340 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
3342 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
3344 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
3346 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
3348 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
3350 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
3352 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
3354 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
3358 let Constraints = "$src1 = $dst" in {
3359 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
3360 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
3361 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
3362 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
3363 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
3364 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
3365 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
3366 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
3367 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
3370 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
3371 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
3372 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
3373 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
3374 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
3375 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
3376 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
3377 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
3378 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
3379 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
3380 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
3381 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
3382 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
3383 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
3384 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
3385 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
3386 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
3387 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
3388 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
3390 } // Constraints = "$src1 = $dst"
3392 //===---------------------------------------------------------------------===//
3393 // SSE2 - Packed Integer Logical Instructions
3394 //===---------------------------------------------------------------------===//
3396 let Predicates = [HasAVX] in {
3397 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3398 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
3400 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3401 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
3403 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3404 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
3407 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3408 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
3410 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3411 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
3413 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3414 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
3417 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3418 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
3420 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3421 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
3424 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
3425 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
3426 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
3428 let ExeDomain = SSEPackedInt in {
3429 let neverHasSideEffects = 1 in {
3430 // 128-bit logical shifts.
3431 def VPSLLDQri : PDIi8<0x73, MRM7r,
3432 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3433 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3435 def VPSRLDQri : PDIi8<0x73, MRM3r,
3436 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3437 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3439 // PSRADQri doesn't exist in SSE[1-3].
3441 def VPANDNrr : PDI<0xDF, MRMSrcReg,
3442 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3443 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3445 (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
3447 def VPANDNrm : PDI<0xDF, MRMSrcMem,
3448 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3449 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3450 [(set VR128:$dst, (X86andnp VR128:$src1,
3451 (memopv2i64 addr:$src2)))]>, VEX_4V;
3455 let Constraints = "$src1 = $dst" in {
3456 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3457 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
3458 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3459 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
3460 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3461 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
3463 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3464 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
3465 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3466 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
3467 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3468 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
3470 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3471 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
3472 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3473 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
3475 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
3476 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
3477 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
3479 let ExeDomain = SSEPackedInt in {
3480 let neverHasSideEffects = 1 in {
3481 // 128-bit logical shifts.
3482 def PSLLDQri : PDIi8<0x73, MRM7r,
3483 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3484 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3485 def PSRLDQri : PDIi8<0x73, MRM3r,
3486 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3487 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3488 // PSRADQri doesn't exist in SSE[1-3].
3490 def PANDNrr : PDI<0xDF, MRMSrcReg,
3491 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3492 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3494 def PANDNrm : PDI<0xDF, MRMSrcMem,
3495 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3496 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3498 } // Constraints = "$src1 = $dst"
3500 let Predicates = [HasAVX] in {
3501 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3502 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3503 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3504 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3505 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3506 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
3507 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3508 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
3509 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3510 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3512 // Shift up / down and insert zero's.
3513 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3514 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3515 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3516 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3519 let Predicates = [HasSSE2] in {
3520 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3521 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3522 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3523 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3524 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3525 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
3526 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3527 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
3528 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3529 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3531 // Shift up / down and insert zero's.
3532 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3533 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3534 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3535 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3538 //===---------------------------------------------------------------------===//
3539 // SSE2 - Packed Integer Comparison Instructions
3540 //===---------------------------------------------------------------------===//
3542 let Predicates = [HasAVX] in {
3543 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
3545 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
3547 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
3549 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
3551 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
3553 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
3556 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3557 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3558 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3559 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3560 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3561 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3562 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3563 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3564 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3565 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3566 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3567 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3569 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3570 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3571 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3572 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3573 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3574 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3575 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3576 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3577 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3578 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3579 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3580 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3583 let Constraints = "$src1 = $dst" in {
3584 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
3585 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
3586 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
3587 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
3588 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
3589 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
3590 } // Constraints = "$src1 = $dst"
3592 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3593 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3594 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3595 (PCMPEQBrm VR128:$src1, addr:$src2)>;
3596 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3597 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
3598 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3599 (PCMPEQWrm VR128:$src1, addr:$src2)>;
3600 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3601 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
3602 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3603 (PCMPEQDrm VR128:$src1, addr:$src2)>;
3605 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3606 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
3607 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3608 (PCMPGTBrm VR128:$src1, addr:$src2)>;
3609 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3610 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
3611 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3612 (PCMPGTWrm VR128:$src1, addr:$src2)>;
3613 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3614 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
3615 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3616 (PCMPGTDrm VR128:$src1, addr:$src2)>;
3618 //===---------------------------------------------------------------------===//
3619 // SSE2 - Packed Integer Pack Instructions
3620 //===---------------------------------------------------------------------===//
3622 let Predicates = [HasAVX] in {
3623 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3625 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3627 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3631 let Constraints = "$src1 = $dst" in {
3632 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
3633 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
3634 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
3635 } // Constraints = "$src1 = $dst"
3637 //===---------------------------------------------------------------------===//
3638 // SSE2 - Packed Integer Shuffle Instructions
3639 //===---------------------------------------------------------------------===//
3641 let ExeDomain = SSEPackedInt in {
3642 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
3644 def ri : Ii8<0x70, MRMSrcReg,
3645 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3646 !strconcat(OpcodeStr,
3647 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3648 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
3650 def mi : Ii8<0x70, MRMSrcMem,
3651 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3652 !strconcat(OpcodeStr,
3653 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3654 [(set VR128:$dst, (vt (pshuf_frag:$src2
3655 (bc_frag (memopv2i64 addr:$src1)),
3658 } // ExeDomain = SSEPackedInt
3660 let Predicates = [HasAVX] in {
3661 let AddedComplexity = 5 in
3662 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
3665 // SSE2 with ImmT == Imm8 and XS prefix.
3666 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
3669 // SSE2 with ImmT == Imm8 and XD prefix.
3670 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
3673 let AddedComplexity = 5 in
3674 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3675 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3676 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
3677 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3678 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3680 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3682 (VPSHUFDmi addr:$src1, imm:$imm)>;
3683 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3685 (VPSHUFDmi addr:$src1, imm:$imm)>;
3686 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3687 (VPSHUFDri VR128:$src1, imm:$imm)>;
3688 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3689 (VPSHUFDri VR128:$src1, imm:$imm)>;
3690 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3691 (VPSHUFHWri VR128:$src, imm:$imm)>;
3692 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3694 (VPSHUFHWmi addr:$src, imm:$imm)>;
3695 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3696 (VPSHUFLWri VR128:$src, imm:$imm)>;
3697 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3699 (VPSHUFLWmi addr:$src, imm:$imm)>;
3702 let Predicates = [HasSSE2] in {
3703 let AddedComplexity = 5 in
3704 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
3706 // SSE2 with ImmT == Imm8 and XS prefix.
3707 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
3709 // SSE2 with ImmT == Imm8 and XD prefix.
3710 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
3712 let AddedComplexity = 5 in
3713 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3714 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3715 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3716 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3717 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3719 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3721 (PSHUFDmi addr:$src1, imm:$imm)>;
3722 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3724 (PSHUFDmi addr:$src1, imm:$imm)>;
3725 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3726 (PSHUFDri VR128:$src1, imm:$imm)>;
3727 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3728 (PSHUFDri VR128:$src1, imm:$imm)>;
3729 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3730 (PSHUFHWri VR128:$src, imm:$imm)>;
3731 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3733 (PSHUFHWmi addr:$src, imm:$imm)>;
3734 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3735 (PSHUFLWri VR128:$src, imm:$imm)>;
3736 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3738 (PSHUFLWmi addr:$src, imm:$imm)>;
3741 //===---------------------------------------------------------------------===//
3742 // SSE2 - Packed Integer Unpack Instructions
3743 //===---------------------------------------------------------------------===//
3745 let ExeDomain = SSEPackedInt in {
3746 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3747 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3748 def rr : PDI<opc, MRMSrcReg,
3749 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3751 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3752 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3753 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
3754 def rm : PDI<opc, MRMSrcMem,
3755 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3757 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3758 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3759 [(set VR128:$dst, (OpNode VR128:$src1,
3760 (bc_frag (memopv2i64
3764 let Predicates = [HasAVX] in {
3765 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Punpcklbw,
3766 bc_v16i8, 0>, VEX_4V;
3767 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Punpcklwd,
3768 bc_v8i16, 0>, VEX_4V;
3769 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Punpckldq,
3770 bc_v4i32, 0>, VEX_4V;
3772 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3773 /// knew to collapse (bitconvert VT to VT) into its operand.
3774 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3775 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3776 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3777 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3778 VR128:$src2)))]>, VEX_4V;
3779 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3780 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3781 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3782 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3783 (memopv2i64 addr:$src2))))]>, VEX_4V;
3785 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Punpckhbw,
3786 bc_v16i8, 0>, VEX_4V;
3787 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Punpckhwd,
3788 bc_v8i16, 0>, VEX_4V;
3789 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Punpckhdq,
3790 bc_v4i32, 0>, VEX_4V;
3792 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3793 /// knew to collapse (bitconvert VT to VT) into its operand.
3794 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3795 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3796 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3797 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3798 VR128:$src2)))]>, VEX_4V;
3799 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3800 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3801 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3802 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3803 (memopv2i64 addr:$src2))))]>, VEX_4V;
3806 let Constraints = "$src1 = $dst" in {
3807 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Punpcklbw, bc_v16i8>;
3808 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Punpcklwd, bc_v8i16>;
3809 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Punpckldq, bc_v4i32>;
3811 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3812 /// knew to collapse (bitconvert VT to VT) into its operand.
3813 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3814 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3815 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3817 (v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)))]>;
3818 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3819 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3820 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3822 (v2i64 (X86Punpcklqdq VR128:$src1,
3823 (memopv2i64 addr:$src2))))]>;
3825 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Punpckhbw, bc_v16i8>;
3826 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Punpckhwd, bc_v8i16>;
3827 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Punpckhdq, bc_v4i32>;
3829 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3830 /// knew to collapse (bitconvert VT to VT) into its operand.
3831 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3832 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3833 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3835 (v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)))]>;
3836 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3837 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3838 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3840 (v2i64 (X86Punpckhqdq VR128:$src1,
3841 (memopv2i64 addr:$src2))))]>;
3844 } // ExeDomain = SSEPackedInt
3846 //===---------------------------------------------------------------------===//
3847 // SSE2 - Packed Integer Extract and Insert
3848 //===---------------------------------------------------------------------===//
3850 let ExeDomain = SSEPackedInt in {
3851 multiclass sse2_pinsrw<bit Is2Addr = 1> {
3852 def rri : Ii8<0xC4, MRMSrcReg,
3853 (outs VR128:$dst), (ins VR128:$src1,
3854 GR32:$src2, i32i8imm:$src3),
3856 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3857 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3859 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
3860 def rmi : Ii8<0xC4, MRMSrcMem,
3861 (outs VR128:$dst), (ins VR128:$src1,
3862 i16mem:$src2, i32i8imm:$src3),
3864 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3865 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3867 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
3872 let Predicates = [HasAVX] in
3873 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
3874 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3875 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3876 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3877 imm:$src2))]>, TB, OpSize, VEX;
3878 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
3879 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3880 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3881 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3885 let Predicates = [HasAVX] in {
3886 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
3887 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
3888 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
3889 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
3890 []>, TB, OpSize, VEX_4V;
3893 let Constraints = "$src1 = $dst" in
3894 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
3896 } // ExeDomain = SSEPackedInt
3898 //===---------------------------------------------------------------------===//
3899 // SSE2 - Packed Mask Creation
3900 //===---------------------------------------------------------------------===//
3902 let ExeDomain = SSEPackedInt in {
3904 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3905 "pmovmskb\t{$src, $dst|$dst, $src}",
3906 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
3907 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
3908 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
3909 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3910 "pmovmskb\t{$src, $dst|$dst, $src}",
3911 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
3913 } // ExeDomain = SSEPackedInt
3915 //===---------------------------------------------------------------------===//
3916 // SSE2 - Conditional Store
3917 //===---------------------------------------------------------------------===//
3919 let ExeDomain = SSEPackedInt in {
3922 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
3923 (ins VR128:$src, VR128:$mask),
3924 "maskmovdqu\t{$mask, $src|$src, $mask}",
3925 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
3927 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
3928 (ins VR128:$src, VR128:$mask),
3929 "maskmovdqu\t{$mask, $src|$src, $mask}",
3930 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
3933 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3934 "maskmovdqu\t{$mask, $src|$src, $mask}",
3935 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
3937 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3938 "maskmovdqu\t{$mask, $src|$src, $mask}",
3939 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
3941 } // ExeDomain = SSEPackedInt
3943 //===---------------------------------------------------------------------===//
3944 // SSE2 - Move Doubleword
3945 //===---------------------------------------------------------------------===//
3947 //===---------------------------------------------------------------------===//
3948 // Move Int Doubleword to Packed Double Int
3950 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3951 "movd\t{$src, $dst|$dst, $src}",
3953 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
3954 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3955 "movd\t{$src, $dst|$dst, $src}",
3957 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
3959 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3960 "mov{d|q}\t{$src, $dst|$dst, $src}",
3962 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
3963 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3964 "mov{d|q}\t{$src, $dst|$dst, $src}",
3965 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
3967 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3968 "movd\t{$src, $dst|$dst, $src}",
3970 (v4i32 (scalar_to_vector GR32:$src)))]>;
3971 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3972 "movd\t{$src, $dst|$dst, $src}",
3974 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
3975 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3976 "mov{d|q}\t{$src, $dst|$dst, $src}",
3978 (v2i64 (scalar_to_vector GR64:$src)))]>;
3979 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3980 "mov{d|q}\t{$src, $dst|$dst, $src}",
3981 [(set FR64:$dst, (bitconvert GR64:$src))]>;
3983 //===---------------------------------------------------------------------===//
3984 // Move Int Doubleword to Single Scalar
3986 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3987 "movd\t{$src, $dst|$dst, $src}",
3988 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
3990 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3991 "movd\t{$src, $dst|$dst, $src}",
3992 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
3994 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3995 "movd\t{$src, $dst|$dst, $src}",
3996 [(set FR32:$dst, (bitconvert GR32:$src))]>;
3998 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3999 "movd\t{$src, $dst|$dst, $src}",
4000 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4002 //===---------------------------------------------------------------------===//
4003 // Move Packed Doubleword Int to Packed Double Int
4005 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4006 "movd\t{$src, $dst|$dst, $src}",
4007 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4009 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4010 (ins i32mem:$dst, VR128:$src),
4011 "movd\t{$src, $dst|$dst, $src}",
4012 [(store (i32 (vector_extract (v4i32 VR128:$src),
4013 (iPTR 0))), addr:$dst)]>, VEX;
4014 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4015 "movd\t{$src, $dst|$dst, $src}",
4016 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4018 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4019 "movd\t{$src, $dst|$dst, $src}",
4020 [(store (i32 (vector_extract (v4i32 VR128:$src),
4021 (iPTR 0))), addr:$dst)]>;
4023 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4024 "mov{d|q}\t{$src, $dst|$dst, $src}",
4025 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4027 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4028 "movq\t{$src, $dst|$dst, $src}",
4029 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4031 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4032 "mov{d|q}\t{$src, $dst|$dst, $src}",
4033 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4034 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4035 "movq\t{$src, $dst|$dst, $src}",
4036 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4038 //===---------------------------------------------------------------------===//
4039 // Move Scalar Single to Double Int
4041 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4042 "movd\t{$src, $dst|$dst, $src}",
4043 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4044 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4045 "movd\t{$src, $dst|$dst, $src}",
4046 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4047 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4048 "movd\t{$src, $dst|$dst, $src}",
4049 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4050 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4051 "movd\t{$src, $dst|$dst, $src}",
4052 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4054 //===---------------------------------------------------------------------===//
4055 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4057 let AddedComplexity = 15 in {
4058 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4059 "movd\t{$src, $dst|$dst, $src}",
4060 [(set VR128:$dst, (v4i32 (X86vzmovl
4061 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4063 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4064 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4065 [(set VR128:$dst, (v2i64 (X86vzmovl
4066 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4069 let AddedComplexity = 15 in {
4070 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4071 "movd\t{$src, $dst|$dst, $src}",
4072 [(set VR128:$dst, (v4i32 (X86vzmovl
4073 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4074 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4075 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4076 [(set VR128:$dst, (v2i64 (X86vzmovl
4077 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4080 let AddedComplexity = 20 in {
4081 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4082 "movd\t{$src, $dst|$dst, $src}",
4084 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4085 (loadi32 addr:$src))))))]>,
4087 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4088 "movd\t{$src, $dst|$dst, $src}",
4090 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4091 (loadi32 addr:$src))))))]>;
4093 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4094 (MOVZDI2PDIrm addr:$src)>;
4095 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4096 (MOVZDI2PDIrm addr:$src)>;
4097 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4098 (MOVZDI2PDIrm addr:$src)>;
4101 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4102 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4103 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4104 (v4i32 (scalar_to_vector GR32:$src)), (i32 0)))),
4105 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4106 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4107 (v2i64 (scalar_to_vector GR64:$src)), (i32 0)))),
4108 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4110 // These are the correct encodings of the instructions so that we know how to
4111 // read correct assembly, even though we continue to emit the wrong ones for
4112 // compatibility with Darwin's buggy assembler.
4113 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4114 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4115 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4116 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4117 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4118 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4119 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4120 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4121 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4122 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4123 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4124 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4126 //===---------------------------------------------------------------------===//
4127 // SSE2 - Move Quadword
4128 //===---------------------------------------------------------------------===//
4130 //===---------------------------------------------------------------------===//
4131 // Move Quadword Int to Packed Quadword Int
4133 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4134 "vmovq\t{$src, $dst|$dst, $src}",
4136 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4137 VEX, Requires<[HasAVX]>;
4138 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4139 "movq\t{$src, $dst|$dst, $src}",
4141 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4142 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4144 //===---------------------------------------------------------------------===//
4145 // Move Packed Quadword Int to Quadword Int
4147 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4148 "movq\t{$src, $dst|$dst, $src}",
4149 [(store (i64 (vector_extract (v2i64 VR128:$src),
4150 (iPTR 0))), addr:$dst)]>, VEX;
4151 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4152 "movq\t{$src, $dst|$dst, $src}",
4153 [(store (i64 (vector_extract (v2i64 VR128:$src),
4154 (iPTR 0))), addr:$dst)]>;
4156 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
4157 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
4159 //===---------------------------------------------------------------------===//
4160 // Store / copy lower 64-bits of a XMM register.
4162 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4163 "movq\t{$src, $dst|$dst, $src}",
4164 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4165 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4166 "movq\t{$src, $dst|$dst, $src}",
4167 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4169 let AddedComplexity = 20 in
4170 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4171 "vmovq\t{$src, $dst|$dst, $src}",
4173 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4174 (loadi64 addr:$src))))))]>,
4175 XS, VEX, Requires<[HasAVX]>;
4177 let AddedComplexity = 20 in {
4178 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4179 "movq\t{$src, $dst|$dst, $src}",
4181 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4182 (loadi64 addr:$src))))))]>,
4183 XS, Requires<[HasSSE2]>;
4185 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4186 (MOVZQI2PQIrm addr:$src)>;
4187 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4188 (MOVZQI2PQIrm addr:$src)>;
4189 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4192 //===---------------------------------------------------------------------===//
4193 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4194 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4196 let AddedComplexity = 15 in
4197 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4198 "vmovq\t{$src, $dst|$dst, $src}",
4199 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4200 XS, VEX, Requires<[HasAVX]>;
4201 let AddedComplexity = 15 in
4202 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4203 "movq\t{$src, $dst|$dst, $src}",
4204 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4205 XS, Requires<[HasSSE2]>;
4207 let AddedComplexity = 20 in
4208 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4209 "vmovq\t{$src, $dst|$dst, $src}",
4210 [(set VR128:$dst, (v2i64 (X86vzmovl
4211 (loadv2i64 addr:$src))))]>,
4212 XS, VEX, Requires<[HasAVX]>;
4213 let AddedComplexity = 20 in {
4214 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4215 "movq\t{$src, $dst|$dst, $src}",
4216 [(set VR128:$dst, (v2i64 (X86vzmovl
4217 (loadv2i64 addr:$src))))]>,
4218 XS, Requires<[HasSSE2]>;
4220 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4221 (MOVZPQILo2PQIrm addr:$src)>;
4224 // Instructions to match in the assembler
4225 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4226 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4227 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4228 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4229 // Recognize "movd" with GR64 destination, but encode as a "movq"
4230 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4231 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4233 // Instructions for the disassembler
4234 // xr = XMM register
4237 let Predicates = [HasAVX] in
4238 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4239 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4240 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4241 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4243 //===---------------------------------------------------------------------===//
4244 // SSE2 - Misc Instructions
4245 //===---------------------------------------------------------------------===//
4248 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
4249 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
4250 TB, Requires<[HasSSE2]>;
4252 // Load, store, and memory fence
4253 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
4254 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
4255 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
4256 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
4257 def : Pat<(X86LFence), (LFENCE)>;
4258 def : Pat<(X86MFence), (MFENCE)>;
4261 // Pause. This "instruction" is encoded as "rep; nop", so even though it
4262 // was introduced with SSE2, it's backward compatible.
4263 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
4265 // Alias instructions that map zero vector to pxor / xorp* for sse.
4266 // We set canFoldAsLoad because this can be converted to a constant-pool
4267 // load of an all-ones value if folding it would be beneficial.
4268 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
4269 // JIT implementation, it does not expand the instructions below like
4270 // X86MCInstLower does.
4271 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
4272 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
4273 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
4274 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
4275 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
4276 isCodeGenOnly = 1, ExeDomain = SSEPackedInt, Predicates = [HasAVX] in
4277 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
4278 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
4280 //===---------------------------------------------------------------------===//
4281 // SSE3 - Conversion Instructions
4282 //===---------------------------------------------------------------------===//
4284 // Convert Packed Double FP to Packed DW Integers
4285 let Predicates = [HasAVX] in {
4286 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4287 // register, but the same isn't true when using memory operands instead.
4288 // Provide other assembly rr and rm forms to address this explicitly.
4289 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4290 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4291 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4292 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4295 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4296 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4297 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4298 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4301 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4302 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4303 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4304 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4307 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4308 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4309 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4310 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4312 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4313 (VCVTPD2DQYrr VR256:$src)>;
4314 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4315 (VCVTPD2DQYrm addr:$src)>;
4317 // Convert Packed DW Integers to Packed Double FP
4318 let Predicates = [HasAVX] in {
4319 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4320 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4321 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4322 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4323 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4324 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4325 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4326 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4329 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4330 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4331 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4332 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4334 // AVX 256-bit register conversion intrinsics
4335 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4336 (VCVTDQ2PDYrr VR128:$src)>;
4337 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
4338 (VCVTDQ2PDYrm addr:$src)>;
4340 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4341 (VCVTPD2DQYrr VR256:$src)>;
4342 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4343 (VCVTPD2DQYrm addr:$src)>;
4345 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4346 (VCVTDQ2PDYrr VR128:$src)>;
4347 def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
4348 (VCVTDQ2PDYrm addr:$src)>;
4350 //===---------------------------------------------------------------------===//
4351 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4352 //===---------------------------------------------------------------------===//
4353 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4354 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4355 X86MemOperand x86memop> {
4356 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4357 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4358 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4359 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4360 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4361 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4364 let Predicates = [HasAVX] in {
4365 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4366 v4f32, VR128, memopv4f32, f128mem>, VEX;
4367 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4368 v4f32, VR128, memopv4f32, f128mem>, VEX;
4369 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4370 v8f32, VR256, memopv8f32, f256mem>, VEX;
4371 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4372 v8f32, VR256, memopv8f32, f256mem>, VEX;
4374 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4375 memopv4f32, f128mem>;
4376 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4377 memopv4f32, f128mem>;
4379 let Predicates = [HasSSE3] in {
4380 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4381 (MOVSHDUPrr VR128:$src)>;
4382 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4383 (MOVSHDUPrm addr:$src)>;
4384 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4385 (MOVSLDUPrr VR128:$src)>;
4386 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4387 (MOVSLDUPrm addr:$src)>;
4390 let Predicates = [HasAVX] in {
4391 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4392 (VMOVSHDUPrr VR128:$src)>;
4393 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4394 (VMOVSHDUPrm addr:$src)>;
4395 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4396 (VMOVSLDUPrr VR128:$src)>;
4397 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4398 (VMOVSLDUPrm addr:$src)>;
4399 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4400 (VMOVSHDUPYrr VR256:$src)>;
4401 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4402 (VMOVSHDUPYrm addr:$src)>;
4403 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4404 (VMOVSLDUPYrr VR256:$src)>;
4405 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4406 (VMOVSLDUPYrm addr:$src)>;
4409 //===---------------------------------------------------------------------===//
4410 // SSE3 - Replicate Double FP - MOVDDUP
4411 //===---------------------------------------------------------------------===//
4413 multiclass sse3_replicate_dfp<string OpcodeStr> {
4414 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4415 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4416 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4417 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4418 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4420 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4424 // FIXME: Merge with above classe when there're patterns for the ymm version
4425 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4426 let Predicates = [HasAVX] in {
4427 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4428 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4430 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4431 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4436 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4437 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4438 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4440 let Predicates = [HasSSE3] in {
4441 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4443 (MOVDDUPrm addr:$src)>;
4444 let AddedComplexity = 5 in {
4445 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4446 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4447 (MOVDDUPrm addr:$src)>;
4448 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4449 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4450 (MOVDDUPrm addr:$src)>;
4452 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4453 (MOVDDUPrm addr:$src)>;
4454 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4455 (MOVDDUPrm addr:$src)>;
4456 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4457 (MOVDDUPrm addr:$src)>;
4458 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4459 (MOVDDUPrm addr:$src)>;
4460 def : Pat<(X86Movddup (bc_v2f64
4461 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4462 (MOVDDUPrm addr:$src)>;
4465 let Predicates = [HasAVX] in {
4466 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4468 (VMOVDDUPrm addr:$src)>;
4469 let AddedComplexity = 5 in {
4470 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4471 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4472 (VMOVDDUPrm addr:$src)>;
4473 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4474 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4475 (VMOVDDUPrm addr:$src)>;
4477 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4478 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4479 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4480 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4481 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4482 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4483 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4484 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4485 def : Pat<(X86Movddup (bc_v2f64
4486 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4487 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4490 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4491 (VMOVDDUPYrm addr:$src)>;
4492 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4493 (VMOVDDUPYrm addr:$src)>;
4494 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4495 (VMOVDDUPYrm addr:$src)>;
4496 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4497 (VMOVDDUPYrm addr:$src)>;
4498 def : Pat<(X86Movddup (v4f64 VR256:$src)),
4499 (VMOVDDUPYrr VR256:$src)>;
4500 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4501 (VMOVDDUPYrr VR256:$src)>;
4504 //===---------------------------------------------------------------------===//
4505 // SSE3 - Move Unaligned Integer
4506 //===---------------------------------------------------------------------===//
4508 let Predicates = [HasAVX] in {
4509 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4510 "vlddqu\t{$src, $dst|$dst, $src}",
4511 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4512 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4513 "vlddqu\t{$src, $dst|$dst, $src}",
4514 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4516 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4517 "lddqu\t{$src, $dst|$dst, $src}",
4518 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4520 //===---------------------------------------------------------------------===//
4521 // SSE3 - Arithmetic
4522 //===---------------------------------------------------------------------===//
4524 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4525 X86MemOperand x86memop, bit Is2Addr = 1> {
4526 def rr : I<0xD0, MRMSrcReg,
4527 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4529 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4530 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4531 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4532 def rm : I<0xD0, MRMSrcMem,
4533 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4535 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4536 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4537 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4540 let Predicates = [HasAVX],
4541 ExeDomain = SSEPackedDouble in {
4542 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4543 f128mem, 0>, TB, XD, VEX_4V;
4544 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4545 f128mem, 0>, TB, OpSize, VEX_4V;
4546 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4547 f256mem, 0>, TB, XD, VEX_4V;
4548 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4549 f256mem, 0>, TB, OpSize, VEX_4V;
4551 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
4552 ExeDomain = SSEPackedDouble in {
4553 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4555 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4556 f128mem>, TB, OpSize;
4559 //===---------------------------------------------------------------------===//
4560 // SSE3 Instructions
4561 //===---------------------------------------------------------------------===//
4564 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4565 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
4566 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4568 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4569 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4570 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
4572 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4574 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4575 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4576 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
4578 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4579 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
4580 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4582 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4583 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4584 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
4586 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4588 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4589 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4590 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
4593 let Predicates = [HasAVX] in {
4594 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4595 int_x86_sse3_hadd_ps, 0>, VEX_4V;
4596 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4597 int_x86_sse3_hadd_pd, 0>, VEX_4V;
4598 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4599 int_x86_sse3_hsub_ps, 0>, VEX_4V;
4600 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4601 int_x86_sse3_hsub_pd, 0>, VEX_4V;
4602 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4603 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
4604 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4605 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
4606 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4607 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
4608 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4609 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
4612 let Constraints = "$src1 = $dst" in {
4613 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
4614 int_x86_sse3_hadd_ps>;
4615 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
4616 int_x86_sse3_hadd_pd>;
4617 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
4618 int_x86_sse3_hsub_ps>;
4619 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
4620 int_x86_sse3_hsub_pd>;
4623 //===---------------------------------------------------------------------===//
4624 // SSSE3 - Packed Absolute Instructions
4625 //===---------------------------------------------------------------------===//
4628 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4629 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4630 PatFrag mem_frag128, Intrinsic IntId128> {
4631 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4633 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4634 [(set VR128:$dst, (IntId128 VR128:$src))]>,
4637 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4639 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4642 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
4645 let Predicates = [HasAVX] in {
4646 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
4647 int_x86_ssse3_pabs_b_128>, VEX;
4648 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
4649 int_x86_ssse3_pabs_w_128>, VEX;
4650 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
4651 int_x86_ssse3_pabs_d_128>, VEX;
4654 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
4655 int_x86_ssse3_pabs_b_128>;
4656 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
4657 int_x86_ssse3_pabs_w_128>;
4658 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
4659 int_x86_ssse3_pabs_d_128>;
4661 //===---------------------------------------------------------------------===//
4662 // SSSE3 - Packed Binary Operator Instructions
4663 //===---------------------------------------------------------------------===//
4665 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4666 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4667 PatFrag mem_frag128, Intrinsic IntId128,
4669 let isCommutable = 1 in
4670 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4671 (ins VR128:$src1, VR128:$src2),
4673 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4674 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4675 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4677 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4678 (ins VR128:$src1, i128mem:$src2),
4680 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4681 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4683 (IntId128 VR128:$src1,
4684 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4687 let Predicates = [HasAVX] in {
4688 let isCommutable = 0 in {
4689 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
4690 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
4691 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
4692 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
4693 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
4694 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
4695 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
4696 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
4697 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
4698 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
4699 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
4700 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
4701 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
4702 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
4703 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
4704 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
4705 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
4706 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
4707 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
4708 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
4709 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
4710 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
4712 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
4713 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
4716 // None of these have i8 immediate fields.
4717 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
4718 let isCommutable = 0 in {
4719 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
4720 int_x86_ssse3_phadd_w_128>;
4721 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
4722 int_x86_ssse3_phadd_d_128>;
4723 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
4724 int_x86_ssse3_phadd_sw_128>;
4725 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
4726 int_x86_ssse3_phsub_w_128>;
4727 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
4728 int_x86_ssse3_phsub_d_128>;
4729 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
4730 int_x86_ssse3_phsub_sw_128>;
4731 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
4732 int_x86_ssse3_pmadd_ub_sw_128>;
4733 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
4734 int_x86_ssse3_pshuf_b_128>;
4735 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
4736 int_x86_ssse3_psign_b_128>;
4737 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
4738 int_x86_ssse3_psign_w_128>;
4739 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
4740 int_x86_ssse3_psign_d_128>;
4742 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
4743 int_x86_ssse3_pmul_hr_sw_128>;
4746 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
4747 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
4748 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
4749 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
4751 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
4752 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4753 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
4754 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4755 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
4756 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4758 //===---------------------------------------------------------------------===//
4759 // SSSE3 - Packed Align Instruction Patterns
4760 //===---------------------------------------------------------------------===//
4762 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
4763 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
4764 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4766 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4768 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4770 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
4771 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4773 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4775 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4779 let Predicates = [HasAVX] in
4780 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
4781 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
4782 defm PALIGN : ssse3_palign<"palignr">;
4784 let Predicates = [HasSSSE3] in {
4785 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4786 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4787 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4788 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4789 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4790 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4791 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4792 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4795 let Predicates = [HasAVX] in {
4796 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4797 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4798 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4799 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4800 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4801 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4802 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4803 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4806 //===---------------------------------------------------------------------===//
4807 // SSSE3 Misc Instructions
4808 //===---------------------------------------------------------------------===//
4810 // Thread synchronization
4811 let usesCustomInserter = 1 in {
4812 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
4813 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
4814 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
4815 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
4818 let Uses = [EAX, ECX, EDX] in
4819 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
4820 Requires<[HasSSE3]>;
4821 let Uses = [ECX, EAX] in
4822 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
4823 Requires<[HasSSE3]>;
4825 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
4826 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
4828 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
4829 Requires<[In32BitMode]>;
4830 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
4831 Requires<[In64BitMode]>;
4833 // extload f32 -> f64. This matches load+fextend because we have a hack in
4834 // the isel (PreprocessForFPConvert) that can introduce loads after dag
4836 // Since these loads aren't folded into the fextend, we have to match it
4838 let Predicates = [HasSSE2] in
4839 def : Pat<(fextend (loadf32 addr:$src)),
4840 (CVTSS2SDrm addr:$src)>;
4842 // Splat v2f64 / v2i64
4843 let AddedComplexity = 10 in {
4844 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4845 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4848 // Set lowest element and zero upper elements.
4849 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4850 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
4852 //===----------------------------------------------------------------------===//
4853 // SSE4.1 - Packed Move with Sign/Zero Extend
4854 //===----------------------------------------------------------------------===//
4856 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4857 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4858 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4859 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4861 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4862 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4864 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
4868 let Predicates = [HasAVX] in {
4869 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
4871 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
4873 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
4875 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
4877 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
4879 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4883 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4884 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4885 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4886 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4887 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4888 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4890 // Common patterns involving scalar load.
4891 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4892 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4893 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4894 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4896 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4897 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4898 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4899 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4901 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4902 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4903 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4904 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4906 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4907 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4908 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4909 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4911 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4912 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4913 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4914 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4916 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4917 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4918 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4919 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4922 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4923 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4924 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4925 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4927 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4928 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4930 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4934 let Predicates = [HasAVX] in {
4935 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4937 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4939 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4941 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4945 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4946 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4947 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4948 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4950 // Common patterns involving scalar load
4951 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4952 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4953 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4954 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4956 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4957 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4958 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4959 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4962 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4963 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4964 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4965 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4967 // Expecting a i16 load any extended to i32 value.
4968 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4969 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4970 [(set VR128:$dst, (IntId (bitconvert
4971 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4975 let Predicates = [HasAVX] in {
4976 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4978 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4981 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4982 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4984 // Common patterns involving scalar load
4985 def : Pat<(int_x86_sse41_pmovsxbq
4986 (bitconvert (v4i32 (X86vzmovl
4987 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4988 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4990 def : Pat<(int_x86_sse41_pmovzxbq
4991 (bitconvert (v4i32 (X86vzmovl
4992 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4993 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4995 //===----------------------------------------------------------------------===//
4996 // SSE4.1 - Extract Instructions
4997 //===----------------------------------------------------------------------===//
4999 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5000 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5001 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5002 (ins VR128:$src1, i32i8imm:$src2),
5003 !strconcat(OpcodeStr,
5004 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5005 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5007 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5008 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5009 !strconcat(OpcodeStr,
5010 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5013 // There's an AssertZext in the way of writing the store pattern
5014 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5017 let Predicates = [HasAVX] in {
5018 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5019 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5020 (ins VR128:$src1, i32i8imm:$src2),
5021 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5024 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5027 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5028 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5029 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5030 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5031 !strconcat(OpcodeStr,
5032 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5035 // There's an AssertZext in the way of writing the store pattern
5036 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5039 let Predicates = [HasAVX] in
5040 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5042 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5045 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5046 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5047 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5048 (ins VR128:$src1, i32i8imm:$src2),
5049 !strconcat(OpcodeStr,
5050 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5052 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5053 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5054 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5055 !strconcat(OpcodeStr,
5056 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5057 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5058 addr:$dst)]>, OpSize;
5061 let Predicates = [HasAVX] in
5062 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5064 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5066 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5067 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5068 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5069 (ins VR128:$src1, i32i8imm:$src2),
5070 !strconcat(OpcodeStr,
5071 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5073 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5074 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5075 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5076 !strconcat(OpcodeStr,
5077 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5078 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5079 addr:$dst)]>, OpSize, REX_W;
5082 let Predicates = [HasAVX] in
5083 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5085 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5087 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5089 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5090 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5091 (ins VR128:$src1, i32i8imm:$src2),
5092 !strconcat(OpcodeStr,
5093 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5095 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5097 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5098 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5099 !strconcat(OpcodeStr,
5100 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5101 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5102 addr:$dst)]>, OpSize;
5105 let Predicates = [HasAVX] in {
5106 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5107 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5108 (ins VR128:$src1, i32i8imm:$src2),
5109 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5112 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5114 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5115 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5118 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5119 Requires<[HasSSE41]>;
5121 //===----------------------------------------------------------------------===//
5122 // SSE4.1 - Insert Instructions
5123 //===----------------------------------------------------------------------===//
5125 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5126 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5127 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5129 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5131 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5133 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5134 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5135 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5137 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5139 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5141 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5142 imm:$src3))]>, OpSize;
5145 let Predicates = [HasAVX] in
5146 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5147 let Constraints = "$src1 = $dst" in
5148 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5150 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5151 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5152 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5154 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5156 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5158 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5160 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5161 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5163 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5165 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5167 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5168 imm:$src3)))]>, OpSize;
5171 let Predicates = [HasAVX] in
5172 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5173 let Constraints = "$src1 = $dst" in
5174 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5176 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5177 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5178 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5180 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5182 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5184 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5186 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5187 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5189 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5191 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5193 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5194 imm:$src3)))]>, OpSize;
5197 let Predicates = [HasAVX] in
5198 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5199 let Constraints = "$src1 = $dst" in
5200 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5202 // insertps has a few different modes, there's the first two here below which
5203 // are optimized inserts that won't zero arbitrary elements in the destination
5204 // vector. The next one matches the intrinsic and could zero arbitrary elements
5205 // in the target vector.
5206 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5207 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5208 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5210 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5212 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5214 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5216 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5217 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5219 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5221 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5223 (X86insrtps VR128:$src1,
5224 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5225 imm:$src3))]>, OpSize;
5228 let Constraints = "$src1 = $dst" in
5229 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5230 let Predicates = [HasAVX] in
5231 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5233 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5234 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5236 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5237 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5238 Requires<[HasSSE41]>;
5240 //===----------------------------------------------------------------------===//
5241 // SSE4.1 - Round Instructions
5242 //===----------------------------------------------------------------------===//
5244 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5245 X86MemOperand x86memop, RegisterClass RC,
5246 PatFrag mem_frag32, PatFrag mem_frag64,
5247 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5248 // Intrinsic operation, reg.
5249 // Vector intrinsic operation, reg
5250 def PSr : SS4AIi8<opcps, MRMSrcReg,
5251 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5252 !strconcat(OpcodeStr,
5253 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5254 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5257 // Vector intrinsic operation, mem
5258 def PSm : Ii8<opcps, MRMSrcMem,
5259 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
5260 !strconcat(OpcodeStr,
5261 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5263 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5265 Requires<[HasSSE41]>;
5267 // Vector intrinsic operation, reg
5268 def PDr : SS4AIi8<opcpd, MRMSrcReg,
5269 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5270 !strconcat(OpcodeStr,
5271 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5272 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
5275 // Vector intrinsic operation, mem
5276 def PDm : SS4AIi8<opcpd, MRMSrcMem,
5277 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
5278 !strconcat(OpcodeStr,
5279 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5281 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
5285 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
5286 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
5287 // Intrinsic operation, reg.
5288 // Vector intrinsic operation, reg
5289 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
5290 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5291 !strconcat(OpcodeStr,
5292 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5295 // Vector intrinsic operation, mem
5296 def PSm_AVX : Ii8<opcps, MRMSrcMem,
5297 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5298 !strconcat(OpcodeStr,
5299 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5300 []>, TA, OpSize, Requires<[HasSSE41]>;
5302 // Vector intrinsic operation, reg
5303 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
5304 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5305 !strconcat(OpcodeStr,
5306 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5309 // Vector intrinsic operation, mem
5310 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
5311 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5312 !strconcat(OpcodeStr,
5313 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5317 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
5320 Intrinsic F64Int, bit Is2Addr = 1> {
5321 // Intrinsic operation, reg.
5322 def SSr : SS4AIi8<opcss, MRMSrcReg,
5323 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5325 !strconcat(OpcodeStr,
5326 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5327 !strconcat(OpcodeStr,
5328 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5329 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5332 // Intrinsic operation, mem.
5333 def SSm : SS4AIi8<opcss, MRMSrcMem,
5334 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5336 !strconcat(OpcodeStr,
5337 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5338 !strconcat(OpcodeStr,
5339 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5341 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
5344 // Intrinsic operation, reg.
5345 def SDr : SS4AIi8<opcsd, MRMSrcReg,
5346 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5348 !strconcat(OpcodeStr,
5349 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5350 !strconcat(OpcodeStr,
5351 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5352 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5355 // Intrinsic operation, mem.
5356 def SDm : SS4AIi8<opcsd, MRMSrcMem,
5357 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5359 !strconcat(OpcodeStr,
5360 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5361 !strconcat(OpcodeStr,
5362 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5364 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
5368 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
5370 // Intrinsic operation, reg.
5371 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
5372 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5373 !strconcat(OpcodeStr,
5374 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5377 // Intrinsic operation, mem.
5378 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
5379 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5380 !strconcat(OpcodeStr,
5381 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5384 // Intrinsic operation, reg.
5385 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
5386 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5387 !strconcat(OpcodeStr,
5388 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5391 // Intrinsic operation, mem.
5392 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
5393 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5394 !strconcat(OpcodeStr,
5395 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5399 // FP round - roundss, roundps, roundsd, roundpd
5400 let Predicates = [HasAVX] in {
5402 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
5403 memopv4f32, memopv2f64,
5404 int_x86_sse41_round_ps,
5405 int_x86_sse41_round_pd>, VEX;
5406 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
5407 memopv8f32, memopv4f64,
5408 int_x86_avx_round_ps_256,
5409 int_x86_avx_round_pd_256>, VEX;
5410 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
5411 int_x86_sse41_round_ss,
5412 int_x86_sse41_round_sd, 0>, VEX_4V;
5414 // Instructions for the assembler
5415 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
5417 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
5419 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
5422 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
5423 memopv4f32, memopv2f64,
5424 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
5425 let Constraints = "$src1 = $dst" in
5426 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
5427 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
5429 //===----------------------------------------------------------------------===//
5430 // SSE4.1 - Packed Bit Test
5431 //===----------------------------------------------------------------------===//
5433 // ptest instruction we'll lower to this in X86ISelLowering primarily from
5434 // the intel intrinsic that corresponds to this.
5435 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5436 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5437 "vptest\t{$src2, $src1|$src1, $src2}",
5438 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5440 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5441 "vptest\t{$src2, $src1|$src1, $src2}",
5442 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5445 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
5446 "vptest\t{$src2, $src1|$src1, $src2}",
5447 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
5449 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
5450 "vptest\t{$src2, $src1|$src1, $src2}",
5451 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
5455 let Defs = [EFLAGS] in {
5456 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5457 "ptest \t{$src2, $src1|$src1, $src2}",
5458 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5460 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5461 "ptest \t{$src2, $src1|$src1, $src2}",
5462 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5466 // The bit test instructions below are AVX only
5467 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
5468 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
5469 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
5470 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5471 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
5472 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
5473 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5474 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
5478 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5479 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
5480 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
5481 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
5482 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
5485 //===----------------------------------------------------------------------===//
5486 // SSE4.1 - Misc Instructions
5487 //===----------------------------------------------------------------------===//
5489 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
5490 "popcnt{w}\t{$src, $dst|$dst, $src}",
5491 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
5492 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
5493 "popcnt{w}\t{$src, $dst|$dst, $src}",
5494 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
5496 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
5497 "popcnt{l}\t{$src, $dst|$dst, $src}",
5498 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
5499 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
5500 "popcnt{l}\t{$src, $dst|$dst, $src}",
5501 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
5503 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
5504 "popcnt{q}\t{$src, $dst|$dst, $src}",
5505 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
5506 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
5507 "popcnt{q}\t{$src, $dst|$dst, $src}",
5508 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
5512 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
5513 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
5514 Intrinsic IntId128> {
5515 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5517 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5518 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
5519 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5521 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5524 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
5527 let Predicates = [HasAVX] in
5528 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
5529 int_x86_sse41_phminposuw>, VEX;
5530 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
5531 int_x86_sse41_phminposuw>;
5533 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
5534 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
5535 Intrinsic IntId128, bit Is2Addr = 1> {
5536 let isCommutable = 1 in
5537 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5538 (ins VR128:$src1, VR128:$src2),
5540 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5541 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5542 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
5543 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5544 (ins VR128:$src1, i128mem:$src2),
5546 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5547 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5549 (IntId128 VR128:$src1,
5550 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5553 let Predicates = [HasAVX] in {
5554 let isCommutable = 0 in
5555 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
5557 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
5559 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
5561 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
5563 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
5565 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
5567 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
5569 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
5571 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
5573 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
5575 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
5578 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5579 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
5580 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5581 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
5584 let Constraints = "$src1 = $dst" in {
5585 let isCommutable = 0 in
5586 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
5587 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
5588 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
5589 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
5590 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
5591 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
5592 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
5593 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
5594 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
5595 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
5596 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
5599 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5600 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
5601 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5602 (PCMPEQQrm VR128:$src1, addr:$src2)>;
5604 /// SS48I_binop_rm - Simple SSE41 binary operator.
5605 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5606 ValueType OpVT, bit Is2Addr = 1> {
5607 let isCommutable = 1 in
5608 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5609 (ins VR128:$src1, VR128:$src2),
5611 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5612 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5613 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
5615 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5616 (ins VR128:$src1, i128mem:$src2),
5618 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5619 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5620 [(set VR128:$dst, (OpNode VR128:$src1,
5621 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
5625 let Predicates = [HasAVX] in
5626 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
5627 let Constraints = "$src1 = $dst" in
5628 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
5630 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
5631 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
5632 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
5633 X86MemOperand x86memop, bit Is2Addr = 1> {
5634 let isCommutable = 1 in
5635 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
5636 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
5638 !strconcat(OpcodeStr,
5639 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5640 !strconcat(OpcodeStr,
5641 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5642 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
5644 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
5645 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
5647 !strconcat(OpcodeStr,
5648 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5649 !strconcat(OpcodeStr,
5650 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5653 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
5657 let Predicates = [HasAVX] in {
5658 let isCommutable = 0 in {
5659 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
5660 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5661 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
5662 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5663 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
5664 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5665 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
5666 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5667 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
5668 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5669 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
5670 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5672 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
5673 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5674 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
5675 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5676 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
5677 VR256, memopv32i8, i256mem, 0>, VEX_4V;
5680 let Constraints = "$src1 = $dst" in {
5681 let isCommutable = 0 in {
5682 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
5683 VR128, memopv16i8, i128mem>;
5684 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
5685 VR128, memopv16i8, i128mem>;
5686 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
5687 VR128, memopv16i8, i128mem>;
5688 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
5689 VR128, memopv16i8, i128mem>;
5691 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
5692 VR128, memopv16i8, i128mem>;
5693 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
5694 VR128, memopv16i8, i128mem>;
5697 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
5698 let Predicates = [HasAVX] in {
5699 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
5700 RegisterClass RC, X86MemOperand x86memop,
5701 PatFrag mem_frag, Intrinsic IntId> {
5702 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
5703 (ins RC:$src1, RC:$src2, RC:$src3),
5704 !strconcat(OpcodeStr,
5705 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5706 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
5707 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5709 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
5710 (ins RC:$src1, x86memop:$src2, RC:$src3),
5711 !strconcat(OpcodeStr,
5712 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5714 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
5716 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5720 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
5721 memopv16i8, int_x86_sse41_blendvpd>;
5722 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
5723 memopv16i8, int_x86_sse41_blendvps>;
5724 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
5725 memopv16i8, int_x86_sse41_pblendvb>;
5726 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
5727 memopv32i8, int_x86_avx_blendv_pd_256>;
5728 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
5729 memopv32i8, int_x86_avx_blendv_ps_256>;
5731 /// SS41I_ternary_int - SSE 4.1 ternary operator
5732 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
5733 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5734 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5735 (ins VR128:$src1, VR128:$src2),
5736 !strconcat(OpcodeStr,
5737 "\t{$src2, $dst|$dst, $src2}"),
5738 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
5741 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5742 (ins VR128:$src1, i128mem:$src2),
5743 !strconcat(OpcodeStr,
5744 "\t{$src2, $dst|$dst, $src2}"),
5747 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
5751 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
5752 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
5753 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
5755 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
5756 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
5758 let Predicates = [HasAVX] in
5759 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5760 "vmovntdqa\t{$src, $dst|$dst, $src}",
5761 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5763 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5764 "movntdqa\t{$src, $dst|$dst, $src}",
5765 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5768 //===----------------------------------------------------------------------===//
5769 // SSE4.2 - Compare Instructions
5770 //===----------------------------------------------------------------------===//
5772 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
5773 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
5774 Intrinsic IntId128, bit Is2Addr = 1> {
5775 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
5776 (ins VR128:$src1, VR128:$src2),
5778 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5779 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5780 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5782 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
5783 (ins VR128:$src1, i128mem:$src2),
5785 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5786 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5788 (IntId128 VR128:$src1,
5789 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5792 let Predicates = [HasAVX] in {
5793 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
5796 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
5797 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
5798 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
5799 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
5802 let Constraints = "$src1 = $dst" in
5803 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
5805 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
5806 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
5807 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
5808 (PCMPGTQrm VR128:$src1, addr:$src2)>;
5810 //===----------------------------------------------------------------------===//
5811 // SSE4.2 - String/text Processing Instructions
5812 //===----------------------------------------------------------------------===//
5814 // Packed Compare Implicit Length Strings, Return Mask
5815 multiclass pseudo_pcmpistrm<string asm> {
5816 def REG : PseudoI<(outs VR128:$dst),
5817 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5818 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
5820 def MEM : PseudoI<(outs VR128:$dst),
5821 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5822 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
5823 VR128:$src1, (load addr:$src2), imm:$src3))]>;
5826 let Defs = [EFLAGS], usesCustomInserter = 1 in {
5827 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
5828 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
5831 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
5832 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5833 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5834 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5835 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5836 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5837 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5840 let Defs = [XMM0, EFLAGS] in {
5841 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5842 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5843 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5844 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5845 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5846 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5849 // Packed Compare Explicit Length Strings, Return Mask
5850 multiclass pseudo_pcmpestrm<string asm> {
5851 def REG : PseudoI<(outs VR128:$dst),
5852 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5853 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5854 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
5855 def MEM : PseudoI<(outs VR128:$dst),
5856 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5857 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5858 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
5861 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
5862 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
5863 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
5866 let Predicates = [HasAVX],
5867 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5868 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5869 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5870 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5871 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5872 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5873 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5876 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5877 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5878 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5879 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5880 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5881 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5882 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5885 // Packed Compare Implicit Length Strings, Return Index
5886 let Defs = [ECX, EFLAGS] in {
5887 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
5888 def rr : SS42AI<0x63, MRMSrcReg, (outs),
5889 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5890 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5891 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
5892 (implicit EFLAGS)]>, OpSize;
5893 def rm : SS42AI<0x63, MRMSrcMem, (outs),
5894 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5895 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5896 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
5897 (implicit EFLAGS)]>, OpSize;
5901 let Predicates = [HasAVX] in {
5902 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
5904 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
5906 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
5908 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
5910 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
5912 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5916 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5917 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5918 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5919 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5920 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5921 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5923 // Packed Compare Explicit Length Strings, Return Index
5924 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5925 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5926 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5927 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5928 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5929 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5930 (implicit EFLAGS)]>, OpSize;
5931 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5932 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5933 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5935 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5936 (implicit EFLAGS)]>, OpSize;
5940 let Predicates = [HasAVX] in {
5941 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5943 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5945 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5947 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5949 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5951 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5955 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5956 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5957 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5958 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5959 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5960 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5962 //===----------------------------------------------------------------------===//
5963 // SSE4.2 - CRC Instructions
5964 //===----------------------------------------------------------------------===//
5966 // No CRC instructions have AVX equivalents
5968 // crc intrinsic instruction
5969 // This set of instructions are only rm, the only difference is the size
5971 let Constraints = "$src1 = $dst" in {
5972 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5973 (ins GR32:$src1, i8mem:$src2),
5974 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5976 (int_x86_sse42_crc32_32_8 GR32:$src1,
5977 (load addr:$src2)))]>;
5978 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5979 (ins GR32:$src1, GR8:$src2),
5980 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5982 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5983 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5984 (ins GR32:$src1, i16mem:$src2),
5985 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5987 (int_x86_sse42_crc32_32_16 GR32:$src1,
5988 (load addr:$src2)))]>,
5990 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5991 (ins GR32:$src1, GR16:$src2),
5992 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5994 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5996 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5997 (ins GR32:$src1, i32mem:$src2),
5998 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6000 (int_x86_sse42_crc32_32_32 GR32:$src1,
6001 (load addr:$src2)))]>;
6002 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6003 (ins GR32:$src1, GR32:$src2),
6004 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6006 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6007 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6008 (ins GR64:$src1, i8mem:$src2),
6009 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6011 (int_x86_sse42_crc32_64_8 GR64:$src1,
6012 (load addr:$src2)))]>,
6014 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6015 (ins GR64:$src1, GR8:$src2),
6016 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6018 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6020 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6021 (ins GR64:$src1, i64mem:$src2),
6022 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6024 (int_x86_sse42_crc32_64_64 GR64:$src1,
6025 (load addr:$src2)))]>,
6027 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6028 (ins GR64:$src1, GR64:$src2),
6029 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6031 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6035 //===----------------------------------------------------------------------===//
6036 // AES-NI Instructions
6037 //===----------------------------------------------------------------------===//
6039 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6040 Intrinsic IntId128, bit Is2Addr = 1> {
6041 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6042 (ins VR128:$src1, VR128:$src2),
6044 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6045 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6046 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6048 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6049 (ins VR128:$src1, i128mem:$src2),
6051 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6052 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6054 (IntId128 VR128:$src1,
6055 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
6058 // Perform One Round of an AES Encryption/Decryption Flow
6059 let Predicates = [HasAVX, HasAES] in {
6060 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6061 int_x86_aesni_aesenc, 0>, VEX_4V;
6062 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6063 int_x86_aesni_aesenclast, 0>, VEX_4V;
6064 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6065 int_x86_aesni_aesdec, 0>, VEX_4V;
6066 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6067 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6070 let Constraints = "$src1 = $dst" in {
6071 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6072 int_x86_aesni_aesenc>;
6073 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
6074 int_x86_aesni_aesenclast>;
6075 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
6076 int_x86_aesni_aesdec>;
6077 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
6078 int_x86_aesni_aesdeclast>;
6081 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
6082 (AESENCrr VR128:$src1, VR128:$src2)>;
6083 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
6084 (AESENCrm VR128:$src1, addr:$src2)>;
6085 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
6086 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
6087 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
6088 (AESENCLASTrm VR128:$src1, addr:$src2)>;
6089 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
6090 (AESDECrr VR128:$src1, VR128:$src2)>;
6091 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
6092 (AESDECrm VR128:$src1, addr:$src2)>;
6093 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
6094 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
6095 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
6096 (AESDECLASTrm VR128:$src1, addr:$src2)>;
6098 // Perform the AES InvMixColumn Transformation
6099 let Predicates = [HasAVX, HasAES] in {
6100 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6102 "vaesimc\t{$src1, $dst|$dst, $src1}",
6104 (int_x86_aesni_aesimc VR128:$src1))]>,
6106 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6107 (ins i128mem:$src1),
6108 "vaesimc\t{$src1, $dst|$dst, $src1}",
6110 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
6113 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6115 "aesimc\t{$src1, $dst|$dst, $src1}",
6117 (int_x86_aesni_aesimc VR128:$src1))]>,
6119 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6120 (ins i128mem:$src1),
6121 "aesimc\t{$src1, $dst|$dst, $src1}",
6123 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
6126 // AES Round Key Generation Assist
6127 let Predicates = [HasAVX, HasAES] in {
6128 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6129 (ins VR128:$src1, i8imm:$src2),
6130 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6132 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6134 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6135 (ins i128mem:$src1, i8imm:$src2),
6136 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6138 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
6142 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6143 (ins VR128:$src1, i8imm:$src2),
6144 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6146 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6148 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6149 (ins i128mem:$src1, i8imm:$src2),
6150 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6152 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
6156 //===----------------------------------------------------------------------===//
6157 // CLMUL Instructions
6158 //===----------------------------------------------------------------------===//
6160 // Carry-less Multiplication instructions
6161 let Constraints = "$src1 = $dst" in {
6162 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6163 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6164 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6167 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6168 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6169 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6173 // AVX carry-less Multiplication instructions
6174 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6175 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6176 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6179 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6180 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6181 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6185 multiclass pclmul_alias<string asm, int immop> {
6186 def : InstAlias<!strconcat("pclmul", asm,
6187 "dq {$src, $dst|$dst, $src}"),
6188 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
6190 def : InstAlias<!strconcat("pclmul", asm,
6191 "dq {$src, $dst|$dst, $src}"),
6192 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
6194 def : InstAlias<!strconcat("vpclmul", asm,
6195 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6196 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
6198 def : InstAlias<!strconcat("vpclmul", asm,
6199 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6200 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
6202 defm : pclmul_alias<"hqhq", 0x11>;
6203 defm : pclmul_alias<"hqlq", 0x01>;
6204 defm : pclmul_alias<"lqhq", 0x10>;
6205 defm : pclmul_alias<"lqlq", 0x00>;
6207 //===----------------------------------------------------------------------===//
6209 //===----------------------------------------------------------------------===//
6211 //===----------------------------------------------------------------------===//
6212 // VBROADCAST - Load from memory and broadcast to all elements of the
6213 // destination operand
6215 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
6216 X86MemOperand x86memop, Intrinsic Int> :
6217 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
6218 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6219 [(set RC:$dst, (Int addr:$src))]>, VEX;
6221 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
6222 int_x86_avx_vbroadcastss>;
6223 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
6224 int_x86_avx_vbroadcastss_256>;
6225 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
6226 int_x86_avx_vbroadcast_sd_256>;
6227 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
6228 int_x86_avx_vbroadcastf128_pd_256>;
6230 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
6231 (VBROADCASTF128 addr:$src)>;
6233 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
6234 (VBROADCASTSSY addr:$src)>;
6235 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
6236 (VBROADCASTSD addr:$src)>;
6237 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
6238 (VBROADCASTSSY addr:$src)>;
6239 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
6240 (VBROADCASTSD addr:$src)>;
6242 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
6243 (VBROADCASTSS addr:$src)>;
6244 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
6245 (VBROADCASTSS addr:$src)>;
6247 //===----------------------------------------------------------------------===//
6248 // VINSERTF128 - Insert packed floating-point values
6250 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
6251 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
6252 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6254 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
6255 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
6256 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6259 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
6260 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6261 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
6262 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6263 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
6264 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6266 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
6268 (VINSERTF128rr VR256:$src1, VR128:$src2,
6269 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6270 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
6272 (VINSERTF128rr VR256:$src1, VR128:$src2,
6273 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6274 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
6276 (VINSERTF128rr VR256:$src1, VR128:$src2,
6277 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6278 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
6280 (VINSERTF128rr VR256:$src1, VR128:$src2,
6281 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6282 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
6284 (VINSERTF128rr VR256:$src1, VR128:$src2,
6285 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6286 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
6288 (VINSERTF128rr VR256:$src1, VR128:$src2,
6289 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6291 //===----------------------------------------------------------------------===//
6292 // VEXTRACTF128 - Extract packed floating-point values
6294 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
6295 (ins VR256:$src1, i8imm:$src2),
6296 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6298 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
6299 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
6300 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6303 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
6304 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6305 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
6306 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6307 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
6308 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6310 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6311 (v4f32 (VEXTRACTF128rr
6312 (v8f32 VR256:$src1),
6313 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6314 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6315 (v2f64 (VEXTRACTF128rr
6316 (v4f64 VR256:$src1),
6317 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6318 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6319 (v4i32 (VEXTRACTF128rr
6320 (v8i32 VR256:$src1),
6321 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6322 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6323 (v2i64 (VEXTRACTF128rr
6324 (v4i64 VR256:$src1),
6325 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6326 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6327 (v8i16 (VEXTRACTF128rr
6328 (v16i16 VR256:$src1),
6329 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6330 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6331 (v16i8 (VEXTRACTF128rr
6332 (v32i8 VR256:$src1),
6333 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6335 //===----------------------------------------------------------------------===//
6336 // VMASKMOV - Conditional SIMD Packed Loads and Stores
6338 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
6339 Intrinsic IntLd, Intrinsic IntLd256,
6340 Intrinsic IntSt, Intrinsic IntSt256,
6341 PatFrag pf128, PatFrag pf256> {
6342 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
6343 (ins VR128:$src1, f128mem:$src2),
6344 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6345 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
6347 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
6348 (ins VR256:$src1, f256mem:$src2),
6349 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6350 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
6352 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
6353 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
6354 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6355 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
6356 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
6357 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
6358 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6359 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
6362 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
6363 int_x86_avx_maskload_ps,
6364 int_x86_avx_maskload_ps_256,
6365 int_x86_avx_maskstore_ps,
6366 int_x86_avx_maskstore_ps_256,
6367 memopv4f32, memopv8f32>;
6368 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
6369 int_x86_avx_maskload_pd,
6370 int_x86_avx_maskload_pd_256,
6371 int_x86_avx_maskstore_pd,
6372 int_x86_avx_maskstore_pd_256,
6373 memopv2f64, memopv4f64>;
6375 //===----------------------------------------------------------------------===//
6376 // VPERMIL - Permute Single and Double Floating-Point Values
6378 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
6379 RegisterClass RC, X86MemOperand x86memop_f,
6380 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
6381 Intrinsic IntVar, Intrinsic IntImm> {
6382 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
6383 (ins RC:$src1, RC:$src2),
6384 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6385 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
6386 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
6387 (ins RC:$src1, x86memop_i:$src2),
6388 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6389 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
6391 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
6392 (ins RC:$src1, i8imm:$src2),
6393 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6394 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
6395 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
6396 (ins x86memop_f:$src1, i8imm:$src2),
6397 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6398 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
6401 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
6402 memopv4f32, memopv4i32,
6403 int_x86_avx_vpermilvar_ps,
6404 int_x86_avx_vpermil_ps>;
6405 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
6406 memopv8f32, memopv8i32,
6407 int_x86_avx_vpermilvar_ps_256,
6408 int_x86_avx_vpermil_ps_256>;
6409 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
6410 memopv2f64, memopv2i64,
6411 int_x86_avx_vpermilvar_pd,
6412 int_x86_avx_vpermil_pd>;
6413 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
6414 memopv4f64, memopv4i64,
6415 int_x86_avx_vpermilvar_pd_256,
6416 int_x86_avx_vpermil_pd_256>;
6418 def : Pat<(v8f32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6419 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6420 def : Pat<(v4f64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6421 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6422 def : Pat<(v8i32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6423 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6424 def : Pat<(v4i64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6425 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6427 //===----------------------------------------------------------------------===//
6428 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
6430 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
6431 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
6432 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6434 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
6435 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
6436 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6439 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
6440 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6441 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
6442 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6443 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
6444 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6446 def : Pat<(int_x86_avx_vperm2f128_ps_256
6447 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
6448 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6449 def : Pat<(int_x86_avx_vperm2f128_pd_256
6450 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
6451 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6452 def : Pat<(int_x86_avx_vperm2f128_si_256
6453 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
6454 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6456 def : Pat<(v8f32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6457 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6458 def : Pat<(v8i32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6459 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6460 def : Pat<(v4i64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6461 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6462 def : Pat<(v4f64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6463 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6464 def : Pat<(v32i8 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6465 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6466 def : Pat<(v16i16 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6467 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6469 //===----------------------------------------------------------------------===//
6470 // VZERO - Zero YMM registers
6472 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
6473 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
6474 // Zero All YMM registers
6475 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
6476 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
6478 // Zero Upper bits of YMM registers
6479 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
6480 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;