//===----------------------------------------------------------------------===//
-//===----------------------------------------------------------------------===//
-// SSE scalar FP Instructions
-//===----------------------------------------------------------------------===//
-
-// CMOV* - Used to implement the SSE SELECT DAG operation. Expanded after
-// instruction selection into a branch sequence.
-let Uses = [EFLAGS], usesCustomInserter = 1 in {
- def CMOV_FR32 : I<0, Pseudo,
- (outs FR32:$dst), (ins FR32:$t, FR32:$f, i8imm:$cond),
- "#CMOV_FR32 PSEUDO!",
- [(set FR32:$dst, (X86cmov FR32:$t, FR32:$f, imm:$cond,
- EFLAGS))]>;
- def CMOV_FR64 : I<0, Pseudo,
- (outs FR64:$dst), (ins FR64:$t, FR64:$f, i8imm:$cond),
- "#CMOV_FR64 PSEUDO!",
- [(set FR64:$dst, (X86cmov FR64:$t, FR64:$f, imm:$cond,
- EFLAGS))]>;
- def CMOV_V4F32 : I<0, Pseudo,
- (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
- "#CMOV_V4F32 PSEUDO!",
- [(set VR128:$dst,
- (v4f32 (X86cmov VR128:$t, VR128:$f, imm:$cond,
- EFLAGS)))]>;
- def CMOV_V2F64 : I<0, Pseudo,
- (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
- "#CMOV_V2F64 PSEUDO!",
- [(set VR128:$dst,
- (v2f64 (X86cmov VR128:$t, VR128:$f, imm:$cond,
- EFLAGS)))]>;
- def CMOV_V2I64 : I<0, Pseudo,
- (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
- "#CMOV_V2I64 PSEUDO!",
- [(set VR128:$dst,
- (v2i64 (X86cmov VR128:$t, VR128:$f, imm:$cond,
- EFLAGS)))]>;
-}
-
//===----------------------------------------------------------------------===//
// SSE 1 & 2 Instructions Classes
//===----------------------------------------------------------------------===//
!if(Is2Addr,
!strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_sse",
- !strconcat(SSEVer, !strconcat("_",
- !strconcat(OpcodeStr, FPSizeStr))))
+ [(set RC:$dst, (!cast<Intrinsic>(
+ !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
RC:$src1, RC:$src2))]>;
def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
!if(Is2Addr,
!strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_sse",
- !strconcat(SSEVer, !strconcat("_",
- !strconcat(OpcodeStr, FPSizeStr))))
+ [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
+ SSEVer, "_", OpcodeStr, FPSizeStr))
RC:$src1, mem_cpat:$src2))]>;
}
!if(Is2Addr,
!strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_",
- !strconcat(SSEVer, !strconcat("_",
- !strconcat(OpcodeStr, FPSizeStr))))
+ [(set RC:$dst, (!cast<Intrinsic>(
+ !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
RC:$src1, RC:$src2))], d>;
def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
!if(Is2Addr,
!strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_",
- !strconcat(SSEVer, !strconcat("_",
- !strconcat(OpcodeStr, FPSizeStr))))
+ [(set RC:$dst, (!cast<Intrinsic>(
+ !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
RC:$src1, (mem_frag addr:$src2)))], d>;
}
// is used instead. Register-to-register movss/movsd is not modeled as an
// INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
// in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
-let isAsmParserOnly = 1 in {
- def VMOVSSrr : sse12_move_rr<FR32, v4f32,
- "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
- def VMOVSDrr : sse12_move_rr<FR64, v2f64,
- "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
+def VMOVSSrr : sse12_move_rr<FR32, v4f32,
+ "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
+def VMOVSDrr : sse12_move_rr<FR64, v2f64,
+ "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
- let canFoldAsLoad = 1, isReMaterializable = 1 in {
- def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
+let canFoldAsLoad = 1, isReMaterializable = 1 in {
+ def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
- let AddedComplexity = 20 in
- def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
- }
+ let AddedComplexity = 20 in
+ def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
}
let Constraints = "$src1 = $dst" in {
// Implicitly promote a 64-bit scalar to a vector.
def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
(INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
+// Implicitly promote a 32-bit scalar to a vector.
+def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
+ (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
+// Implicitly promote a 64-bit scalar to a vector.
+def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
+ (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
let AddedComplexity = 20 in {
+let Predicates = [HasSSE1] in {
+ // MOVSSrm zeros the high parts of the register; represent this
+ // with SUBREG_TO_REG.
+ def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
+ (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
+ def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
+ (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
+ def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
+ (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
+}
+let Predicates = [HasSSE2] in {
+ // MOVSDrm zeros the high parts of the register; represent this
+ // with SUBREG_TO_REG.
+ def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+ def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+ def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+ def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+ def : Pat<(v2f64 (X86vzload addr:$src)),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+}
+}
+
+let AddedComplexity = 20, Predicates = [HasAVX] in {
// MOVSSrm zeros the high parts of the register; represent this
-// with SUBREG_TO_REG.
+// with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
- (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
+ (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
- (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
+ (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
- (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
+ (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
// MOVSDrm zeros the high parts of the register; represent this
-// with SUBREG_TO_REG.
+// with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
- (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+ (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
- (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+ (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
- (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+ (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
- (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+ (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
def : Pat<(v2f64 (X86vzload addr:$src)),
- (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+ (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
+// Represent the same patterns above but in the form they appear for
+// 256-bit types
+def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
+ (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
+ (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
+def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
+ (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
+ (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
}
// Store scalar value to memory.
"movsd\t{$src, $dst|$dst, $src}",
[(store FR64:$src, addr:$dst)]>;
-let isAsmParserOnly = 1 in {
def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
"movss\t{$src, $dst|$dst, $src}",
[(store FR32:$src, addr:$dst)]>, XS, VEX;
def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
"movsd\t{$src, $dst|$dst, $src}",
[(store FR64:$src, addr:$dst)]>, XD, VEX;
-}
// Extract and store.
def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
[(set RC:$dst, (ld_frag addr:$src))], d>;
}
-let isAsmParserOnly = 1 in {
defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
"movaps", SSEPackedSingle>, VEX;
defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
"movups", SSEPackedSingle>, VEX;
defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
"movupd", SSEPackedDouble, 0>, OpSize, VEX;
-}
defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
"movaps", SSEPackedSingle>, TB;
defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
"movupd", SSEPackedDouble, 0>, TB, OpSize;
-let isAsmParserOnly = 1 in {
def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movaps\t{$src, $dst|$dst, $src}",
[(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
"movupd\t{$src, $dst|$dst, $src}",
[(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
-}
+
+def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
+def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
+ (VMOVUPSYmr addr:$dst, VR256:$src)>;
+
+def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
+def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
+ (VMOVUPDYmr addr:$dst, VR256:$src)>;
+
def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movaps\t{$src, $dst|$dst, $src}",
[(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
[(store (v2f64 VR128:$src), addr:$dst)]>;
// Intrinsic forms of MOVUPS/D load and store
-let isAsmParserOnly = 1 in {
- let canFoldAsLoad = 1, isReMaterializable = 1 in
- def VMOVUPSrm_Int : VPSI<0x10, MRMSrcMem, (outs VR128:$dst),
- (ins f128mem:$src),
- "movups\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>, VEX;
- def VMOVUPDrm_Int : VPDI<0x10, MRMSrcMem, (outs VR128:$dst),
- (ins f128mem:$src),
- "movupd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>, VEX;
- def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
- (ins f128mem:$dst, VR128:$src),
- "movups\t{$src, $dst|$dst, $src}",
- [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
- def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
- (ins f128mem:$dst, VR128:$src),
- "movupd\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
-}
-let canFoldAsLoad = 1, isReMaterializable = 1 in
-def MOVUPSrm_Int : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- "movups\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
-def MOVUPDrm_Int : PDI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- "movupd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>;
+def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
+ (ins f128mem:$dst, VR128:$src),
+ "movups\t{$src, $dst|$dst, $src}",
+ [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
+def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
+ (ins f128mem:$dst, VR128:$src),
+ "movupd\t{$src, $dst|$dst, $src}",
+ [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movups\t{$src, $dst|$dst, $src}",
string asm_opr> {
def PSrm : PI<opc, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
- !strconcat(!strconcat(base_opc,"s"), asm_opr),
+ !strconcat(base_opc, "s", asm_opr),
[(set RC:$dst,
(mov_frag RC:$src1,
(bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
def PDrm : PI<opc, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, f64mem:$src2),
- !strconcat(!strconcat(base_opc,"d"), asm_opr),
+ !strconcat(base_opc, "d", asm_opr),
[(set RC:$dst, (v2f64 (mov_frag RC:$src1,
(scalar_to_vector (loadf64 addr:$src2)))))],
SSEPackedDouble>, TB, OpSize;
}
-let isAsmParserOnly = 1, AddedComplexity = 20 in {
+let AddedComplexity = 20 in {
defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
"\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
"\t{$src2, $dst|$dst, $src2}">;
}
-let isAsmParserOnly = 1 in {
def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movlps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
"movlpd\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract (v2f64 VR128:$src),
(iPTR 0))), addr:$dst)]>, VEX;
-}
def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movlps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
// v2f64 extract element 1 is always custom lowered to unpack high to low
// and extract element 0 so the non-store version isn't too horrible.
-let isAsmParserOnly = 1 in {
def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movhps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract
(v2f64 (unpckh VR128:$src, (undef))),
(iPTR 0))), addr:$dst)]>,
VEX;
-}
def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movhps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract
(v2f64 (unpckh VR128:$src, (undef))),
(iPTR 0))), addr:$dst)]>;
-let isAsmParserOnly = 1, AddedComplexity = 20 in {
+let AddedComplexity = 20 in {
def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
"movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
(v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
}
-def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
- (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
-let AddedComplexity = 20 in {
- def : Pat<(v4f32 (movddup VR128:$src, (undef))),
- (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
- def : Pat<(v2i64 (movddup VR128:$src, (undef))),
- (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
+let Predicates = [HasAVX] in {
+ // MOVHPS patterns
+ def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
+ (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
+ def : Pat<(X86Movlhps VR128:$src1,
+ (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
+ (VMOVHPSrm VR128:$src1, addr:$src2)>;
+ def : Pat<(X86Movlhps VR128:$src1,
+ (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
+ (VMOVHPSrm VR128:$src1, addr:$src2)>;
+
+ // MOVLHPS patterns
+ let AddedComplexity = 20 in {
+ def : Pat<(v4f32 (movddup VR128:$src, (undef))),
+ (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
+ def : Pat<(v2i64 (movddup VR128:$src, (undef))),
+ (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
+
+ // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
+ def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
+ (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
+ }
+ def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
+
+ // MOVHLPS patterns
+ let AddedComplexity = 20 in {
+ // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
+ def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
+ (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
+
+ // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
+ def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
+ (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
+ def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
+ (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
+ }
+
+ def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
+ (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
+ (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
+}
+
+let Predicates = [HasSSE1] in {
+ // MOVHPS patterns
+ def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
+ (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
+ def : Pat<(X86Movlhps VR128:$src1,
+ (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
+ (MOVHPSrm VR128:$src1, addr:$src2)>;
+ def : Pat<(X86Movlhps VR128:$src1,
+ (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
+ (MOVHPSrm VR128:$src1, addr:$src2)>;
+
+ // MOVLHPS patterns
+ let AddedComplexity = 20 in {
+ def : Pat<(v4f32 (movddup VR128:$src, (undef))),
+ (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
+ def : Pat<(v2i64 (movddup VR128:$src, (undef))),
+ (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
+
+ // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
+ def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
+ (MOVLHPSrr VR128:$src1, VR128:$src2)>;
+ }
+ def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (MOVLHPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (MOVLHPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
+ (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
+
+ // MOVHLPS patterns
+ let AddedComplexity = 20 in {
+ // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
+ def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
+ (MOVHLPSrr VR128:$src1, VR128:$src2)>;
+
+ // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
+ def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
+ (MOVHLPSrr VR128:$src1, VR128:$src1)>;
+ def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
+ (MOVHLPSrr VR128:$src1, VR128:$src1)>;
+ }
+
+ def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
+ (MOVHLPSrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
+ (MOVHLPSrr VR128:$src1, VR128:$src2)>;
}
//===----------------------------------------------------------------------===//
!strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
}
-let isAsmParserOnly = 1 in {
defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
"cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
VEX_4V;
defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
VEX_4V, VEX_W;
+
+let Predicates = [HasAVX] in {
+ def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
+ (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
+ def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
+ (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
+ def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
+ (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
+ def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
+ (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
+
+ def : Pat<(f32 (sint_to_fp GR32:$src)),
+ (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
+ def : Pat<(f32 (sint_to_fp GR64:$src)),
+ (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
+ def : Pat<(f64 (sint_to_fp GR32:$src)),
+ (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
+ def : Pat<(f64 (sint_to_fp GR64:$src)),
+ (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
}
defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
// Conversion Instructions Intrinsics - Match intrinsics which expect MM
// and/or XMM operand(s).
-multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
- Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
- string asm, Domain d> {
- def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
- [(set DstRC:$dst, (Int SrcRC:$src))], d>;
- def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
- [(set DstRC:$dst, (Int (ld_frag addr:$src)))], d>;
-}
multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
[(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
}
-multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
- RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
- PatFrag ld_frag, string asm, Domain d> {
- def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
- asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))], d>;
- def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst),
- (ins DstRC:$src1, x86memop:$src2), asm,
- [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))], d>;
-}
-
multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
PatFrag ld_frag, string asm, bit Is2Addr = 1> {
[(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
}
-let isAsmParserOnly = 1 in {
- defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
- f32mem, load, "cvtss2si">, XS, VEX;
- defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
- int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
- XS, VEX, VEX_W;
- defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
- f128mem, load, "cvtsd2si">, XD, VEX;
- defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
- int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
- XD, VEX, VEX_W;
-
- // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
- // Get rid of this hack or rename the intrinsics, there are several
- // intructions that only match with the intrinsic form, why create duplicates
- // to let them be recognized by the assembler?
- defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
- "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
- defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
- "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
-}
-defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
- f32mem, load, "cvtss2si">, XS;
-defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
- f32mem, load, "cvtss2si{q}">, XS, REX_W;
-defm Int_CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
- f128mem, load, "cvtsd2si">, XD;
-defm Int_CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
- f128mem, load, "cvtsd2si">, XD, REX_W;
-
-defm CVTSD2SI64 : sse12_cvt_s_np<0x2D, VR128, GR64, f64mem, "cvtsd2si{q}">, XD,
- REX_W;
-
-let isAsmParserOnly = 1 in {
- defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
- int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
- defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
- int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
- VEX_W;
- defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
- int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
- defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
- int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
- VEX_4V, VEX_W;
-}
+defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
+ f128mem, load, "cvtsd2si">, XD, VEX;
+defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
+ int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
+ XD, VEX, VEX_W;
+
+// FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
+// Get rid of this hack or rename the intrinsics, there are several
+// intructions that only match with the intrinsic form, why create duplicates
+// to let them be recognized by the assembler?
+defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
+ "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
+defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
+ "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
+defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
+ f128mem, load, "cvtsd2si{l}">, XD;
+defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
+ f128mem, load, "cvtsd2si{q}">, XD, REX_W;
+
+
+defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
+ int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
+defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
+ int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
+ VEX_W;
+defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
+ int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
+defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
+ int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
+ VEX_4V, VEX_W;
let Constraints = "$src1 = $dst" in {
defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
"cvtsi2sd">, XD, REX_W;
}
-// Instructions below don't have an AVX form.
-defm Int_CVTPS2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtps2pi,
- f64mem, load, "cvtps2pi\t{$src, $dst|$dst, $src}",
- SSEPackedSingle>, TB;
-defm Int_CVTPD2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtpd2pi,
- f128mem, memop, "cvtpd2pi\t{$src, $dst|$dst, $src}",
- SSEPackedDouble>, TB, OpSize;
-defm Int_CVTTPS2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttps2pi,
- f64mem, load, "cvttps2pi\t{$src, $dst|$dst, $src}",
- SSEPackedSingle>, TB;
-defm Int_CVTTPD2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttpd2pi,
- f128mem, memop, "cvttpd2pi\t{$src, $dst|$dst, $src}",
- SSEPackedDouble>, TB, OpSize;
-defm Int_CVTPI2PD : sse12_cvt_pint<0x2A, VR64, VR128, int_x86_sse_cvtpi2pd,
- i64mem, load, "cvtpi2pd\t{$src, $dst|$dst, $src}",
- SSEPackedDouble>, TB, OpSize;
-let Constraints = "$src1 = $dst" in {
- defm Int_CVTPI2PS : sse12_cvt_pint_3addr<0x2A, VR64, VR128,
- int_x86_sse_cvtpi2ps,
- i64mem, load, "cvtpi2ps\t{$src2, $dst|$dst, $src2}",
- SSEPackedSingle>, TB;
-}
-
/// SSE 1 Only
// Aliases for intrinsics
-let isAsmParserOnly = 1 in {
defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
f32mem, load, "cvttss2si">, XS, VEX;
defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse_cvttss2si64, f32mem, load,
"cvttss2si">, XS, VEX, VEX_W;
defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
- f128mem, load, "cvttss2si">, XD, VEX;
+ f128mem, load, "cvttsd2si">, XD, VEX;
defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse2_cvttsd2si64, f128mem, load,
- "cvttss2si">, XD, VEX, VEX_W;
-}
+ "cvttsd2si">, XD, VEX, VEX_W;
defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
f32mem, load, "cvttss2si">, XS;
defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse_cvttss2si64, f32mem, load,
"cvttss2si{q}">, XS, REX_W;
defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
- f128mem, load, "cvttss2si">, XD;
+ f128mem, load, "cvttsd2si">, XD;
defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse2_cvttsd2si64, f128mem, load,
- "cvttss2si{q}">, XD, REX_W;
+ "cvttsd2si{q}">, XD, REX_W;
-let isAsmParserOnly = 1, Pattern = []<dag> in {
+let Pattern = []<dag> in {
defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
"cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
"cvtdq2ps\t{$src, $dst|$dst, $src}",
SSEPackedSingle>, TB, VEX;
}
+
let Pattern = []<dag> in {
defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
"cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
}
+let Predicates = [HasSSE1] in {
+ def : Pat<(int_x86_sse_cvtss2si VR128:$src),
+ (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
+ def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
+ (CVTSS2SIrm addr:$src)>;
+ def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
+ (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
+ def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
+ (CVTSS2SI64rm addr:$src)>;
+}
+
+let Predicates = [HasAVX] in {
+ def : Pat<(int_x86_sse_cvtss2si VR128:$src),
+ (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
+ def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
+ (VCVTSS2SIrm addr:$src)>;
+ def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
+ (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
+ def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
+ (VCVTSS2SI64rm addr:$src)>;
+}
+
/// SSE 2 Only
// Convert scalar double to scalar single
-let isAsmParserOnly = 1 in {
def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
(ins FR64:$src1, FR64:$src2),
"cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
(ins FR64:$src1, f64mem:$src2),
"vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
-}
+def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
+ Requires<[HasAVX]>;
+
def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
"cvtsd2ss\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (fround FR64:$src))]>;
[(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
Requires<[HasSSE2, OptForSize]>;
-let isAsmParserOnly = 1 in
defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
XS, VEX_4V;
int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
// Convert scalar single to scalar double
-let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
+// SSE2 instructions with XS prefix
def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
(ins FR32:$src1, FR32:$src2),
"vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
(ins FR32:$src1, f32mem:$src2),
"vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
+
+let Predicates = [HasAVX] in {
+ def : Pat<(f64 (fextend FR32:$src)),
+ (VCVTSS2SDrr FR32:$src, FR32:$src)>;
+ def : Pat<(fextend (loadf32 addr:$src)),
+ (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
+ def : Pat<(extloadf32 addr:$src),
+ (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
}
+
def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
"cvtss2sd\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (fextend FR32:$src))]>, XS,
[(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
Requires<[HasSSE2, OptForSize]>;
-let isAsmParserOnly = 1 in {
def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
(load addr:$src2)))]>, XS, VEX_4V,
Requires<[HasAVX]>;
-}
let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
Requires<[HasSSE2, OptForSpeed]>;
// Convert doubleword to packed single/double fp
-let isAsmParserOnly = 1 in { // SSE2 instructions without OpSize prefix
+// SSE2 instructions without OpSize prefix
def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vcvtdq2ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
[(set VR128:$dst, (int_x86_sse2_cvtdq2ps
(bitconvert (memopv2i64 addr:$src))))]>,
TB, VEX, Requires<[HasAVX]>;
-}
def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtdq2ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
TB, Requires<[HasSSE2]>;
// FIXME: why the non-intrinsic version is described as SSE3?
-let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
+// SSE2 instructions with XS prefix
def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vcvtdq2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
[(set VR128:$dst, (int_x86_sse2_cvtdq2pd
(bitconvert (memopv2i64 addr:$src))))]>,
XS, VEX, Requires<[HasAVX]>;
-}
def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtdq2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
// Convert packed single/double fp to doubleword
-let isAsmParserOnly = 1 in {
def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
-}
def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}", []>;
def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}", []>;
-let isAsmParserOnly = 1 in {
def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
"cvtps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2dq
(memop addr:$src)))]>, VEX;
-}
def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
[(set VR128:$dst, (int_x86_sse2_cvtps2dq
(memop addr:$src)))]>;
-let isAsmParserOnly = 1 in { // SSE2 packed instructions with XD prefix
+// SSE2 packed instructions with XD prefix
def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vcvtpd2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
[(set VR128:$dst, (int_x86_sse2_cvtpd2dq
(memop addr:$src)))]>,
XD, VEX, Requires<[HasAVX]>;
-}
def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
// Convert with truncation packed single/double fp to doubleword
-let isAsmParserOnly = 1 in { // SSE2 packed instructions with XS prefix
+// SSE2 packed instructions with XS prefix
def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
-}
def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
+ "cvttps2dq\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst,
+ (int_x86_sse2_cvttps2dq VR128:$src))]>;
def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
-
+ "cvttps2dq\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst,
+ (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
-let isAsmParserOnly = 1 in {
def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vcvttps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
[(set VR128:$dst, (int_x86_sse2_cvttps2dq
(memop addr:$src)))]>,
XS, VEX, Requires<[HasAVX]>;
-}
-def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "cvttps2dq\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst,
- (int_x86_sse2_cvttps2dq VR128:$src))]>,
- XS, Requires<[HasSSE2]>;
-def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- "cvttps2dq\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_cvttps2dq
- (memop addr:$src)))]>,
- XS, Requires<[HasSSE2]>;
-let isAsmParserOnly = 1 in {
+def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
+ (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
+def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
+ (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
+
+def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
+ (Int_VCVTDQ2PSrr VR128:$src)>, Requires<[HasAVX]>;
+def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
+ (VCVTTPS2DQrr VR128:$src)>, Requires<[HasAVX]>;
+def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
+ (VCVTDQ2PSYrr VR256:$src)>, Requires<[HasAVX]>;
+def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
+ (VCVTTPS2DQYrr VR256:$src)>, Requires<[HasAVX]>;
+
def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src),
"cvttpd2dq\t{$src, $dst|$dst, $src}",
"cvttpd2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvttpd2dq
(memop addr:$src)))]>, VEX;
-}
-def Int_CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "cvttpd2dq\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
-def Int_CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
- "cvttpd2dq\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
- (memop addr:$src)))]>;
+def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "cvttpd2dq\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
+def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
+ "cvttpd2dq\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
+ (memop addr:$src)))]>;
-let isAsmParserOnly = 1 in {
// The assembler can recognize rr 256-bit instructions by seeing a ymm
// register, but the same isn't true when using memory operands instead.
// Provide other assembly rr and rm forms to address this explicitly.
"cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
"cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
-}
// Convert packed single to packed double
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
// SSE2 instructions without OpSize prefix
def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
"cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
-let isAsmParserOnly = 1 in {
def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vcvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
[(set VR128:$dst, (int_x86_sse2_cvtps2pd
(load addr:$src)))]>,
VEX, Requires<[HasAVX]>;
-}
def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
TB, Requires<[HasSSE2]>;
// Convert packed double to packed single
-let isAsmParserOnly = 1 in {
// The assembler can recognize rr 256-bit instructions by seeing a ymm
// register, but the same isn't true when using memory operands instead.
// Provide other assembly rr and rm forms to address this explicitly.
"cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
"cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
-}
def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
-let isAsmParserOnly = 1 in {
def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
"cvtpd2ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2ps
(memop addr:$src)))]>;
-}
def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
(VCVTTPS2DQYrm addr:$src)>;
+// Match fround and fextend for 128/256-bit conversions
+def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
+ (VCVTPD2PSYrr VR256:$src)>;
+def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
+ (VCVTPD2PSYrm addr:$src)>;
+
+def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
+ (VCVTPS2PDYrr VR128:$src)>;
+def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
+ (VCVTPS2PDYrm addr:$src)>;
+
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Compare Instructions
//===----------------------------------------------------------------------===//
// sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
string asm, string asm_alt> {
- def rr : SIi8<0xC2, MRMSrcReg,
- (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
- asm, []>;
- let mayLoad = 1 in
- def rm : SIi8<0xC2, MRMSrcMem,
- (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
- asm, []>;
- // Accept explicit immediate argument form instead of comparison code.
let isAsmParserOnly = 1 in {
- def rr_alt : SIi8<0xC2, MRMSrcReg,
- (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
- asm_alt, []>;
+ def rr : SIi8<0xC2, MRMSrcReg,
+ (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
+ asm, []>;
let mayLoad = 1 in
- def rm_alt : SIi8<0xC2, MRMSrcMem,
- (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
- asm_alt, []>;
+ def rm : SIi8<0xC2, MRMSrcMem,
+ (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
+ asm, []>;
}
+
+ // Accept explicit immediate argument form instead of comparison code.
+ def rr_alt : SIi8<0xC2, MRMSrcReg,
+ (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
+ asm_alt, []>;
+ let mayLoad = 1 in
+ def rm_alt : SIi8<0xC2, MRMSrcMem,
+ (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
+ asm_alt, []>;
}
-let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
+let neverHasSideEffects = 1 in {
defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
"cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
"cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
XD, VEX_4V;
}
-let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
- defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
- "cmp${cc}ss\t{$src, $dst|$dst, $src}",
- "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
- defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
- "cmp${cc}sd\t{$src, $dst|$dst, $src}",
- "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
+let Constraints = "$src1 = $dst" in {
+def CMPSSrr : SIi8<0xC2, MRMSrcReg,
+ (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
+ "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
+ [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
+def CMPSSrm : SIi8<0xC2, MRMSrcMem,
+ (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
+ "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
+ [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
+def CMPSDrr : SIi8<0xC2, MRMSrcReg,
+ (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
+ "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
+ [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
+def CMPSDrm : SIi8<0xC2, MRMSrcMem,
+ (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
+ "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
+ [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
}
-
-multiclass sse12_cmp_scalar_int<RegisterClass RC, Operand memopr,
- ComplexPattern mem_cpat, Intrinsic Int, string asm> {
+let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
+def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
+ (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
+ "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
+def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
+ (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
+ "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
+def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
+ (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
+ "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
+def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
+ (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
+ "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
+}
+
+multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
+ Intrinsic Int, string asm> {
def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
[(set VR128:$dst, (Int VR128:$src1,
VR128:$src, imm:$cc))]>;
def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, memopr:$src, SSECC:$cc), asm,
+ (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
[(set VR128:$dst, (Int VR128:$src1,
- mem_cpat:$src, imm:$cc))]>;
+ (load addr:$src), imm:$cc))]>;
}
// Aliases to match intrinsics which expect XMM operand(s).
-
-let isAsmParserOnly = 1 in {
- defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, ssmem, sse_load_f32,
- int_x86_sse_cmp_ss,
- "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
- XS, VEX_4V;
- defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, sdmem, sse_load_f64,
- int_x86_sse2_cmp_sd,
- "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
- XD, VEX_4V;
-}
+defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
+ "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
+ XS, VEX_4V;
+defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
+ "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
+ XD, VEX_4V;
let Constraints = "$src1 = $dst" in {
- defm Int_CMPSS : sse12_cmp_scalar_int<VR128, ssmem, sse_load_f32,
- int_x86_sse_cmp_ss,
+ defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
"cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
- defm Int_CMPSD : sse12_cmp_scalar_int<VR128, sdmem, sse_load_f64,
- int_x86_sse2_cmp_sd,
+ defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
"cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
}
}
let Defs = [EFLAGS] in {
- let isAsmParserOnly = 1 in {
- defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
- "ucomiss", SSEPackedSingle>, VEX;
- defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
- "ucomisd", SSEPackedDouble>, OpSize, VEX;
- let Pattern = []<dag> in {
- defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
- "comiss", SSEPackedSingle>, VEX;
- defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
- "comisd", SSEPackedDouble>, OpSize, VEX;
- }
-
- defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
- load, "ucomiss", SSEPackedSingle>, VEX;
- defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
- load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
-
- defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
- load, "comiss", SSEPackedSingle>, VEX;
- defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
- load, "comisd", SSEPackedDouble>, OpSize, VEX;
+ defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
+ "ucomiss", SSEPackedSingle>, VEX;
+ defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
+ "ucomisd", SSEPackedDouble>, OpSize, VEX;
+ let Pattern = []<dag> in {
+ defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
+ "comiss", SSEPackedSingle>, VEX;
+ defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
+ "comisd", SSEPackedDouble>, OpSize, VEX;
}
+
+ defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
+ load, "ucomiss", SSEPackedSingle>, VEX;
+ defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
+ load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
+
+ defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
+ load, "comiss", SSEPackedSingle>, VEX;
+ defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
+ load, "comisd", SSEPackedDouble>, OpSize, VEX;
defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
"ucomiss", SSEPackedSingle>, TB;
defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
Intrinsic Int, string asm, string asm_alt,
Domain d> {
- def rri : PIi8<0xC2, MRMSrcReg,
- (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
- [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
- def rmi : PIi8<0xC2, MRMSrcMem,
- (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
- [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
- // Accept explicit immediate argument form instead of comparison code.
let isAsmParserOnly = 1 in {
- def rri_alt : PIi8<0xC2, MRMSrcReg,
- (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
- asm_alt, [], d>;
- def rmi_alt : PIi8<0xC2, MRMSrcMem,
- (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
- asm_alt, [], d>;
+ def rri : PIi8<0xC2, MRMSrcReg,
+ (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
+ [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
+ def rmi : PIi8<0xC2, MRMSrcMem,
+ (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
+ [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
}
-}
-let isAsmParserOnly = 1 in {
- defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
- "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
- "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
- SSEPackedSingle>, VEX_4V;
- defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
- "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
- "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
- SSEPackedDouble>, OpSize, VEX_4V;
- let Pattern = []<dag> in {
- defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_sse_cmp_ps,
- "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
- "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
- SSEPackedSingle>, VEX_4V;
- defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_sse2_cmp_pd,
- "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
- "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
- SSEPackedDouble>, OpSize, VEX_4V;
- }
-}
+ // Accept explicit immediate argument form instead of comparison code.
+ def rri_alt : PIi8<0xC2, MRMSrcReg,
+ (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
+ asm_alt, [], d>;
+ def rmi_alt : PIi8<0xC2, MRMSrcMem,
+ (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
+ asm_alt, [], d>;
+}
+
+defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
+ "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
+ "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
+ SSEPackedSingle>, VEX_4V;
+defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
+ "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
+ "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
+ SSEPackedDouble>, OpSize, VEX_4V;
+defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
+ "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
+ "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
+ SSEPackedSingle>, VEX_4V;
+defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
+ "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
+ "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
+ SSEPackedDouble>, OpSize, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
"cmp${cc}ps\t{$src, $dst|$dst, $src}",
SSEPackedDouble>, TB, OpSize;
}
+let Predicates = [HasSSE1] in {
def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
(CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
(CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
+}
+
+let Predicates = [HasSSE2] in {
def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
(CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
(CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
+}
+
+let Predicates = [HasAVX] in {
+def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
+ (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
+def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
+ (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
+def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
+ (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
+def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
+ (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
+
+def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
+ (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
+def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
+ (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
+def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
+ (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
+def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
+ (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
+}
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Shuffle Instructions
(vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
}
-let isAsmParserOnly = 1 in {
- defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
- "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- memopv4f32, SSEPackedSingle>, VEX_4V;
- defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
- "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- memopv8f32, SSEPackedSingle>, VEX_4V;
- defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
- "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
- memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
- defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
- "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
- memopv4f64, SSEPackedDouble>, OpSize, VEX_4V;
-}
+defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
+ "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ memopv4f32, SSEPackedSingle>, TB, VEX_4V;
+defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
+ "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ memopv8f32, SSEPackedSingle>, TB, VEX_4V;
+defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
+ "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
+ memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
+defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
+ "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
+ memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
memopv2f64, SSEPackedDouble>, TB, OpSize;
}
+let Predicates = [HasSSE1] in {
+ def : Pat<(v4f32 (X86Shufps VR128:$src1,
+ (memopv4f32 addr:$src2), (i8 imm:$imm))),
+ (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ def : Pat<(v4i32 (X86Shufps VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
+ (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
+ // fall back to this for SSE1)
+ def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
+ (SHUFPSrri VR128:$src2, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special unary SHUFPSrri case.
+ def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
+ (SHUFPSrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+}
+
+let Predicates = [HasSSE2] in {
+ // Special binary v4i32 shuffle cases with SHUFPS.
+ def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
+ (SHUFPSrri VR128:$src1, VR128:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)))),
+ (SHUFPSrmi VR128:$src1, addr:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special unary SHUFPDrri cases.
+ def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
+ (SHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
+ (SHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special binary v2i64 shuffle cases using SHUFPDrri.
+ def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
+ (SHUFPDrri VR128:$src1, VR128:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Generic SHUFPD patterns
+ def : Pat<(v2f64 (X86Shufps VR128:$src1,
+ (memopv2f64 addr:$src2), (i8 imm:$imm))),
+ (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
+}
+
+let Predicates = [HasAVX] in {
+ def : Pat<(v4f32 (X86Shufps VR128:$src1,
+ (memopv4f32 addr:$src2), (i8 imm:$imm))),
+ (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ def : Pat<(v4i32 (X86Shufps VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
+ (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
+ // fall back to this for SSE1)
+ def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
+ (VSHUFPSrri VR128:$src2, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special unary SHUFPSrri case.
+ def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
+ (VSHUFPSrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special binary v4i32 shuffle cases with SHUFPS.
+ def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
+ (VSHUFPSrri VR128:$src1, VR128:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)))),
+ (VSHUFPSrmi VR128:$src1, addr:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special unary SHUFPDrri cases.
+ def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
+ (VSHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
+ (VSHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Special binary v2i64 shuffle cases using SHUFPDrri.
+ def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
+ (VSHUFPDrri VR128:$src1, VR128:$src2,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+ // Generic VSHUFPD patterns
+ def : Pat<(v2f64 (X86Shufps VR128:$src1,
+ (memopv2f64 addr:$src2), (i8 imm:$imm))),
+ (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
+ def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
+ def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
+}
+
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Unpack Instructions
//===----------------------------------------------------------------------===//
}
let AddedComplexity = 10 in {
- let isAsmParserOnly = 1 in {
- defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
- VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, VEX_4V;
- defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
- VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedDouble>, OpSize, VEX_4V;
- defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
- VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, VEX_4V;
- defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
- VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedDouble>, OpSize, VEX_4V;
-
- defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
- VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, VEX_4V;
- defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
- VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedDouble>, OpSize, VEX_4V;
- defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
- VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, VEX_4V;
- defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
- VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedDouble>, OpSize, VEX_4V;
- }
+ defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
+ VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ SSEPackedSingle>, VEX_4V;
+ defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
+ VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ SSEPackedDouble>, OpSize, VEX_4V;
+ defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
+ VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ SSEPackedSingle>, VEX_4V;
+ defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
+ VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ SSEPackedDouble>, OpSize, VEX_4V;
+
+ defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
+ VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ SSEPackedSingle>, VEX_4V;
+ defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
+ VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ SSEPackedDouble>, OpSize, VEX_4V;
+ defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
+ VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ SSEPackedSingle>, VEX_4V;
+ defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
+ VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ SSEPackedDouble>, OpSize, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
/// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
Domain d> {
- def rr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
- !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
+ def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
+ !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
[(set GR32:$dst, (Int RC:$src))], d>;
+ def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
+ !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
}
-// Mask creation
defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
SSEPackedSingle>, TB;
defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
SSEPackedDouble>, TB, OpSize;
-let isAsmParserOnly = 1 in {
+def : Pat<(i32 (X86fgetsign FR32:$src)),
+ (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
+ sub_ss))>, Requires<[HasSSE1]>;
+def : Pat<(i64 (X86fgetsign FR32:$src)),
+ (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
+ sub_ss))>, Requires<[HasSSE1]>;
+def : Pat<(i32 (X86fgetsign FR64:$src)),
+ (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
+ sub_sd))>, Requires<[HasSSE2]>;
+def : Pat<(i64 (X86fgetsign FR64:$src)),
+ (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
+ sub_sd))>, Requires<[HasSSE2]>;
+
+let Predicates = [HasAVX] in {
defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
- "movmskps", SSEPackedSingle>, VEX;
+ "movmskps", SSEPackedSingle>, TB, VEX;
defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
- "movmskpd", SSEPackedDouble>, OpSize,
+ "movmskpd", SSEPackedDouble>, TB, OpSize,
+ VEX;
+ defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
+ "movmskps", SSEPackedSingle>, TB, VEX;
+ defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
+ "movmskpd", SSEPackedDouble>, TB, OpSize,
VEX;
- // FIXME: merge with multiclass above when the intrinsics come.
+ def : Pat<(i32 (X86fgetsign FR32:$src)),
+ (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
+ sub_ss))>;
+ def : Pat<(i64 (X86fgetsign FR32:$src)),
+ (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
+ sub_ss))>;
+ def : Pat<(i32 (X86fgetsign FR64:$src)),
+ (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
+ sub_sd))>;
+ def : Pat<(i64 (X86fgetsign FR64:$src)),
+ (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
+ sub_sd))>;
+
+ // Assembler Only
def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
"movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
"movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
VEX;
-
- def VMOVMSKPSYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
- "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
- def VMOVMSKPDYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
- "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
- VEX;
-
def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
"movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
[(set FR64:$dst, fpimm0)]>,
Requires<[HasSSE2]>, TB, OpSize;
+def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
+ [(set FR32:$dst, fp32imm0)]>,
+ Requires<[HasAVX]>, TB, OpSize, VEX_4V;
+def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
+ [(set FR64:$dst, fpimm0)]>,
+ Requires<[HasAVX]>, TB, OpSize, VEX_4V;
}
// Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
///
multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
SDNode OpNode> {
- let isAsmParserOnly = 1 in {
- defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
- FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
+ defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
+ FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
- defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
- FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
- }
+ defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
+ FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
/// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
///
multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
- SDNode OpNode, int HasPat = 0,
- list<list<dag>> Pattern = []> {
- let isAsmParserOnly = 1, Pattern = []<dag> in {
- defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
- !strconcat(OpcodeStr, "ps"), f128mem,
- !if(HasPat, Pattern[0], // rr
- [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
- VR128:$src2)))]),
- !if(HasPat, Pattern[2], // rm
- [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
- (memopv2i64 addr:$src2)))]), 0>,
- VEX_4V;
-
- defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
- !strconcat(OpcodeStr, "pd"), f128mem,
- !if(HasPat, Pattern[1], // rr
- [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
- (bc_v2i64 (v2f64
- VR128:$src2))))]),
- !if(HasPat, Pattern[3], // rm
- [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
- (memopv2i64 addr:$src2)))]), 0>,
- OpSize, VEX_4V;
- }
+ SDNode OpNode> {
+ // In AVX no need to add a pattern for 128-bit logical rr ps, because they
+ // are all promoted to v2i64, and the patterns are covered by the int
+ // version. This is needed in SSE only, because v2i64 isn't supported on
+ // SSE1, but only on SSE2.
+ defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
+ !strconcat(OpcodeStr, "ps"), f128mem, [],
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
+ (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
+
+ defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
+ !strconcat(OpcodeStr, "pd"), f128mem,
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (bc_v2i64 (v2f64 VR128:$src2))))],
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (memopv2i64 addr:$src2)))], 0>,
+ TB, OpSize, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
!strconcat(OpcodeStr, "ps"), f128mem,
- !if(HasPat, Pattern[0], // rr
- [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
- VR128:$src2)))]),
- !if(HasPat, Pattern[2], // rm
- [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
- (memopv2i64 addr:$src2)))])>, TB;
+ [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
+ (memopv2i64 addr:$src2)))]>, TB;
defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
!strconcat(OpcodeStr, "pd"), f128mem,
- !if(HasPat, Pattern[1], // rr
- [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
- (bc_v2i64 (v2f64
- VR128:$src2))))]),
- !if(HasPat, Pattern[3], // rm
- [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
- (memopv2i64 addr:$src2)))])>,
- TB, OpSize;
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (bc_v2i64 (v2f64 VR128:$src2))))],
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (memopv2i64 addr:$src2)))]>, TB, OpSize;
}
}
/// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
///
-let isAsmParserOnly = 1 in {
-multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
+multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
+ SDNode OpNode> {
defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
- !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
+ !strconcat(OpcodeStr, "ps"), f256mem,
+ [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
+ [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
+ (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
- !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
-}
+ !strconcat(OpcodeStr, "pd"), f256mem,
+ [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
+ (bc_v4i64 (v4f64 VR256:$src2))))],
+ [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
+ (memopv4i64 addr:$src2)))], 0>,
+ TB, OpSize, VEX_4V;
}
// AVX 256-bit packed logical ops forms
-defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
-defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
-defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
-let isCommutable = 0 in
- defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
+defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
+defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
+defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
+defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
defm AND : sse12_fp_packed_logical<0x54, "and", and>;
defm OR : sse12_fp_packed_logical<0x56, "or", or>;
defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
let isCommutable = 0 in
- defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
- // single r+r
- [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
- (bc_v2i64 (v4i32 immAllOnesV))),
- VR128:$src2)))],
- // double r+r
- [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
- (bc_v2i64 (v2f64 VR128:$src2))))],
- // single r+m
- [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
- (bc_v2i64 (v4i32 immAllOnesV))),
- (memopv2i64 addr:$src2))))],
- // double r+m
- [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
- (memopv2i64 addr:$src2)))]]>;
+ defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Arithmetic Instructions
}
// Binary Arithmetic instructions
-let isAsmParserOnly = 1 in {
- defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
- basic_sse12_fp_binop_s_int<0x58, "add", 0>,
- basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
- basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
- defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
- basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
- basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
- basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
+defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
+ basic_sse12_fp_binop_s_int<0x58, "add", 0>,
+ basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
+ basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
+defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
+ basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
+ basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
+ basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
- let isCommutable = 0 in {
- defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
- basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
- basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
- basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
- defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
- basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
- basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
- basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
- defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
- basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
- basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
- basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
- basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
- basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
- defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
- basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
- basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
- basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
- basic_sse12_fp_binop_p_y_int<0x5D, "min">,
- basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
- }
+let isCommutable = 0 in {
+ defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
+ basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
+ basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
+ basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
+ defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
+ basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
+ basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
+ basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
+ defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
+ basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
+ basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
+ basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
+ basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
+ basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
+ defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
+ basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
+ basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
+ basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
+ basic_sse12_fp_binop_p_y_int<0x5D, "min">,
+ basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
}
let Constraints = "$src1 = $dst" in {
}
/// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
-multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
- SDNode OpNode, Intrinsic F32Int> {
+multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
!strconcat(OpcodeStr,
"ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
- !strconcat(OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>, XS, Requires<[HasAVX, OptForSize]>;
- def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
!strconcat(OpcodeStr,
- "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
- [(set VR128:$dst, (F32Int VR128:$src))]>;
- def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+ def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins ssmem:$src1, VR128:$src2),
!strconcat(OpcodeStr,
- "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
- [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
}
/// sse1_fp_unop_p - SSE1 unops in packed form.
}
/// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
-multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
- SDNode OpNode, Intrinsic F64Int> {
+multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
!strconcat(OpcodeStr,
"sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
- (ins FR64:$src1, f64mem:$src2),
+ def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
+ !strconcat(OpcodeStr,
+ "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+ def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, sdmem:$src2),
!strconcat(OpcodeStr,
"sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
- [(set VR128:$dst, (F64Int VR128:$src))]>;
- def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
- !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
- [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
}
/// sse2_fp_unop_p - SSE2 unops in vector forms.
[(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
// Square root.
- defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
- sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
- VEX_4V;
+ defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
+ sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V;
defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
- defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
- int_x86_sse_rsqrt_ss>, VEX_4V;
+ defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V;
defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
- defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
- VEX_4V;
+ defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V;
defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
}
+def : Pat<(f32 (fsqrt FR32:$src)),
+ (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
+def : Pat<(f32 (fsqrt (load addr:$src))),
+ (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+def : Pat<(f64 (fsqrt FR64:$src)),
+ (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
+def : Pat<(f64 (fsqrt (load addr:$src))),
+ (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+
+def : Pat<(f32 (X86frsqrt FR32:$src)),
+ (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
+def : Pat<(f32 (X86frsqrt (load addr:$src))),
+ (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+
+def : Pat<(f32 (X86frcp FR32:$src)),
+ (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
+def : Pat<(f32 (X86frcp (load addr:$src))),
+ (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+
+let Predicates = [HasAVX] in {
+def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
+ (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
+ (VSQRTSSr (f32 (IMPLICIT_DEF)),
+ (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
+ sub_ss)>;
+def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
+ (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+
+def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
+ (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
+ (VSQRTSDr (f64 (IMPLICIT_DEF)),
+ (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
+ sub_sd)>;
+def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
+ (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
+
+def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
+ (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
+ (VRSQRTSSr (f32 (IMPLICIT_DEF)),
+ (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
+ sub_ss)>;
+def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
+ (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+
+def : Pat<(int_x86_sse_rcp_ss VR128:$src),
+ (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
+ (VRCPSSr (f32 (IMPLICIT_DEF)),
+ (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
+ sub_ss)>;
+def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
+ (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+}
+
// Square root.
defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
// SSE 1 & 2 - Non-temporal stores
//===----------------------------------------------------------------------===//
-let isAsmParserOnly = 1 in {
- def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
- (ins i128mem:$dst, VR128:$src),
- "movntps\t{$src, $dst|$dst, $src}",
- [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
- def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
- (ins i128mem:$dst, VR128:$src),
- "movntpd\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
-
- let ExeDomain = SSEPackedInt in
- def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
+let AddedComplexity = 400 in { // Prefer non-temporal versions
+ def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
(ins f128mem:$dst, VR128:$src),
- "movntdq\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
-
- let AddedComplexity = 400 in { // Prefer non-temporal versions
- def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
- (ins f128mem:$dst, VR128:$src),
- "movntps\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v4f32 VR128:$src),
- addr:$dst)]>, VEX;
- def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
- (ins f128mem:$dst, VR128:$src),
- "movntpd\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v2f64 VR128:$src),
- addr:$dst)]>, VEX;
- def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
- (ins f128mem:$dst, VR128:$src),
- "movntdq\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v2f64 VR128:$src),
- addr:$dst)]>, VEX;
- let ExeDomain = SSEPackedInt in
- def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
+ "movntps\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v4f32 VR128:$src),
+ addr:$dst)]>, VEX;
+ def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
+ (ins f128mem:$dst, VR128:$src),
+ "movntpd\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v2f64 VR128:$src),
+ addr:$dst)]>, VEX;
+ def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
(ins f128mem:$dst, VR128:$src),
"movntdq\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v4f32 VR128:$src),
+ [(alignednontemporalstore (v2f64 VR128:$src),
addr:$dst)]>, VEX;
- def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
- (ins f256mem:$dst, VR256:$src),
- "movntps\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v8f32 VR256:$src),
- addr:$dst)]>, VEX;
- def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
- (ins f256mem:$dst, VR256:$src),
- "movntpd\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v4f64 VR256:$src),
- addr:$dst)]>, VEX;
- def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
- (ins f256mem:$dst, VR256:$src),
- "movntdq\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v4f64 VR256:$src),
- addr:$dst)]>, VEX;
- let ExeDomain = SSEPackedInt in
- def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
+ let ExeDomain = SSEPackedInt in
+ def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
+ (ins f128mem:$dst, VR128:$src),
+ "movntdq\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v4f32 VR128:$src),
+ addr:$dst)]>, VEX;
+
+ def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
+ (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
+
+ def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
+ (ins f256mem:$dst, VR256:$src),
+ "movntps\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v8f32 VR256:$src),
+ addr:$dst)]>, VEX;
+ def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
+ (ins f256mem:$dst, VR256:$src),
+ "movntpd\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v4f64 VR256:$src),
+ addr:$dst)]>, VEX;
+ def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
(ins f256mem:$dst, VR256:$src),
"movntdq\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v8f32 VR256:$src),
+ [(alignednontemporalstore (v4f64 VR256:$src),
addr:$dst)]>, VEX;
- }
+ let ExeDomain = SSEPackedInt in
+ def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
+ (ins f256mem:$dst, VR256:$src),
+ "movntdq\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v8f32 VR256:$src),
+ addr:$dst)]>, VEX;
}
-def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
- "movntps\t{$src, $dst|$dst, $src}",
- [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
-def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
- "movntpd\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
-
-let ExeDomain = SSEPackedInt in
-def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
- "movntdq\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
+def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
+ (VMOVNTDQYmr addr:$dst, VR256:$src)>;
+def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
+ (VMOVNTPDYmr addr:$dst, VR256:$src)>;
+def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
+ (VMOVNTPSYmr addr:$dst, VR256:$src)>;
let AddedComplexity = 400 in { // Prefer non-temporal versions
def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movntdq\t{$src, $dst|$dst, $src}",
[(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
+def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
+ (MOVNTDQmr addr:$dst, VR128:$src)>;
+
// There is no AVX form for instructions below this point
def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
- "movnti\t{$src, $dst|$dst, $src}",
+ "movnti{l}\t{$src, $dst|$dst, $src}",
[(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
TB, Requires<[HasSSE2]>;
-
def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
- "movnti\t{$src, $dst|$dst, $src}",
+ "movnti{q}\t{$src, $dst|$dst, $src}",
[(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
TB, Requires<[HasSSE2]>;
-
}
-def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
- "movnti\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
- TB, Requires<[HasSSE2]>;
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Misc Instructions (No AVX form)
// Prefetch intrinsic.
def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
- "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
+ "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
- "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
+ "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
- "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
+ "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
- "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
+ "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
// Load, store, and memory fence
def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
// Alias instructions that map zero vector to pxor / xorp* for sse.
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-zeros value if folding it would be beneficial.
-// FIXME: Change encoding to pseudo!
+// FIXME: Change encoding to pseudo! This is blocked right now by the x86
+// JIT implementation, it does not expand the instructions below like
+// X86MCInstLower does.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
isCodeGenOnly = 1 in {
def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
[(set VR128:$dst, (v4i32 immAllZerosV))]>;
}
+// The same as done above but for AVX. The 128-bit versions are the
+// same, but re-encoded. The 256-bit does not support PI version, and
+// doesn't need it because on sandy bridge the register is set to zero
+// at the rename stage without using any execution unit, so SET0PSY
+// and SET0PDY can be used for vector int instructions without penalty
+// FIXME: Change encoding to pseudo! This is blocked right now by the x86
+// JIT implementatioan, it does not expand the instructions below like
+// X86MCInstLower does.
+let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
+ isCodeGenOnly = 1, Predicates = [HasAVX] in {
+def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
+def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
+def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
+ [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
+def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
+ [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
+let ExeDomain = SSEPackedInt in
+def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4i32 immAllZerosV))]>;
+}
+
def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
(f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
+// AVX has no support for 256-bit integer instructions, but since the 128-bit
+// VPXOR instruction writes zero to its upper part, it's safe build zeros.
+def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
+def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
+ (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
+
+def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
+def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
+ (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
+
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Load/Store XCSR register
//===----------------------------------------------------------------------===//
-let isAsmParserOnly = 1 in {
- def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
- "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
- def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
- "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
-}
+def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
+ "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
+def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
+ "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
"ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
let ExeDomain = SSEPackedInt in { // SSE integer instructions
-let isAsmParserOnly = 1 in {
- let neverHasSideEffects = 1 in {
- def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
- def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
- }
- def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
- def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
-
- let canFoldAsLoad = 1, mayLoad = 1 in {
- def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
- "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
- def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
- "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
- let Predicates = [HasAVX] in {
- def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
- "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
- def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
- "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
- }
- }
+let neverHasSideEffects = 1 in {
+def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
+def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+ "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
+}
+def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
+def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+ "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
- let mayStore = 1 in {
- def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
- (ins i128mem:$dst, VR128:$src),
- "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
- def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
- (ins i256mem:$dst, VR256:$src),
- "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
- let Predicates = [HasAVX] in {
- def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
+let canFoldAsLoad = 1, mayLoad = 1 in {
+def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
+ "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
+def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
+ "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
+let Predicates = [HasAVX] in {
+ def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
- def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
+ def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
"vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
- }
- }
+}
+}
+
+let mayStore = 1 in {
+def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
+ (ins i128mem:$dst, VR128:$src),
+ "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
+def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
+ (ins i256mem:$dst, VR256:$src),
+ "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
+let Predicates = [HasAVX] in {
+def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
+ "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
+def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
+ "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
+}
}
let neverHasSideEffects = 1 in
def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", []>;
+def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "movdqu\t{$src, $dst|$dst, $src}",
+ []>, XS, Requires<[HasSSE2]>;
+
let canFoldAsLoad = 1, mayLoad = 1 in {
def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movdqa\t{$src, $dst|$dst, $src}",
}
// Intrinsic forms of MOVDQU load and store
-let isAsmParserOnly = 1 in {
-let canFoldAsLoad = 1 in
-def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
- "vmovdqu\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
- XS, VEX, Requires<[HasAVX]>;
def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
"vmovdqu\t{$src, $dst|$dst, $src}",
[(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
XS, VEX, Requires<[HasAVX]>;
-}
-let canFoldAsLoad = 1 in
-def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
- "movdqu\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
- XS, Requires<[HasSSE2]>;
def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
"movdqu\t{$src, $dst|$dst, $src}",
[(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
} // ExeDomain = SSEPackedInt
+def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
+def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
+ (VMOVDQUYmr addr:$dst, VR256:$src)>;
+
//===---------------------------------------------------------------------===//
// SSE2 - Packed Integer Arithmetic Instructions
//===---------------------------------------------------------------------===//
// 128-bit Integer Arithmetic
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
// SSE2 - Packed Integer Logical Instructions
//===---------------------------------------------------------------------===//
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
VEX_4V;
def VPANDNrr : PDI<0xDF, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
- VR128:$src2)))]>, VEX_4V;
+ [(set VR128:$dst,
+ (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
def VPANDNrm : PDI<0xDF, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
"vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
- (memopv2i64 addr:$src2))))]>,
- VEX_4V;
+ [(set VR128:$dst, (X86andnp VR128:$src1,
+ (memopv2i64 addr:$src2)))]>, VEX_4V;
}
}
}
def PANDNrr : PDI<0xDF, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "pandn\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
- VR128:$src2)))]>;
+ "pandn\t{$src2, $dst|$dst, $src2}", []>;
def PANDNrm : PDI<0xDF, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- "pandn\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
- (memopv2i64 addr:$src2))))]>;
+ "pandn\t{$src2, $dst|$dst, $src2}", []>;
}
} // Constraints = "$src1 = $dst"
// SSE2 - Packed Integer Comparison Instructions
//===---------------------------------------------------------------------===//
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
0>, VEX_4V;
defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
0>, VEX_4V;
defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
0>, VEX_4V;
+
+ def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
+ (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
+ (VPCMPEQBrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
+ (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
+ (VPCMPEQWrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
+ (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
+ (VPCMPEQDrm VR128:$src1, addr:$src2)>;
+
+ def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
+ (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
+ (VPCMPGTBrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
+ (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
+ (VPCMPGTWrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
+ (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
+ (VPCMPGTDrm VR128:$src1, addr:$src2)>;
}
let Constraints = "$src1 = $dst" in {
// SSE2 - Packed Integer Pack Instructions
//===---------------------------------------------------------------------===//
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
0, 0>, VEX_4V;
defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
}
} // ExeDomain = SSEPackedInt
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
let AddedComplexity = 5 in
defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
VEX;
// SSE2 with ImmT == Imm8 and XD prefix.
defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
VEX;
+
+ let AddedComplexity = 5 in
+ def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
+ (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
+ // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
+ def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
+ (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
+
+ def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
+ (i8 imm:$imm))),
+ (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
+ def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
+ (i8 imm:$imm))),
+ (VPSHUFDmi addr:$src1, imm:$imm)>;
+ def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
+ def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
+ def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
+ (VPSHUFHWri VR128:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
+ (i8 imm:$imm))),
+ (VPSHUFHWmi addr:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
+ (VPSHUFLWri VR128:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
+ (i8 imm:$imm))),
+ (VPSHUFLWmi addr:$src, imm:$imm)>;
}
let Predicates = [HasSSE2] in {
// SSE2 with ImmT == Imm8 and XD prefix.
defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
+
+ let AddedComplexity = 5 in
+ def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
+ (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
+ // Unary v4f32 shuffle with PSHUF* in order to fold a load.
+ def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
+ (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
+
+ def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
+ (i8 imm:$imm))),
+ (PSHUFDmi addr:$src1, imm:$imm)>;
+ def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
+ (i8 imm:$imm))),
+ (PSHUFDmi addr:$src1, imm:$imm)>;
+ def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (PSHUFDri VR128:$src1, imm:$imm)>;
+ def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (PSHUFDri VR128:$src1, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
+ (PSHUFHWri VR128:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
+ (i8 imm:$imm))),
+ (PSHUFHWmi addr:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
+ (PSHUFLWri VR128:$src, imm:$imm)>;
+ def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
+ (i8 imm:$imm))),
+ (PSHUFLWmi addr:$src, imm:$imm)>;
}
//===---------------------------------------------------------------------===//
let ExeDomain = SSEPackedInt in {
multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
- PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
+ SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
def rr : PDI<opc, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
+ [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
def rm : PDI<opc, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set VR128:$dst, (unp_frag VR128:$src1,
+ [(set VR128:$dst, (OpNode VR128:$src1,
(bc_frag (memopv2i64
addr:$src2))))]>;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
- defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
- 0>, VEX_4V;
- defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
- 0>, VEX_4V;
- defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
- 0>, VEX_4V;
+let Predicates = [HasAVX] in {
+ defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Punpcklbw,
+ bc_v16i8, 0>, VEX_4V;
+ defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Punpcklwd,
+ bc_v8i16, 0>, VEX_4V;
+ defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Punpckldq,
+ bc_v4i32, 0>, VEX_4V;
/// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
/// knew to collapse (bitconvert VT to VT) into its operand.
def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set VR128:$dst,
- (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
+ VR128:$src2)))]>, VEX_4V;
def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set VR128:$dst,
- (v2i64 (unpckl VR128:$src1,
- (memopv2i64 addr:$src2))))]>, VEX_4V;
-
- defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
- 0>, VEX_4V;
- defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
- 0>, VEX_4V;
- defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
- 0>, VEX_4V;
+ (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
+ "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
+ (memopv2i64 addr:$src2))))]>, VEX_4V;
+
+ defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Punpckhbw,
+ bc_v16i8, 0>, VEX_4V;
+ defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Punpckhwd,
+ bc_v8i16, 0>, VEX_4V;
+ defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Punpckhdq,
+ bc_v4i32, 0>, VEX_4V;
/// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
/// knew to collapse (bitconvert VT to VT) into its operand.
def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set VR128:$dst,
- (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
+ VR128:$src2)))]>, VEX_4V;
def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set VR128:$dst,
- (v2i64 (unpckh VR128:$src1,
- (memopv2i64 addr:$src2))))]>, VEX_4V;
+ (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
+ "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
+ (memopv2i64 addr:$src2))))]>, VEX_4V;
}
let Constraints = "$src1 = $dst" in {
- defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
- defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
- defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
+ defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Punpcklbw, bc_v16i8>;
+ defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Punpcklwd, bc_v8i16>;
+ defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Punpckldq, bc_v4i32>;
/// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
/// knew to collapse (bitconvert VT to VT) into its operand.
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"punpcklqdq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
- (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
+ (v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)))]>;
def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
"punpcklqdq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
- (v2i64 (unpckl VR128:$src1,
+ (v2i64 (X86Punpcklqdq VR128:$src1,
(memopv2i64 addr:$src2))))]>;
- defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
- defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
- defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
+ defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Punpckhbw, bc_v16i8>;
+ defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Punpckhwd, bc_v8i16>;
+ defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Punpckhdq, bc_v4i32>;
/// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
/// knew to collapse (bitconvert VT to VT) into its operand.
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"punpckhqdq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
- (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
+ (v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)))]>;
def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
"punpckhqdq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
- (v2i64 (unpckh VR128:$src1,
+ (v2i64 (X86Punpckhqdq VR128:$src1,
(memopv2i64 addr:$src2))))]>;
}
}
// Extract
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
(outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
"vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
imm:$src2))]>;
// Insert
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
let ExeDomain = SSEPackedInt in {
-let isAsmParserOnly = 1 in {
def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
"pmovmskb\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
"pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
-}
def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
"pmovmskb\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
let ExeDomain = SSEPackedInt in {
-let isAsmParserOnly = 1 in {
let Uses = [EDI] in
def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
(ins VR128:$src, VR128:$mask),
(ins VR128:$src, VR128:$mask),
"maskmovdqu\t{$mask, $src|$src, $mask}",
[(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
-}
let Uses = [EDI] in
def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
// SSE2 - Move Doubleword
//===---------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
// Move Int Doubleword to Packed Double Int
-let isAsmParserOnly = 1 in {
+//
def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
[(set VR128:$dst,
(v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
VEX;
-}
+def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
+ "mov{d|q}\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst,
+ (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
+def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
+ "mov{d|q}\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
+
def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
+def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
+ "mov{d|q}\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst,
+ (v2i64 (scalar_to_vector GR64:$src)))]>;
+def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
+ "mov{d|q}\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (bitconvert GR64:$src))]>;
-
+//===---------------------------------------------------------------------===//
// Move Int Doubleword to Single Scalar
-let isAsmParserOnly = 1 in {
+//
def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
"movd\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
VEX;
-}
def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (bitconvert GR32:$src))]>;
"movd\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
+//===---------------------------------------------------------------------===//
// Move Packed Doubleword Int to Packed Double Int
-let isAsmParserOnly = 1 in {
+//
def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(store (i32 (vector_extract (v4i32 VR128:$src),
(iPTR 0))), addr:$dst)]>, VEX;
-}
def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
[(store (i32 (vector_extract (v4i32 VR128:$src),
(iPTR 0))), addr:$dst)]>;
+def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
+ "mov{d|q}\t{$src, $dst|$dst, $src}",
+ [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
+ (iPTR 0)))]>;
+def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
+ "movq\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
+
+def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
+ "mov{d|q}\t{$src, $dst|$dst, $src}",
+ [(set GR64:$dst, (bitconvert FR64:$src))]>;
+def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
+ "movq\t{$src, $dst|$dst, $src}",
+ [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
+
+//===---------------------------------------------------------------------===//
// Move Scalar Single to Double Int
-let isAsmParserOnly = 1 in {
+//
def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
-}
def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (bitconvert FR32:$src))]>;
"movd\t{$src, $dst|$dst, $src}",
[(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
-// movd / movq to XMM register zero-extends
-let AddedComplexity = 15, isAsmParserOnly = 1 in {
+//===---------------------------------------------------------------------===//
+// Patterns and instructions to describe movd/movq to XMM register zero-extends
+//
+let AddedComplexity = 15 in {
def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (v4i32 (X86vzmovl
}
let AddedComplexity = 20 in {
-let isAsmParserOnly = 1 in
def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(MOVZDI2PDIrm addr:$src)>;
}
+// AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
+// Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
+def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
+ (v4i32 (scalar_to_vector GR32:$src)), (i32 0)))),
+ (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
+def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
+ (v2i64 (scalar_to_vector GR64:$src)), (i32 0)))),
+ (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
+
+// These are the correct encodings of the instructions so that we know how to
+// read correct assembly, even though we continue to emit the wrong ones for
+// compatibility with Darwin's buggy assembler.
+def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
+ (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
+def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
+ (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
+def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
+ (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
+ (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
+def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
+ (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
+def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
+ (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
+
//===---------------------------------------------------------------------===//
// SSE2 - Move Quadword
//===---------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
// Move Quadword Int to Packed Quadword Int
-let isAsmParserOnly = 1 in
+//
def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"vmovq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
+//===---------------------------------------------------------------------===//
// Move Packed Quadword Int to Quadword Int
-let isAsmParserOnly = 1 in
+//
def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
"movq\t{$src, $dst|$dst, $src}",
[(store (i64 (vector_extract (v2i64 VR128:$src),
def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
(f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
+//===---------------------------------------------------------------------===//
// Store / copy lower 64-bits of a XMM register.
-let isAsmParserOnly = 1 in
+//
def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
"movq\t{$src, $dst|$dst, $src}",
[(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
"movq\t{$src, $dst|$dst, $src}",
[(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
-let AddedComplexity = 20, isAsmParserOnly = 1 in
+let AddedComplexity = 20 in
def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"vmovq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
}
+//===---------------------------------------------------------------------===//
// Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
// IA32 document. movq xmm1, xmm2 does clear the high bits.
-let isAsmParserOnly = 1, AddedComplexity = 15 in
+//
+let AddedComplexity = 15 in
def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vmovq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
[(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
XS, Requires<[HasSSE2]>;
-let AddedComplexity = 20, isAsmParserOnly = 1 in
+let AddedComplexity = 20 in
def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"vmovq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (v2i64 (X86vzmovl
}
// Instructions to match in the assembler
-let isAsmParserOnly = 1 in {
def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
"movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
// Recognize "movd" with GR64 destination, but encode as a "movq"
def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
"movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
-}
// Instructions for the disassembler
// xr = XMM register
// xm = mem64
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
// Alias instructions that map zero vector to pxor / xorp* for sse.
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-ones value if folding it would be beneficial.
+// FIXME: Change encoding to pseudo! This is blocked right now by the x86
+// JIT implementation, it does not expand the instructions below like
+// X86MCInstLower does.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
- // FIXME: Change encoding to pseudo.
def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
[(set VR128:$dst, (v4i32 immAllOnesV))]>;
+let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
+ isCodeGenOnly = 1, ExeDomain = SSEPackedInt, Predicates = [HasAVX] in
+ def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
//===---------------------------------------------------------------------===//
// SSE3 - Conversion Instructions
//===---------------------------------------------------------------------===//
// Convert Packed Double FP to Packed DW Integers
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
// The assembler can recognize rr 256-bit instructions by seeing a ymm
// register, but the same isn't true when using memory operands instead.
// Provide other assembly rr and rm forms to address this explicitly.
def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
+def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
+ (VCVTPD2DQYrr VR256:$src)>;
+def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
+ (VCVTPD2DQYrm addr:$src)>;
+
// Convert Packed DW Integers to Packed Double FP
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
(VCVTPD2DQYrm addr:$src)>;
+def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
+ (VCVTDQ2PDYrr VR128:$src)>;
+def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
+ (VCVTDQ2PDYrm addr:$src)>;
+
//===---------------------------------------------------------------------===//
-// SSE3 - Move Instructions
+// SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
//===---------------------------------------------------------------------===//
-
-// Replicate Single FP
-multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
-def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
+ ValueType vt, RegisterClass RC, PatFrag mem_frag,
+ X86MemOperand x86memop> {
+def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (v4f32 (rep_frag
- VR128:$src, (undef))))]>;
-def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ [(set RC:$dst, (vt (OpNode RC:$src)))]>;
+def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (rep_frag
- (memopv4f32 addr:$src), (undef)))]>;
+ [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
}
-multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
- string OpcodeStr> {
-def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
-def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
+let Predicates = [HasAVX] in {
+ defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
+ v4f32, VR128, memopv4f32, f128mem>, VEX;
+ defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
+ v4f32, VR128, memopv4f32, f128mem>, VEX;
+ defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
+ v8f32, VR256, memopv8f32, f256mem>, VEX;
+ defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
+ v8f32, VR256, memopv8f32, f256mem>, VEX;
+}
+defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
+ memopv4f32, f128mem>;
+defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
+ memopv4f32, f128mem>;
+
+let Predicates = [HasSSE3] in {
+ def : Pat<(v4i32 (X86Movshdup VR128:$src)),
+ (MOVSHDUPrr VR128:$src)>;
+ def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
+ (MOVSHDUPrm addr:$src)>;
+ def : Pat<(v4i32 (X86Movsldup VR128:$src)),
+ (MOVSLDUPrr VR128:$src)>;
+ def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
+ (MOVSLDUPrm addr:$src)>;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
- // FIXME: Merge above classes when we have patterns for the ymm version
- defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
- defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
- defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
- defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
+let Predicates = [HasAVX] in {
+ def : Pat<(v4i32 (X86Movshdup VR128:$src)),
+ (VMOVSHDUPrr VR128:$src)>;
+ def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
+ (VMOVSHDUPrm addr:$src)>;
+ def : Pat<(v4i32 (X86Movsldup VR128:$src)),
+ (VMOVSLDUPrr VR128:$src)>;
+ def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
+ (VMOVSLDUPrm addr:$src)>;
+ def : Pat<(v8i32 (X86Movshdup VR256:$src)),
+ (VMOVSHDUPYrr VR256:$src)>;
+ def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
+ (VMOVSHDUPYrm addr:$src)>;
+ def : Pat<(v8i32 (X86Movsldup VR256:$src)),
+ (VMOVSLDUPYrr VR256:$src)>;
+ def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
+ (VMOVSLDUPYrm addr:$src)>;
}
-defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
-defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
-// Replicate Double FP
+//===---------------------------------------------------------------------===//
+// SSE3 - Replicate Double FP - MOVDDUP
+//===---------------------------------------------------------------------===//
+
multiclass sse3_replicate_dfp<string OpcodeStr> {
def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
(undef))))]>;
}
+// FIXME: Merge with above classe when there're patterns for the ymm version
multiclass sse3_replicate_dfp_y<string OpcodeStr> {
-def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- []>;
-def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- []>;
+let Predicates = [HasAVX] in {
+ def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ []>;
+ def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ []>;
+ }
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
- // FIXME: Merge above classes when we have patterns for the ymm version
- defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
- defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
-}
defm MOVDDUP : sse3_replicate_dfp<"movddup">;
+defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
+defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
-// Move Unaligned Integer
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
- def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
- "vlddqu\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
- def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
- "vlddqu\t{$src, $dst|$dst, $src}", []>, VEX;
+let Predicates = [HasSSE3] in {
+ def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
+ (undef)),
+ (MOVDDUPrm addr:$src)>;
+ let AddedComplexity = 5 in {
+ def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
+ (MOVDDUPrm addr:$src)>;
+ }
+ def : Pat<(X86Movddup (memopv2f64 addr:$src)),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
+ (MOVDDUPrm addr:$src)>;
+ def : Pat<(X86Movddup (bc_v2f64
+ (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
+ (MOVDDUPrm addr:$src)>;
}
-def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
- "lddqu\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
-def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
+let Predicates = [HasAVX] in {
+ def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
(undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
-
-// Several Move patterns
-let AddedComplexity = 5 in {
-def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
-def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
-def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
-def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
- (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
+ (VMOVDDUPrm addr:$src)>;
+ let AddedComplexity = 5 in {
+ def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
+ (VMOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
+ def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
+ (VMOVDDUPrm addr:$src)>;
+ }
+ def : Pat<(X86Movddup (memopv2f64 addr:$src)),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
+ def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
+ def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
+ def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
+ def : Pat<(X86Movddup (bc_v2f64
+ (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
+ (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
}
-// vector_shuffle v1, <undef> <1, 1, 3, 3>
-let AddedComplexity = 15 in
-def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
- (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
-let AddedComplexity = 20 in
-def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
- (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
+//===---------------------------------------------------------------------===//
+// SSE3 - Move Unaligned Integer
+//===---------------------------------------------------------------------===//
-// vector_shuffle v1, <undef> <0, 0, 2, 2>
-let AddedComplexity = 15 in
- def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
- (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
-let AddedComplexity = 20 in
- def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
- (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
+let Predicates = [HasAVX] in {
+ def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
+ "vlddqu\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
+ def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
+ "vlddqu\t{$src, $dst|$dst, $src}",
+ [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
+}
+def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
+ "lddqu\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
//===---------------------------------------------------------------------===//
// SSE3 - Arithmetic
[(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX],
+let Predicates = [HasAVX],
ExeDomain = SSEPackedDouble in {
defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
- f128mem, 0>, XD, VEX_4V;
+ f128mem, 0>, TB, XD, VEX_4V;
defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
- f128mem, 0>, OpSize, VEX_4V;
+ f128mem, 0>, TB, OpSize, VEX_4V;
defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
- f256mem, 0>, XD, VEX_4V;
+ f256mem, 0>, TB, XD, VEX_4V;
defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
- f256mem, 0>, OpSize, VEX_4V;
+ f256mem, 0>, TB, OpSize, VEX_4V;
}
let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
ExeDomain = SSEPackedDouble in {
defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
- f128mem>, XD;
+ f128mem>, TB, XD;
defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
f128mem>, TB, OpSize;
}
[(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
int_x86_sse3_hadd_ps, 0>, VEX_4V;
defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
// SSSE3 - Packed Absolute Instructions
//===---------------------------------------------------------------------===//
+
/// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
- PatFrag mem_frag64, PatFrag mem_frag128,
- Intrinsic IntId64, Intrinsic IntId128> {
- def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR64:$dst, (IntId64 VR64:$src))]>;
-
- def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR64:$dst,
- (IntId64 (bitconvert (mem_frag64 addr:$src))))]>;
-
+ PatFrag mem_frag128, Intrinsic IntId128> {
def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
(bitconvert (mem_frag128 addr:$src))))]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
- defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv8i8, memopv16i8,
- int_x86_ssse3_pabs_b,
+let Predicates = [HasAVX] in {
+ defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
int_x86_ssse3_pabs_b_128>, VEX;
- defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv4i16, memopv8i16,
- int_x86_ssse3_pabs_w,
+ defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
int_x86_ssse3_pabs_w_128>, VEX;
- defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv2i32, memopv4i32,
- int_x86_ssse3_pabs_d,
+ defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
int_x86_ssse3_pabs_d_128>, VEX;
}
-defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv8i8, memopv16i8,
- int_x86_ssse3_pabs_b,
- int_x86_ssse3_pabs_b_128>;
-defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv4i16, memopv8i16,
- int_x86_ssse3_pabs_w,
- int_x86_ssse3_pabs_w_128>;
-defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv2i32, memopv4i32,
- int_x86_ssse3_pabs_d,
- int_x86_ssse3_pabs_d_128>;
+defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
+ int_x86_ssse3_pabs_b_128>;
+defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
+ int_x86_ssse3_pabs_w_128>;
+defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
+ int_x86_ssse3_pabs_d_128>;
//===---------------------------------------------------------------------===//
// SSSE3 - Packed Binary Operator Instructions
/// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
- PatFrag mem_frag64, PatFrag mem_frag128,
- Intrinsic IntId64, Intrinsic IntId128,
+ PatFrag mem_frag128, Intrinsic IntId128,
bit Is2Addr = 1> {
let isCommutable = 1 in
- def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst),
- (ins VR64:$src1, VR64:$src2),
- !if(Is2Addr,
- !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
- !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))]>;
- def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst),
- (ins VR64:$src1, i64mem:$src2),
- !if(Is2Addr,
- !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
- !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set VR64:$dst,
- (IntId64 VR64:$src1,
- (bitconvert (memopv8i8 addr:$src2))))]>;
-
- let isCommutable = 1 in
def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
!if(Is2Addr,
(bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
let isCommutable = 0 in {
- defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv4i16, memopv8i16,
- int_x86_ssse3_phadd_w,
+ defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
- defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv2i32, memopv4i32,
- int_x86_ssse3_phadd_d,
+ defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
- defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv4i16, memopv8i16,
- int_x86_ssse3_phadd_sw,
+ defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
- defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv4i16, memopv8i16,
- int_x86_ssse3_phsub_w,
+ defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
- defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv2i32, memopv4i32,
- int_x86_ssse3_phsub_d,
+ defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
- defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv4i16, memopv8i16,
- int_x86_ssse3_phsub_sw,
+ defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
- defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv8i8, memopv16i8,
- int_x86_ssse3_pmadd_ub_sw,
+ defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
- defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv8i8, memopv16i8,
- int_x86_ssse3_pshuf_b,
+ defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
- defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv8i8, memopv16i8,
- int_x86_ssse3_psign_b,
+ defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
int_x86_ssse3_psign_b_128, 0>, VEX_4V;
- defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv4i16, memopv8i16,
- int_x86_ssse3_psign_w,
+ defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
int_x86_ssse3_psign_w_128, 0>, VEX_4V;
- defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv2i32, memopv4i32,
- int_x86_ssse3_psign_d,
+ defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
int_x86_ssse3_psign_d_128, 0>, VEX_4V;
}
-defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv4i16, memopv8i16,
- int_x86_ssse3_pmul_hr_sw,
+defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
}
// None of these have i8 immediate fields.
let ImmT = NoImm, Constraints = "$src1 = $dst" in {
let isCommutable = 0 in {
- defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv4i16, memopv8i16,
- int_x86_ssse3_phadd_w,
+ defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
int_x86_ssse3_phadd_w_128>;
- defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv2i32, memopv4i32,
- int_x86_ssse3_phadd_d,
+ defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
int_x86_ssse3_phadd_d_128>;
- defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv4i16, memopv8i16,
- int_x86_ssse3_phadd_sw,
+ defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
int_x86_ssse3_phadd_sw_128>;
- defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv4i16, memopv8i16,
- int_x86_ssse3_phsub_w,
+ defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
int_x86_ssse3_phsub_w_128>;
- defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv2i32, memopv4i32,
- int_x86_ssse3_phsub_d,
+ defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
int_x86_ssse3_phsub_d_128>;
- defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv4i16, memopv8i16,
- int_x86_ssse3_phsub_sw,
+ defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
int_x86_ssse3_phsub_sw_128>;
- defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv8i8, memopv16i8,
- int_x86_ssse3_pmadd_ub_sw,
+ defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
int_x86_ssse3_pmadd_ub_sw_128>;
- defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv8i8, memopv16i8,
- int_x86_ssse3_pshuf_b,
+ defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
int_x86_ssse3_pshuf_b_128>;
- defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv8i8, memopv16i8,
- int_x86_ssse3_psign_b,
+ defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
int_x86_ssse3_psign_b_128>;
- defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv4i16, memopv8i16,
- int_x86_ssse3_psign_w,
+ defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
int_x86_ssse3_psign_w_128>;
- defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv2i32, memopv4i32,
- int_x86_ssse3_psign_d,
+ defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
int_x86_ssse3_psign_d_128>;
}
-defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv4i16, memopv8i16,
- int_x86_ssse3_pmul_hr_sw,
+defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
int_x86_ssse3_pmul_hr_sw_128>;
}
def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
(PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
+def : Pat<(X86psignb VR128:$src1, VR128:$src2),
+ (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
+def : Pat<(X86psignw VR128:$src1, VR128:$src2),
+ (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
+def : Pat<(X86psignd VR128:$src1, VR128:$src2),
+ (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
+
//===---------------------------------------------------------------------===//
// SSSE3 - Packed Align Instruction Patterns
//===---------------------------------------------------------------------===//
-multiclass sse3_palign<string asm, bit Is2Addr = 1> {
- def R64rr : SS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
- (ins VR64:$src1, VR64:$src2, i8imm:$src3),
- !if(Is2Addr,
- !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- !strconcat(asm,
- "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- []>;
- def R64rm : SS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
- (ins VR64:$src1, i64mem:$src2, i8imm:$src3),
- !if(Is2Addr,
- !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- !strconcat(asm,
- "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- []>;
-
+multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
!if(Is2Addr,
[]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
- defm VPALIGN : sse3_palign<"vpalignr", 0>, VEX_4V;
-let Constraints = "$src1 = $dst" in
- defm PALIGN : sse3_palign<"palignr">;
-
-let AddedComplexity = 5 in {
-
-def : Pat<(v1i64 (palign:$src3 VR64:$src1, VR64:$src2)),
- (PALIGNR64rr VR64:$src2, VR64:$src1,
- (SHUFFLE_get_palign_imm VR64:$src3))>,
- Requires<[HasSSSE3]>;
-def : Pat<(v2i32 (palign:$src3 VR64:$src1, VR64:$src2)),
- (PALIGNR64rr VR64:$src2, VR64:$src1,
- (SHUFFLE_get_palign_imm VR64:$src3))>,
- Requires<[HasSSSE3]>;
-def : Pat<(v4i16 (palign:$src3 VR64:$src1, VR64:$src2)),
- (PALIGNR64rr VR64:$src2, VR64:$src1,
- (SHUFFLE_get_palign_imm VR64:$src3))>,
- Requires<[HasSSSE3]>;
-def : Pat<(v8i8 (palign:$src3 VR64:$src1, VR64:$src2)),
- (PALIGNR64rr VR64:$src2, VR64:$src1,
- (SHUFFLE_get_palign_imm VR64:$src3))>,
- Requires<[HasSSSE3]>;
-
-def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
- (PALIGNR128rr VR128:$src2, VR128:$src1,
- (SHUFFLE_get_palign_imm VR128:$src3))>,
- Requires<[HasSSSE3]>;
-def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
- (PALIGNR128rr VR128:$src2, VR128:$src1,
- (SHUFFLE_get_palign_imm VR128:$src3))>,
- Requires<[HasSSSE3]>;
-def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
- (PALIGNR128rr VR128:$src2, VR128:$src1,
- (SHUFFLE_get_palign_imm VR128:$src3))>,
- Requires<[HasSSSE3]>;
-def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
- (PALIGNR128rr VR128:$src2, VR128:$src1,
- (SHUFFLE_get_palign_imm VR128:$src3))>,
- Requires<[HasSSSE3]>;
+let Predicates = [HasAVX] in
+ defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
+let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
+ defm PALIGN : ssse3_palign<"palignr">;
+
+let Predicates = [HasSSSE3] in {
+def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
+def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
+def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
+def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
+}
+
+let Predicates = [HasAVX] in {
+def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
+def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
+def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
+def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+ (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
}
//===---------------------------------------------------------------------===//
//===---------------------------------------------------------------------===//
// Thread synchronization
-def MONITOR : I<0x01, MRM_C8, (outs), (ins), "monitor",
- [(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
-def MWAIT : I<0x01, MRM_C9, (outs), (ins), "mwait",
- [(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
+let usesCustomInserter = 1 in {
+def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
+ [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
+def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
+ [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
+}
+
+let Uses = [EAX, ECX, EDX] in
+def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
+ Requires<[HasSSE3]>;
+let Uses = [ECX, EAX] in
+def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
+ Requires<[HasSSE3]>;
+
+def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
+def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
+
+def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
+ Requires<[In32BitMode]>;
+def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
+ Requires<[In64BitMode]>;
//===---------------------------------------------------------------------===//
// Non-Instruction Patterns
def : Pat<(fextend (loadf32 addr:$src)),
(CVTSS2SDrm addr:$src)>;
-// bit_convert
-let Predicates = [HasSSE2] in {
+// Bitcasts between 128-bit vector types. Return the original type since
+// no instruction is needed for the conversion
+let Predicates = [HasXMMInt] in {
def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
}
+// Bitcasts between 256-bit vector types. Return the original type since
+// no instruction is needed for the conversion
+let Predicates = [HasAVX] in {
+ def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
+ def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
+ def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
+ def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
+ def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
+ def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
+ def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
+ def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
+ def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
+ def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
+ def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
+ def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
+ def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
+ def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
+ def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
+ def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
+ def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
+ def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
+ def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
+ def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
+ def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
+ def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
+ def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
+ def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
+ def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
+ def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
+ def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
+ def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
+ def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
+ def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
+}
+
// Move scalar to XMM zero-extended
// movd to XMM register zero-extends
let AddedComplexity = 15 in {
let AddedComplexity = 10 in {
def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
(UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
-def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
- (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
(PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
-def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
- (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
-}
-
-// Special unary SHUFPSrri case.
-def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
- (SHUFPSrri VR128:$src1, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>;
-let AddedComplexity = 5 in
-def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
- (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
- Requires<[HasSSE2]>;
-// Special unary SHUFPDrri case.
-def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
- (SHUFPDrri VR128:$src1, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-// Special unary SHUFPDrri case.
-def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
- (SHUFPDrri VR128:$src1, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-// Unary v4f32 shuffle with PSHUF* in order to fold a load.
-def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
- (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
- Requires<[HasSSE2]>;
-
-// Special binary v4i32 shuffle cases with SHUFPS.
-def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
- (SHUFPSrri VR128:$src1, VR128:$src2,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
- (SHUFPSrmi VR128:$src1, addr:$src2,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-// Special binary v2i64 shuffle cases using SHUFPDrri.
-def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
- (SHUFPDrri VR128:$src1, VR128:$src2,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE2]>;
-
-// vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
-let AddedComplexity = 15 in {
-def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
- (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
- Requires<[OptForSpeed, HasSSE2]>;
-def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
- (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
- Requires<[OptForSpeed, HasSSE2]>;
-}
-let AddedComplexity = 10 in {
-def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
- (UNPCKLPSrr VR128:$src, VR128:$src)>;
-def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
- (PUNPCKLBWrr VR128:$src, VR128:$src)>;
-def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
- (PUNPCKLWDrr VR128:$src, VR128:$src)>;
-def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
- (PUNPCKLDQrr VR128:$src, VR128:$src)>;
-}
-
-// vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
-let AddedComplexity = 15 in {
-def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
- (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
- Requires<[OptForSpeed, HasSSE2]>;
-def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
- (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
- Requires<[OptForSpeed, HasSSE2]>;
-}
-let AddedComplexity = 10 in {
-def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
- (UNPCKHPSrr VR128:$src, VR128:$src)>;
-def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
- (PUNPCKHBWrr VR128:$src, VR128:$src)>;
-def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
- (PUNPCKHWDrr VR128:$src, VR128:$src)>;
-def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
- (PUNPCKHDQrr VR128:$src, VR128:$src)>;
-}
-
-let AddedComplexity = 20 in {
-// vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
-def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
- (MOVLHPSrr VR128:$src1, VR128:$src2)>;
-
-// vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
-def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
- (MOVHLPSrr VR128:$src1, VR128:$src2)>;
-
-// vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
-def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
- (MOVHLPSrr VR128:$src1, VR128:$src1)>;
-def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
- (MOVHLPSrr VR128:$src1, VR128:$src1)>;
}
let AddedComplexity = 20 in {
Requires<[HasSSE2]>;
}
-// vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
-// fall back to this for SSE1)
-def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
- (SHUFPSrri VR128:$src2, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>;
-
// Set lowest element and zero upper elements.
def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
(MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
-// Some special case pandn patterns.
-def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
- VR128:$src2)),
- (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
-def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
- VR128:$src2)),
- (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
-def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
- VR128:$src2)),
- (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
-
-def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
- (memop addr:$src2))),
- (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
-def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
- (memop addr:$src2))),
- (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
-def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
- (memop addr:$src2))),
- (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
-
-// vector -> vector casts
-def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
- (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
-def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
- (Int_CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
-def : Pat<(v2f64 (sint_to_fp (v2i32 VR64:$src))),
- (Int_CVTPI2PDrr VR64:$src)>, Requires<[HasSSE2]>;
-def : Pat<(v2i32 (fp_to_sint (v2f64 VR128:$src))),
- (Int_CVTTPD2PIrr VR128:$src)>, Requires<[HasSSE2]>;
-
// Use movaps / movups for SSE integer load / store (one byte shorter).
-def : Pat<(alignedloadv4i32 addr:$src),
- (MOVAPSrm addr:$src)>;
-def : Pat<(loadv4i32 addr:$src),
- (MOVUPSrm addr:$src)>;
-def : Pat<(alignedloadv2i64 addr:$src),
- (MOVAPSrm addr:$src)>;
-def : Pat<(loadv2i64 addr:$src),
- (MOVUPSrm addr:$src)>;
-
-def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
- (MOVAPSmr addr:$dst, VR128:$src)>;
-def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
- (MOVAPSmr addr:$dst, VR128:$src)>;
-def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
- (MOVAPSmr addr:$dst, VR128:$src)>;
-def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
- (MOVAPSmr addr:$dst, VR128:$src)>;
-def : Pat<(store (v2i64 VR128:$src), addr:$dst),
- (MOVUPSmr addr:$dst, VR128:$src)>;
-def : Pat<(store (v4i32 VR128:$src), addr:$dst),
- (MOVUPSmr addr:$dst, VR128:$src)>;
-def : Pat<(store (v8i16 VR128:$src), addr:$dst),
- (MOVUPSmr addr:$dst, VR128:$src)>;
-def : Pat<(store (v16i8 VR128:$src), addr:$dst),
- (MOVUPSmr addr:$dst, VR128:$src)>;
+// The instructions selected below are then converted to MOVDQA/MOVDQU
+// during the SSE domain pass.
+let Predicates = [HasSSE1] in {
+ def : Pat<(alignedloadv4i32 addr:$src),
+ (MOVAPSrm addr:$src)>;
+ def : Pat<(loadv4i32 addr:$src),
+ (MOVUPSrm addr:$src)>;
+ def : Pat<(alignedloadv2i64 addr:$src),
+ (MOVAPSrm addr:$src)>;
+ def : Pat<(loadv2i64 addr:$src),
+ (MOVUPSrm addr:$src)>;
+
+ def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
+ (MOVAPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
+ (MOVAPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
+ (MOVAPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
+ (MOVAPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(store (v2i64 VR128:$src), addr:$dst),
+ (MOVUPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(store (v4i32 VR128:$src), addr:$dst),
+ (MOVUPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(store (v8i16 VR128:$src), addr:$dst),
+ (MOVUPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(store (v16i8 VR128:$src), addr:$dst),
+ (MOVUPSmr addr:$dst, VR128:$src)>;
+}
+
+// Use vmovaps/vmovups for AVX integer load/store.
+let Predicates = [HasAVX] in {
+ // 128-bit load/store
+ def : Pat<(alignedloadv4i32 addr:$src),
+ (VMOVAPSrm addr:$src)>;
+ def : Pat<(loadv4i32 addr:$src),
+ (VMOVUPSrm addr:$src)>;
+ def : Pat<(alignedloadv2i64 addr:$src),
+ (VMOVAPSrm addr:$src)>;
+ def : Pat<(loadv2i64 addr:$src),
+ (VMOVUPSrm addr:$src)>;
+
+ def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
+ (VMOVAPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
+ (VMOVAPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
+ (VMOVAPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
+ (VMOVAPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(store (v2i64 VR128:$src), addr:$dst),
+ (VMOVUPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(store (v4i32 VR128:$src), addr:$dst),
+ (VMOVUPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(store (v8i16 VR128:$src), addr:$dst),
+ (VMOVUPSmr addr:$dst, VR128:$src)>;
+ def : Pat<(store (v16i8 VR128:$src), addr:$dst),
+ (VMOVUPSmr addr:$dst, VR128:$src)>;
+
+ // 256-bit load/store
+ def : Pat<(alignedloadv4i64 addr:$src),
+ (VMOVAPSYrm addr:$src)>;
+ def : Pat<(loadv4i64 addr:$src),
+ (VMOVUPSYrm addr:$src)>;
+ def : Pat<(alignedloadv8i32 addr:$src),
+ (VMOVAPSYrm addr:$src)>;
+ def : Pat<(loadv8i32 addr:$src),
+ (VMOVUPSYrm addr:$src)>;
+ def : Pat<(alignedstore (v4i64 VR256:$src), addr:$dst),
+ (VMOVAPSYmr addr:$dst, VR256:$src)>;
+ def : Pat<(alignedstore (v8i32 VR256:$src), addr:$dst),
+ (VMOVAPSYmr addr:$dst, VR256:$src)>;
+ def : Pat<(alignedstore (v16i16 VR256:$src), addr:$dst),
+ (VMOVAPSYmr addr:$dst, VR256:$src)>;
+ def : Pat<(alignedstore (v32i8 VR256:$src), addr:$dst),
+ (VMOVAPSYmr addr:$dst, VR256:$src)>;
+ def : Pat<(store (v4i64 VR256:$src), addr:$dst),
+ (VMOVUPSYmr addr:$dst, VR256:$src)>;
+ def : Pat<(store (v8i32 VR256:$src), addr:$dst),
+ (VMOVUPSYmr addr:$dst, VR256:$src)>;
+ def : Pat<(store (v16i16 VR256:$src), addr:$dst),
+ (VMOVUPSYmr addr:$dst, VR256:$src)>;
+ def : Pat<(store (v32i8 VR256:$src), addr:$dst),
+ (VMOVUPSYmr addr:$dst, VR256:$src)>;
+}
//===----------------------------------------------------------------------===//
// SSE4.1 - Packed Move with Sign/Zero Extend
OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
VEX;
defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
VEX;
defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
VEX;
defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
// (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
(ins VR128:$src1, i32i8imm:$src2),
// (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
addr:$dst)]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
addr:$dst)]>, OpSize, REX_W;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
addr:$dst)]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
(ins VR128:$src1, i32i8imm:$src2),
imm:$src3))]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
imm:$src3)))]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
imm:$src3)))]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
let Constraints = "$src1 = $dst" in
defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
// in the target vector.
multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
(X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
OpSize;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
+ (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
let Constraints = "$src1 = $dst" in
defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
Intrinsic V4F32Int, Intrinsic V2F64Int> {
// Intrinsic operation, reg.
// Vector intrinsic operation, reg
- def PSr_Int : SS4AIi8<opcps, MRMSrcReg,
+ def PSr : SS4AIi8<opcps, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
OpSize;
// Vector intrinsic operation, mem
- def PSm_Int : Ii8<opcps, MRMSrcMem,
+ def PSm : Ii8<opcps, MRMSrcMem,
(outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
Requires<[HasSSE41]>;
// Vector intrinsic operation, reg
- def PDr_Int : SS4AIi8<opcpd, MRMSrcReg,
+ def PDr : SS4AIi8<opcpd, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
OpSize;
// Vector intrinsic operation, mem
- def PDm_Int : SS4AIi8<opcpd, MRMSrcMem,
+ def PDm : SS4AIi8<opcpd, MRMSrcMem,
(outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
// Intrinsic operation, reg.
// Vector intrinsic operation, reg
- def PSr : SS4AIi8<opcps, MRMSrcReg,
+ def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[]>, OpSize;
// Vector intrinsic operation, mem
- def PSm : Ii8<opcps, MRMSrcMem,
+ def PSm_AVX : Ii8<opcps, MRMSrcMem,
(outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[]>, TA, OpSize, Requires<[HasSSE41]>;
// Vector intrinsic operation, reg
- def PDr : SS4AIi8<opcpd, MRMSrcReg,
+ def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[]>, OpSize;
// Vector intrinsic operation, mem
- def PDm : SS4AIi8<opcpd, MRMSrcMem,
+ def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
(outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
Intrinsic F32Int,
Intrinsic F64Int, bit Is2Addr = 1> {
// Intrinsic operation, reg.
- def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
+ def SSr : SS4AIi8<opcss, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
OpSize;
// Intrinsic operation, mem.
- def SSm_Int : SS4AIi8<opcss, MRMSrcMem,
+ def SSm : SS4AIi8<opcss, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
OpSize;
// Intrinsic operation, reg.
- def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
+ def SDr : SS4AIi8<opcsd, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
OpSize;
// Intrinsic operation, mem.
- def SDm_Int : SS4AIi8<opcsd, MRMSrcMem,
+ def SDm : SS4AIi8<opcsd, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
string OpcodeStr> {
// Intrinsic operation, reg.
- def SSr : SS4AIi8<opcss, MRMSrcReg,
+ def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[]>, OpSize;
// Intrinsic operation, mem.
- def SSm : SS4AIi8<opcss, MRMSrcMem,
+ def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[]>, OpSize;
// Intrinsic operation, reg.
- def SDr : SS4AIi8<opcsd, MRMSrcReg,
+ def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[]>, OpSize;
// Intrinsic operation, mem.
- def SDm : SS4AIi8<opcsd, MRMSrcMem,
+ def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
}
// FP round - roundss, roundps, roundsd, roundpd
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
// Intrinsic form
defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
memopv4f32, memopv2f64,
// ptest instruction we'll lower to this in X86ISelLowering primarily from
// the intel intrinsic that corresponds to this.
-let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Defs = [EFLAGS], Predicates = [HasAVX] in {
def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
- "vptest\t{$src2, $src1|$src1, $src2}",
- [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
- OpSize, VEX;
-def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
- "vptest\t{$src2, $src1|$src1, $src2}", []>, OpSize, VEX;
+ "vptest\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
+ OpSize, VEX;
+def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
+ "vptest\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
+ OpSize, VEX;
-def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
- "vptest\t{$src2, $src1|$src1, $src2}",
- [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
- OpSize, VEX;
+def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
+ "vptest\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
+ OpSize, VEX;
def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
- "vptest\t{$src2, $src1|$src1, $src2}", []>, OpSize, VEX;
+ "vptest\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
+ OpSize, VEX;
}
let Defs = [EFLAGS] in {
def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
- "ptest \t{$src2, $src1|$src1, $src2}",
- [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
+ "ptest \t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
OpSize;
-def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
- "ptest \t{$src2, $src1|$src1, $src2}",
- [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
+def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
+ "ptest \t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
OpSize;
}
// The bit test instructions below are AVX only
multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
- X86MemOperand x86memop> {
- def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- []>, OpSize, VEX;
- def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- []>, OpSize, VEX;
+ X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
+ def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
+ !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
+ [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
+ def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
+ !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
+ [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
+ OpSize, VEX;
}
-let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
- defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem>;
- defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem>;
- defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem>;
- defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem>;
+let Defs = [EFLAGS], Predicates = [HasAVX] in {
+defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
+defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
+defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
+defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
}
//===----------------------------------------------------------------------===//
// SSE4.1 - Misc Instructions
//===----------------------------------------------------------------------===//
+def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
+ "popcnt{w}\t{$src, $dst|$dst, $src}",
+ [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
+def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
+ "popcnt{w}\t{$src, $dst|$dst, $src}",
+ [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
+
+def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
+ "popcnt{l}\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
+def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
+ "popcnt{l}\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
+
+def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
+ "popcnt{q}\t{$src, $dst|$dst, $src}",
+ [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
+def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
+ "popcnt{q}\t{$src, $dst|$dst, $src}",
+ [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
+
+
+
// SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
Intrinsic IntId128> {
(bitconvert (memopv8i16 addr:$src))))]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
int_x86_sse41_phminposuw>, VEX;
defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
(bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
let isCommutable = 0 in
defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
0>, VEX_4V;
0>, VEX_4V;
defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
0>, VEX_4V;
+
+ def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
+ (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
+ (VPCMPEQQrm VR128:$src1, addr:$src2)>;
}
let Constraints = "$src1 = $dst" in {
OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in
defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
X86MemOperand x86memop, bit Is2Addr = 1> {
let isCommutable = 1 in
def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2, i32i8imm:$src3),
+ (ins RC:$src1, RC:$src2, u32u8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
[(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
OpSize;
def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
+ (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
let isCommutable = 0 in {
defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
VR128, memopv16i8, i128mem, 0>, VEX_4V;
defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
VR128, memopv16i8, i128mem, 0>, VEX_4V;
- let Pattern = []<dag> in {
- defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
- VR256, memopv32i8, i256mem, 0>, VEX_4V;
- defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
- VR256, memopv32i8, i256mem, 0>, VEX_4V;
- }
+ defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
+ int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
+ defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
+ int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
VR128, memopv16i8, i128mem, 0>, VEX_4V;
defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
VR128, memopv16i8, i128mem, 0>, VEX_4V;
defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
VR128, memopv16i8, i128mem, 0>, VEX_4V;
- let Pattern = []<dag> in
- defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
- VR256, memopv32i8, i256mem, 0>, VEX_4V;
+ defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
+ VR256, memopv32i8, i256mem, 0>, VEX_4V;
}
let Constraints = "$src1 = $dst" in {
}
/// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
RegisterClass RC, X86MemOperand x86memop,
PatFrag mem_frag, Intrinsic IntId> {
memopv16i8, int_x86_sse41_blendvps>;
defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
memopv16i8, int_x86_sse41_pblendvb>;
-
-let Pattern = []<dag> in { // FIXME: implement 256 intrinsics here.
defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
- memopv32i8, int_x86_sse41_blendvpd>;
+ memopv32i8, int_x86_avx_blendv_pd_256>;
defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
- memopv32i8, int_x86_sse41_blendvps>;
-}
+ memopv32i8, int_x86_avx_blendv_ps_256>;
/// SS41I_ternary_int - SSE 4.1 ternary operator
let Uses = [XMM0], Constraints = "$src1 = $dst" in {
def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
!strconcat(OpcodeStr,
- "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
+ "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
OpSize;
def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr,
- "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
+ "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst,
(IntId VR128:$src1,
(bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
+ (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
+
+let Predicates = [HasAVX] in
def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"vmovntdqa\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
(bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in
+let Predicates = [HasAVX] in {
defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
0>, VEX_4V;
+
+ def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
+ (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
+ def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
+ (VPCMPGTQrm VR128:$src1, addr:$src2)>;
+}
+
let Constraints = "$src1 = $dst" in
defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
// Packed Compare Implicit Length Strings, Return Mask
multiclass pseudo_pcmpistrm<string asm> {
- def REG : Ii8<0, Pseudo, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3), !strconcat(asm, "rr PSEUDO"),
+ def REG : PseudoI<(outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
[(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
imm:$src3))]>;
- def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3), !strconcat(asm, "rm PSEUDO"),
+ def MEM : PseudoI<(outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
[(set VR128:$dst, (int_x86_sse42_pcmpistrm128
VR128:$src1, (load addr:$src2), imm:$src3))]>;
}
defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
}
-let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
- Predicates = [HasAVX] in {
+let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
// Packed Compare Explicit Length Strings, Return Mask
multiclass pseudo_pcmpestrm<string asm> {
- def REG : Ii8<0, Pseudo, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5), !strconcat(asm, "rr PSEUDO"),
+ def REG : PseudoI<(outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
[(set VR128:$dst, (int_x86_sse42_pcmpestrm128
VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
- def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5), !strconcat(asm, "rm PSEUDO"),
+ def MEM : PseudoI<(outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
[(set VR128:$dst, (int_x86_sse42_pcmpestrm128
VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
}
defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
}
-let isAsmParserOnly = 1, Predicates = [HasAVX],
+let Predicates = [HasAVX],
Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src3, i8imm:$src5),
}
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
VEX;
defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
}
}
-let isAsmParserOnly = 1, Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in {
defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
VEX;
defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
// This set of instructions are only rm, the only difference is the size
// of r and m.
let Constraints = "$src1 = $dst" in {
- def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
+ def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
(ins GR32:$src1, i8mem:$src2),
"crc32{b} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
- (int_x86_sse42_crc32_8 GR32:$src1,
+ (int_x86_sse42_crc32_32_8 GR32:$src1,
(load addr:$src2)))]>;
- def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
+ def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
(ins GR32:$src1, GR8:$src2),
"crc32{b} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
- (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
- def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
+ (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
+ def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
(ins GR32:$src1, i16mem:$src2),
"crc32{w} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
- (int_x86_sse42_crc32_16 GR32:$src1,
+ (int_x86_sse42_crc32_32_16 GR32:$src1,
(load addr:$src2)))]>,
OpSize;
- def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
+ def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
(ins GR32:$src1, GR16:$src2),
"crc32{w} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
- (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
+ (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
OpSize;
- def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
+ def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
(ins GR32:$src1, i32mem:$src2),
"crc32{l} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
- (int_x86_sse42_crc32_32 GR32:$src1,
+ (int_x86_sse42_crc32_32_32 GR32:$src1,
(load addr:$src2)))]>;
- def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
+ def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
(ins GR32:$src1, GR32:$src2),
"crc32{l} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
- (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
- def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
+ (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
+ def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
(ins GR64:$src1, i8mem:$src2),
"crc32{b} \t{$src2, $src1|$src1, $src2}",
[(set GR64:$dst,
- (int_x86_sse42_crc64_8 GR64:$src1,
+ (int_x86_sse42_crc32_64_8 GR64:$src1,
(load addr:$src2)))]>,
REX_W;
- def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
+ def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
(ins GR64:$src1, GR8:$src2),
"crc32{b} \t{$src2, $src1|$src1, $src2}",
[(set GR64:$dst,
- (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
+ (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
REX_W;
- def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
+ def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
(ins GR64:$src1, i64mem:$src2),
"crc32{q} \t{$src2, $src1|$src1, $src2}",
[(set GR64:$dst,
- (int_x86_sse42_crc64_64 GR64:$src1,
+ (int_x86_sse42_crc32_64_64 GR64:$src1,
(load addr:$src2)))]>,
REX_W;
- def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
+ def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
(ins GR64:$src1, GR64:$src2),
"crc32{q} \t{$src2, $src1|$src1, $src2}",
[(set GR64:$dst,
- (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
+ (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
REX_W;
}
}
// Perform One Round of an AES Encryption/Decryption Flow
-let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
+let Predicates = [HasAVX, HasAES] in {
defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
int_x86_aesni_aesenc, 0>, VEX_4V;
defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
(AESDECLASTrm VR128:$src1, addr:$src2)>;
// Perform the AES InvMixColumn Transformation
-let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
+let Predicates = [HasAVX, HasAES] in {
def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1),
"vaesimc\t{$src1, $dst|$dst, $src1}",
OpSize;
// AES Round Key Generation Assist
-let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
+let Predicates = [HasAVX, HasAES] in {
def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, i8imm:$src2),
"vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
// CLMUL Instructions
//===----------------------------------------------------------------------===//
-// Only the AVX version of CLMUL instructions are described here.
-
// Carry-less Multiplication instructions
-let isAsmParserOnly = 1 in {
-def VPCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
+let Constraints = "$src1 = $dst" in {
+def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
+ []>;
+
+def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
+ []>;
+}
+
+// AVX carry-less Multiplication instructions
+def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>;
-def VPCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
+def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>;
-// Assembler Only
-multiclass avx_vpclmul<string asm> {
- def rr : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>;
- def rm : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>;
-}
-defm VPCLMULHQHQDQ : avx_vpclmul<"vpclmulhqhqdq">;
-defm VPCLMULHQLQDQ : avx_vpclmul<"vpclmulhqlqdq">;
-defm VPCLMULLQHQDQ : avx_vpclmul<"vpclmullqhqdq">;
-defm VPCLMULLQLQDQ : avx_vpclmul<"vpclmullqlqdq">;
+multiclass pclmul_alias<string asm, int immop> {
+ def : InstAlias<!strconcat("pclmul", asm,
+ "dq {$src, $dst|$dst, $src}"),
+ (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
+
+ def : InstAlias<!strconcat("pclmul", asm,
+ "dq {$src, $dst|$dst, $src}"),
+ (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
-} // isAsmParserOnly
+ def : InstAlias<!strconcat("vpclmul", asm,
+ "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
+ (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
+
+ def : InstAlias<!strconcat("vpclmul", asm,
+ "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
+ (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
+}
+defm : pclmul_alias<"hqhq", 0x11>;
+defm : pclmul_alias<"hqlq", 0x01>;
+defm : pclmul_alias<"lqhq", 0x10>;
+defm : pclmul_alias<"lqlq", 0x00>;
//===----------------------------------------------------------------------===//
// AVX Instructions
//===----------------------------------------------------------------------===//
-let isAsmParserOnly = 1 in {
-
-// Load from memory and broadcast to all elements of the destination operand
+//===----------------------------------------------------------------------===//
+// VBROADCAST - Load from memory and broadcast to all elements of the
+// destination operand
+//
class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
X86MemOperand x86memop, Intrinsic Int> :
AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
int_x86_avx_vbroadcastf128_pd_256>;
-// Insert packed floating-point values
+def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
+ (VBROADCASTF128 addr:$src)>;
+
+def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
+ (VBROADCASTSSY addr:$src)>;
+def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
+ (VBROADCASTSD addr:$src)>;
+def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
+ (VBROADCASTSSY addr:$src)>;
+def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
+ (VBROADCASTSD addr:$src)>;
+
+def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
+ (VBROADCASTSS addr:$src)>;
+def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
+ (VBROADCASTSS addr:$src)>;
+
+//===----------------------------------------------------------------------===//
+// VINSERTF128 - Insert packed floating-point values
+//
def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR128:$src2, i8imm:$src3),
"vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
"vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, VEX_4V;
-// Extract packed floating-point values
+def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
+ (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
+def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
+ (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
+def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
+ (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
+
+def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+
+// Special COPY patterns
+def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
+ (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
+def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
+ (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
+def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
+ (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
+def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
+ (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
+def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
+ (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
+def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
+ (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
+
+//===----------------------------------------------------------------------===//
+// VEXTRACTF128 - Extract packed floating-point values
+//
def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
(ins VR256:$src1, i8imm:$src2),
"vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[]>, VEX;
-// Conditional SIMD Packed Loads and Stores
+def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
+ (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
+def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
+ (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
+def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
+ (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
+
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v4f32 (VEXTRACTF128rr
+ (v8f32 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v2f64 (VEXTRACTF128rr
+ (v4f64 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v4i32 (VEXTRACTF128rr
+ (v8i32 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v2i64 (VEXTRACTF128rr
+ (v4i64 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v8i16 (VEXTRACTF128rr
+ (v16i16 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v16i8 (VEXTRACTF128rr
+ (v32i8 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+
+// Special COPY patterns
+def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
+ (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
+def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
+ (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
+
+def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
+ (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
+def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
+ (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
+
+def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
+ (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
+def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
+ (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
+
+
+//===----------------------------------------------------------------------===//
+// VMASKMOV - Conditional SIMD Packed Loads and Stores
+//
multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
Intrinsic IntLd, Intrinsic IntLd256,
Intrinsic IntSt, Intrinsic IntSt256,
int_x86_avx_maskstore_pd_256,
memopv2f64, memopv4f64>;
-// Permute Floating-Point Values
+//===----------------------------------------------------------------------===//
+// VPERMIL - Permute Single and Double Floating-Point Values
+//
multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
RegisterClass RC, X86MemOperand x86memop_f,
X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
int_x86_avx_vpermilvar_pd_256,
int_x86_avx_vpermil_pd_256>;
+def : Pat<(v8f32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
+ (VPERMILPSYri VR256:$src1, imm:$imm)>;
+def : Pat<(v4f64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
+ (VPERMILPDYri VR256:$src1, imm:$imm)>;
+def : Pat<(v8i32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
+ (VPERMILPSYri VR256:$src1, imm:$imm)>;
+def : Pat<(v4i64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
+ (VPERMILPDYri VR256:$src1, imm:$imm)>;
+
+//===----------------------------------------------------------------------===//
+// VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
+//
def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2, i8imm:$src3),
"vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
"vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[]>, VEX_4V;
-// Zero All YMM registers
-def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
- [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
-
-// Zero Upper bits of YMM registers
-def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
- [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
-
-} // isAsmParserOnly
-
-def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
- (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
-def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
- (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
-def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
- (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
-
-def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
- (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
-def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
- (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
-def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
- (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
-
-def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
- (VBROADCASTF128 addr:$src)>;
-
def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
(VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
+def : Pat<(v8f32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v8i32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v4i64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v4f64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v32i8 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v16i16 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+
+//===----------------------------------------------------------------------===//
+// VZERO - Zero YMM registers
+//
+let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
+ YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
+ // Zero All YMM registers
+ def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
+ [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
+
+}
+
+// Zero Upper bits of YMM registers
+def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
+ [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
+
+//===----------------------------------------------------------------------===//
+// SSE Shuffle pattern fragments
+//===----------------------------------------------------------------------===//
+
+// This is part of a "work in progress" refactoring. The idea is that all
+// vector shuffles are going to be translated into target specific nodes and
+// directly matched by the patterns below (which can be changed along the way)
+// The AVX version of some but not all of them are described here, and more
+// should come in a near future.
+
+
+
+// Shuffle with UNPCKLPS
+def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
+ (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
+ (UNPCKLPSrm VR128:$src1, addr:$src2)>;
+
+def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
+ (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
+ (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
+
+// Shuffle with VUNPCKHPSY
+def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
+ (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
+ (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
+ (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, (memopv8i32 addr:$src2))),
+ (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
+
+// Shuffle with UNPCKHPS
+def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
+ (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
+ (UNPCKHPSrm VR128:$src1, addr:$src2)>;
+
+def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
+ (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
+ (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
+
+// Shuffle with VUNPCKHPSY
+def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, (memopv8f32 addr:$src2))),
+ (VUNPCKHPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
+ (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
+
+def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, (memopv8i32 addr:$src2))),
+ (VUNPCKHPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
+ (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
+
+// Shuffle with UNPCKLPD
+def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
+ (VUNPCKLPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
+ (UNPCKLPDrm VR128:$src1, addr:$src2)>;
+
+def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
+ (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
+ (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
+
+// Shuffle with VUNPCKLPDY
+def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
+ (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
+ (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
+
+def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, (memopv4i64 addr:$src2))),
+ (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
+ (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
+
+// Shuffle with UNPCKHPD
+def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
+ (VUNPCKHPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
+ (UNPCKHPDrm VR128:$src1, addr:$src2)>;
+
+def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
+ (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
+ (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
+
+// Shuffle with VUNPCKHPDY
+def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, (memopv4f64 addr:$src2))),
+ (VUNPCKHPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
+ (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, (memopv4i64 addr:$src2))),
+ (VUNPCKHPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
+def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
+ (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
+
+// FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
+// is during lowering, where it's not possible to recognize the load fold cause
+// it has two uses through a bitcast. One use disappears at isel time and the
+// fold opportunity reappears.
+def : Pat<(v2f64 (X86Movddup VR128:$src)),
+ (UNPCKLPDrr VR128:$src, VR128:$src)>;
+
+// Shuffle with MOVLHPD
+def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
+ (scalar_to_vector (loadf64 addr:$src2)))),
+ (MOVHPDrm VR128:$src1, addr:$src2)>;
+
+// FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
+// is during lowering, where it's not possible to recognize the load fold cause
+// it has two uses through a bitcast. One use disappears at isel time and the
+// fold opportunity reappears.
+def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
+ (scalar_to_vector (loadf64 addr:$src2)))),
+ (MOVHPDrm VR128:$src1, addr:$src2)>;
+
+// Shuffle with MOVSS
+def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
+ (MOVSSrr VR128:$src1, FR32:$src2)>;
+def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
+ (MOVSSrr (v4i32 VR128:$src1),
+ (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
+def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
+ (MOVSSrr (v4f32 VR128:$src1),
+ (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
+
+// Shuffle with MOVSD
+def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
+ (MOVSDrr VR128:$src1, FR64:$src2)>;
+def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
+ (MOVSDrr (v2i64 VR128:$src1),
+ (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
+def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
+ (MOVSDrr (v2f64 VR128:$src1),
+ (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
+def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
+ (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
+def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
+ (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
+
+// Shuffle with MOVLPS
+def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
+ (MOVLPSrm VR128:$src1, addr:$src2)>;
+def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
+ (MOVLPSrm VR128:$src1, addr:$src2)>;
+def : Pat<(X86Movlps VR128:$src1,
+ (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
+ (MOVLPSrm VR128:$src1, addr:$src2)>;
+// FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
+// is during lowering, where it's not possible to recognize the load fold cause
+// it has two uses through a bitcast. One use disappears at isel time and the
+// fold opportunity reappears.
+def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
+ (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
+
+def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
+ (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
+
+// Shuffle with MOVLPD
+def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
+ (MOVLPDrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
+ (MOVLPDrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2f64 (X86Movlpd VR128:$src1,
+ (scalar_to_vector (loadf64 addr:$src2)))),
+ (MOVLPDrm VR128:$src1, addr:$src2)>;
+
+// Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
+def : Pat<(store (f64 (vector_extract
+ (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
+ (MOVHPSmr addr:$dst, VR128:$src)>;
+def : Pat<(store (f64 (vector_extract
+ (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
+ (MOVHPDmr addr:$dst, VR128:$src)>;
+
+def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
+ (MOVLPSmr addr:$src1, VR128:$src2)>;
+def : Pat<(store (v4i32 (X86Movlps
+ (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
+ (MOVLPSmr addr:$src1, VR128:$src2)>;
+
+def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
+ (MOVLPDmr addr:$src1, VR128:$src2)>;
+def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
+ (MOVLPDmr addr:$src1, VR128:$src2)>;