// A 128-bit subvector insert to the first 256-bit vector position
// is a subregister copy that needs no instruction.
+let AddedComplexity = 25 in { // to give priority over vinsertf128rm
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:$src), (i32 0)),
(INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
(INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
+}
// Implicitly promote a 32-bit scalar to a vector.
def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
}
-// Alias instructions that map fld0 to pxor for sse.
+// Alias instructions that map fld0 to xorps for sse or vxorps for avx.
// This is expanded by ExpandPostRAPseudos.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
isPseudo = 1 in {
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-zeros value if folding it would be beneficial.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
- isPseudo = 1, neverHasSideEffects = 1 in {
-def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
+ isPseudo = 1 in {
+def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4f32 immAllZerosV))]>;
}
-def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
-// The same as done above but for AVX. The 256-bit ISA does not support PI,
+// The same as done above but for AVX. The 256-bit AVX1 ISA doesn't support PI,
// 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 in {
-let Predicates = [HasAVX] in {
-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 Predicates = [HasAVX2], neverHasSideEffects = 1 in
-def AVX2_SET0 : PDI<0xef, MRMInitReg, (outs VR256:$dst), (ins), "",
- []>, VEX_4V;
+ isPseudo = 1, Predicates = [HasAVX] in {
+def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
+ [(set VR256:$dst, (v8f32 immAllZerosV))]>;
}
-let Predicates = [HasAVX2], AddedComplexity = 5 in {
- def : Pat<(v4i64 immAllZerosV), (AVX2_SET0)>;
- def : Pat<(v8i32 immAllZerosV), (AVX2_SET0)>;
- def : Pat<(v16i16 immAllZerosV), (AVX2_SET0)>;
- def : Pat<(v32i8 immAllZerosV), (AVX2_SET0)>;
+let Predicates = [HasAVX] in
+ def : Pat<(v4f64 immAllZerosV), (AVX_SET0)>;
+
+let Predicates = [HasAVX2] in {
+ def : Pat<(v4i64 immAllZerosV), (AVX_SET0)>;
+ def : Pat<(v8i32 immAllZerosV), (AVX_SET0)>;
+ def : Pat<(v16i16 immAllZerosV), (AVX_SET0)>;
+ def : Pat<(v32i8 immAllZerosV), (AVX_SET0)>;
}
-// AVX has no support for 256-bit integer instructions, but since the 128-bit
+// AVX1 has no support for 256-bit integer instructions, but since the 128-bit
// VPXOR instruction writes zero to its upper part, it's safe build zeros.
+let Predicates = [HasAVX1Only] in {
def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
(SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
(SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
+}
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-ones value if folding it would be beneficial.
-// FIXME: Change encoding to pseudo! This is blocked right now by the x86
-// JIT implementation, it does not expand the instructions below like
-// X86MCInstLower does.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
- isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
- let Predicates = [HasAVX] in
- def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
- [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
- def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
- [(set VR128:$dst, (v4i32 immAllOnesV))]>;
+ isPseudo = 1 in {
+ def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4i32 immAllOnesV))]>;
let Predicates = [HasAVX2] in
- def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
- [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
+ def AVX2_SETALLONES : I<0, Pseudo, (outs VR256:$dst), (ins), "",
+ [(set VR256:$dst, (v8i32 immAllOnesV))]>;
}
(VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
}
-let Predicates = [HasSSE1] in {
+let Predicates = [UseSSE1] in {
let AddedComplexity = 15 in {
// Move scalar to XMM zero-extended, zeroing a VR128 then do a
// MOVSS to the lower bits.
(MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
let AddedComplexity = 15 in {
// Move scalar to XMM zero-extended, zeroing a VR128 then do a
// MOVSD to the lower bits.
(VMOVUPDmr addr:$dst, VR128:$src)>;
}
-let Predicates = [HasSSE1] in
+let Predicates = [UseSSE1] in
def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
(MOVUPSmr addr:$dst, VR128:$src)>;
-let Predicates = [HasSSE2] in
+let Predicates = [UseSSE2] in
def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
(MOVUPDmr addr:$dst, VR128:$src)>;
(VMOVUPSYmr addr:$dst, VR256:$src)>;
def : Pat<(store (v32i8 VR256:$src), addr:$dst),
(VMOVUPSYmr addr:$dst, VR256:$src)>;
+
+ // Special patterns for storing subvector extracts of lower 128-bits
+ // Its cheaper to just use VMOVAPS/VMOVUPS instead of VEXTRACTF128mr
+ def : Pat<(alignedstore (v2f64 (extract_subvector
+ (v4f64 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVAPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(alignedstore (v4f32 (extract_subvector
+ (v8f32 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVAPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(alignedstore (v2i64 (extract_subvector
+ (v4i64 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVAPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(alignedstore (v4i32 (extract_subvector
+ (v8i32 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVAPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(alignedstore (v8i16 (extract_subvector
+ (v16i16 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(alignedstore (v16i8 (extract_subvector
+ (v32i8 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVAPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+
+ def : Pat<(store (v2f64 (extract_subvector
+ (v4f64 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVUPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(store (v4f32 (extract_subvector
+ (v8f32 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVUPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(store (v2i64 (extract_subvector
+ (v4i64 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVUPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(store (v4i32 (extract_subvector
+ (v8i32 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(store (v8i16 (extract_subvector
+ (v16i16 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ def : Pat<(store (v16i8 (extract_subvector
+ (v32i8 VR256:$src), (i32 0))), addr:$dst),
+ (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
}
// Use movaps / movups for SSE integer load / store (one byte shorter).
// The instructions selected below are then converted to MOVDQA/MOVDQU
// during the SSE domain pass.
-let Predicates = [HasSSE1] in {
+let Predicates = [UseSSE1] in {
def : Pat<(alignedloadv2i64 addr:$src),
(MOVAPSrm addr:$src)>;
def : Pat<(loadv2i64 addr:$src),
(VMOVLPDmr addr:$src1, VR128:$src2)>;
}
-let Predicates = [HasSSE1] in {
+let Predicates = [UseSSE1] in {
// (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
(iPTR 0))), addr:$src1),
(MOVLPSmr addr:$src1, VR128:$src2)>;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
// Shuffle with MOVLPD
def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
(MOVLPDrm VR128:$src1, addr:$src2)>;
(VMOVHPDrm VR128:$src1, addr:$src2)>;
}
-let Predicates = [HasSSE1] in {
+let Predicates = [UseSSE1] in {
// MOVHPS patterns
def : Pat<(X86Movlhps VR128:$src1,
(bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
(MOVHPSrm VR128:$src1, addr:$src2)>;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
// FIXME: Instead of X86Unpckl, 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
(VMOVHLPSrr VR128:$src1, VR128:$src2)>;
}
-let Predicates = [HasSSE1] in {
+let Predicates = [UseSSE1] in {
// MOVLHPS patterns
def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
(MOVLHPSrr VR128:$src1, VR128:$src2)>;
def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
(VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
-let Predicates = [HasAVX], AddedComplexity = 1 in {
+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:$src))),
defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
"cvtdq2ps\t{$src, $dst|$dst, $src}",
SSEPackedSingle, SSE_CVT_PS>,
- TB, Requires<[HasSSE2]>;
+ TB, Requires<[UseSSE2]>;
/// SSE 2 Only
[(set FR32:$dst, (fround (loadf64 addr:$src)))],
IIC_SSE_CVT_Scalar_RM>,
XD,
- Requires<[HasSSE2, OptForSize]>;
+ Requires<[UseSSE2, OptForSize]>;
def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
- IIC_SSE_CVT_Scalar_RR>, XD, Requires<[HasSSE2]>;
+ IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>;
def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
"cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst, (int_x86_sse2_cvtsd2ss
VR128:$src1, sse_load_f64:$src2))],
- IIC_SSE_CVT_Scalar_RM>, XD, Requires<[HasSSE2]>;
+ IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>;
}
// Convert scalar single to scalar double
XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
}
-let AddedComplexity = 1 in { // give AVX priority
- def : Pat<(f64 (fextend FR32:$src)),
- (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[HasAVX]>;
- def : Pat<(fextend (loadf32 addr:$src)),
- (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX]>;
+def : Pat<(f64 (fextend FR32:$src)),
+ (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[HasAVX]>;
+def : Pat<(fextend (loadf32 addr:$src)),
+ (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX]>;
- def : Pat<(extloadf32 addr:$src),
- (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[HasAVX, OptForSize]>;
- def : Pat<(extloadf32 addr:$src),
- (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
- Requires<[HasAVX, OptForSpeed]>;
-} // AddedComplexity = 1
+def : Pat<(extloadf32 addr:$src),
+ (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+def : Pat<(extloadf32 addr:$src),
+ (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
+ Requires<[HasAVX, OptForSpeed]>;
def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
"cvtss2sd\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (fextend FR32:$src))],
IIC_SSE_CVT_Scalar_RR>, XS,
- Requires<[HasSSE2]>;
+ Requires<[UseSSE2]>;
def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
"cvtss2sd\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (extloadf32 addr:$src))],
IIC_SSE_CVT_Scalar_RM>, XS,
- Requires<[HasSSE2, OptForSize]>;
+ Requires<[UseSSE2, OptForSize]>;
// extload f32 -> f64. This matches load+fextend because we have a hack in
// the isel (PreprocessForFPConvert) that can introduce loads after dag
// Since these loads aren't folded into the fextend, we have to match it
// explicitly here.
def : Pat<(fextend (loadf32 addr:$src)),
- (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
+ (CVTSS2SDrm addr:$src)>, Requires<[UseSSE2]>;
def : Pat<(extloadf32 addr:$src),
- (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
+ (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"cvtss2sd\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
- IIC_SSE_CVT_Scalar_RR>, XS, Requires<[HasSSE2]>;
+ IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>;
def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
"cvtss2sd\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
- IIC_SSE_CVT_Scalar_RM>, XS, Requires<[HasSSE2]>;
+ IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>;
}
// Convert packed single/double fp to doubleword
(VCVTTPS2DQYrm addr:$src)>;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
(CVTDQ2PSrr VR128:$src)>;
def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
IIC_SSE_CVT_PD_RM>, TB, VEX;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
(VCVTPS2PDYrm addr:$src)>;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
// Match fextend for 128 conversions
def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
(CVTPS2PDrr VR128:$src)>;
(VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
}
-let Predicates = [HasSSE1] in {
+let Predicates = [UseSSE1] in {
def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
(CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
(CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
(CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
(VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
}
-let Predicates = [HasSSE1] in {
+let Predicates = [UseSSE1] in {
def : Pat<(v4i32 (X86Shufp VR128:$src1,
(bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
(SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
(SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
// Generic SHUFPD patterns
def : Pat<(v2i64 (X86Shufp VR128:$src1,
(memopv2i64 addr:$src2), (i8 imm:$imm))),
SSEPackedDouble>, TB, OpSize;
} // Constraints = "$src1 = $dst"
-let Predicates = [HasAVX], AddedComplexity = 1 in {
+let Predicates = [HasAVX1Only] in {
+ def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
+ (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
+ (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
+ def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
+ (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
+ (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
+
+ def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
+ (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
+ (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
+ def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
+ (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
+ (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
+}
+
+let Predicates = [HasAVX] in {
// FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
// problem is during lowering, where it's not possible to recognize the load
// fold cause it has two uses through a bitcast. One use disappears at isel
(VUNPCKLPDrr VR128:$src, VR128:$src)>;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
// FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
// problem is during lowering, where it's not possible to recognize the load
// fold cause it has two uses through a bitcast. One use disappears at isel
def : Pat<(i32 (X86fgetsign FR32:$src)),
(MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
- Requires<[HasSSE1]>;
+ Requires<[UseSSE1]>;
def : Pat<(i64 (X86fgetsign FR32:$src)),
(MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
- Requires<[HasSSE1]>;
+ Requires<[UseSSE1]>;
def : Pat<(i32 (X86fgetsign FR64:$src)),
(MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
- Requires<[HasSSE2]>;
+ Requires<[UseSSE2]>;
def : Pat<(i64 (X86fgetsign FR64:$src)),
(MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
- Requires<[HasSSE2]>;
+ Requires<[UseSSE2]>;
//===---------------------------------------------------------------------===//
// SSE2 - Packed Integer Logical Instructions
}
}
-let isCommutable = 1, isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1 in {
defm VMAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S, 0>,
VEX_4V, VEX_LIG;
defm VMAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P, 0>,
def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
!strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
[(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
- Requires<[HasSSE1, OptForSize]>;
+ Requires<[UseSSE1, OptForSize]>;
def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
!strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
[(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
- Requires<[HasSSE2, OptForSize]>;
+ Requires<[UseSSE2, OptForSize]>;
def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
}
/// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
+let hasSideEffects = 0 in
multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
- let neverHasSideEffects = 1 in {
def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
!strconcat(OpcodeStr,
"sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- let mayLoad = 1 in
+ let mayLoad = 1 in {
def SDm : SDI<opc, MRMSrcMem, (outs FR64:$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, $src2}"), []>;
+ }
}
/// sse2_fp_unop_p - SSE2 unops in vector forms.
SSE_RCPP>, VEX;
}
-let AddedComplexity = 1 in {
def : Pat<(f32 (fsqrt FR32:$src)),
(VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
def : Pat<(f32 (fsqrt (load addr:$src))),
def : Pat<(f32 (X86frcp (load addr:$src))),
(VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
Requires<[HasAVX, OptForSize]>;
-}
-let Predicates = [HasAVX], AddedComplexity = 1 in {
+let Predicates = [HasAVX] in {
def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
(COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
(COPY_TO_REGCLASS VR128:$src, FR32)),
IIC_SSE_MOVNT>;
def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
- (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
+ (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
// There is no AVX form for instructions below this point
def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movdqu\t{$src, $dst|$dst, $src}",
- [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
+ [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
// For Disassembler
let isCodeGenOnly = 1 in {
def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
"movdqu\t{$src, $dst|$dst, $src}",
- [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
+ [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
}
let canFoldAsLoad = 1, mayLoad = 1 in {
"movdqu\t{$src, $dst|$dst, $src}",
[/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
IIC_SSE_MOVU_P_RM>,
- XS, Requires<[HasSSE2]>;
+ XS, Requires<[UseSSE2]>;
}
let mayStore = 1 in {
"movdqu\t{$src, $dst|$dst, $src}",
[/*(store (v2i64 VR128:$src), addr:$dst)*/],
IIC_SSE_MOVU_P_MR>,
- XS, Requires<[HasSSE2]>;
+ XS, Requires<[UseSSE2]>;
}
// Intrinsic forms of MOVDQU load and store
"movdqu\t{$src, $dst|$dst, $src}",
[(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
IIC_SSE_MOVU_P_MR>,
- XS, Requires<[HasSSE2]>;
+ XS, Requires<[UseSSE2]>;
} // ExeDomain = SSEPackedInt
(VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
(PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
}
-let Predicates = [HasSSE2] in {
+let Predicates = [UseSSE2] in {
let AddedComplexity = 5 in
defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
}
} // ExeDomain = SSEPackedInt
-// Patterns for using AVX1 instructions with integer vectors
-// Here to give AVX2 priority
-let Predicates = [HasAVX] in {
- def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
- (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
- def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
- (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
- def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
- (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
- def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
- (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
-
- def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
- (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
- def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
- (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
- def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
- (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
- def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
- (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
-}
-
//===---------------------------------------------------------------------===//
// SSE2 - Packed Integer Extract and Insert
//===---------------------------------------------------------------------===//
}
let Constraints = "$src1 = $dst" in
- defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
+ defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
} // ExeDomain = SSEPackedInt
// Move Packed Doubleword Int first element to Doubleword Int
//
def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
- "mov{d|q}\t{$src, $dst|$dst, $src}",
+ "vmov{d|q}\t{$src, $dst|$dst, $src}",
[(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
(iPTR 0)))],
IIC_SSE_MOVD_ToGP>,
(SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
}
-let Predicates = [HasSSE2], AddedComplexity = 20 in {
+let Predicates = [UseSSE2], AddedComplexity = 20 in {
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
(MOVZDI2PDIrm addr:$src)>;
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
[(set VR128:$dst,
(v2i64 (scalar_to_vector (loadi64 addr:$src))))],
IIC_SSE_MOVDQ>, XS,
- Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
+ Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
//===---------------------------------------------------------------------===//
// Move Packed Quadword Int to Quadword Int
(v2i64 (X86vzmovl (v2i64 (scalar_to_vector
(loadi64 addr:$src))))))],
IIC_SSE_MOVDQ>,
- XS, Requires<[HasSSE2]>;
+ XS, Requires<[UseSSE2]>;
let Predicates = [HasAVX], AddedComplexity = 20 in {
def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
(VMOVZQI2PQIrm addr:$src)>;
}
-let Predicates = [HasSSE2], AddedComplexity = 20 in {
+let Predicates = [UseSSE2], AddedComplexity = 20 in {
def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
(MOVZQI2PQIrm addr:$src)>;
def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
"movq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
IIC_SSE_MOVQ_RR>,
- XS, Requires<[HasSSE2]>;
+ XS, Requires<[UseSSE2]>;
let AddedComplexity = 20 in
def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
[(set VR128:$dst, (v2i64 (X86vzmovl
(loadv2i64 addr:$src))))],
IIC_SSE_MOVDQ>,
- XS, Requires<[HasSSE2]>;
+ XS, Requires<[UseSSE2]>;
}
let AddedComplexity = 20 in {
def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
(VMOVZPQILo2PQIrr VR128:$src)>;
}
- let Predicates = [HasSSE2] in {
+ let Predicates = [UseSSE2] in {
def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
(MOVZPQILo2PQIrm addr:$src)>;
def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
(VMOVSLDUPYrm addr:$src)>;
}
-let Predicates = [HasSSE3] in {
+let Predicates = [UseSSE3] in {
def : Pat<(v4i32 (X86Movshdup VR128:$src)),
(MOVSHDUPrr VR128:$src)>;
def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
(VMOVDDUPYrr VR256:$src)>;
}
-let Predicates = [HasSSE3] in {
+let Predicates = [UseSSE3] in {
def : Pat<(X86Movddup (memopv2f64 addr:$src)),
(MOVDDUPrm addr:$src)>;
def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
}
}
-let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
+let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
let ExeDomain = SSEPackedSingle in
defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
f128mem, SSE_ALU_F32P>, TB, XD;
defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
let Predicates = [HasAVX2] in
defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
-let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
+let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
defm PALIGN : ssse3_palign<"palignr">;
let Predicates = [HasAVX2] in {
(VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
}
-let Predicates = [HasSSSE3] in {
+let Predicates = [UseSSSE3] 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:$imm))),
(VPMOVZXDQrm addr:$src)>;
}
-let Predicates = [HasSSE41] in {
+let Predicates = [UseSSE41] in {
// Common patterns involving scalar load.
def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
(PMOVSXBWrm addr:$src)>;
def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
}
-let Predicates = [HasSSE41] in {
+let Predicates = [UseSSE41] in {
def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
}
(VPMOVZXWQrm addr:$src)>;
}
-let Predicates = [HasSSE41] in {
+let Predicates = [UseSSE41] in {
// Common patterns involving scalar load
def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
(PMOVSXBDrm addr:$src)>;
(VPMOVZXBQrm addr:$src)>;
}
-let Predicates = [HasSSE41] in {
+let Predicates = [UseSSE41] in {
// Common patterns involving scalar load
def : Pat<(int_x86_sse41_pmovsxbq
(bitconvert (v4i32 (X86vzmovl
imm:$src2))),
addr:$dst),
(EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
- Requires<[HasSSE41]>;
+ Requires<[UseSSE41]>;
//===----------------------------------------------------------------------===//
// SSE4.1 - Insert Instructions
(bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
}
-/// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
+/// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
Intrinsic IntId256> {
let isCommutable = 1 in
def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
(PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
-let Predicates = [HasSSE41] in {
+let Predicates = [UseSSE41] in {
def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
(v16i8 VR128:$src2))),
(PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
}
let Defs = [EFLAGS], usesCustomInserter = 1 in {
- let AddedComplexity = 1 in
- defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
- defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
+ defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
+ defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
}
let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
}
let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
- let AddedComplexity = 1 in
- defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
- defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
+ defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
+ defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
}
let Predicates = [HasAVX],
(i32 imm)),
(VINSERTF128rr VR256:$src1, VR128:$src2,
(INSERT_get_vinsertf128_imm VR256:$ins))>;
+
+def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (memopv4f32 addr:$src2),
+ (i32 imm)),
+ (VINSERTF128rm VR256:$src1, addr:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (memopv2f64 addr:$src2),
+ (i32 imm)),
+ (VINSERTF128rm VR256:$src1, addr:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+}
+
+let Predicates = [HasAVX1Only] in {
def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
(i32 imm)),
(VINSERTF128rr VR256:$src1, VR128:$src2,
(VINSERTF128rr VR256:$src1, VR128:$src2,
(INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
+def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
(i32 imm)),
(VINSERTF128rm VR256:$src1, addr:$src2,
(INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
+def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
+ (bc_v4i32 (memopv2i64 addr:$src2)),
(i32 imm)),
(VINSERTF128rm VR256:$src1, addr:$src2,
(INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
+def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
+ (bc_v16i8 (memopv2i64 addr:$src2)),
+ (i32 imm)),
+ (VINSERTF128rm VR256:$src1, addr:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
+ (bc_v8i16 (memopv2i64 addr:$src2)),
(i32 imm)),
(VINSERTF128rm VR256:$src1, addr:$src2,
(INSERT_get_vinsertf128_imm VR256:$ins))>;
[]>, VEX;
}
-// Extract and store.
-let Predicates = [HasAVX] in {
- def : Pat<(alignedstore (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2), addr:$dst),
- (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
- def : Pat<(alignedstore (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2), addr:$dst),
- (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
- def : Pat<(alignedstore (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2), addr:$dst),
- (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
-
- def : Pat<(int_x86_sse_storeu_ps addr:$dst, (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2)),
- (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
- def : Pat<(int_x86_sse2_storeu_pd addr:$dst, (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2)),
- (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
- def : Pat<(int_x86_sse2_storeu_dq addr:$dst, (bc_v16i8 (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2))),
- (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
-}
-
// AVX1 patterns
let Predicates = [HasAVX] in {
-def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$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),
(v2f64 (VEXTRACTF128rr
(v4f64 VR256:$src1),
(EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+
+def : Pat<(alignedstore (v4f32 (vextractf128_extract:$ext (v8f32 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTF128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v2f64 (vextractf128_extract:$ext (v4f64 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTF128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+}
+
+let Predicates = [HasAVX1Only] in {
def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
(v2i64 (VEXTRACTF128rr
- (v4i64 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+ (v4i64 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)))>;
+ (v8i32 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)))>;
+ (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)))>;
+ (v32i8 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+
+def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTF128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTF128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTF128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTF128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
}
//===----------------------------------------------------------------------===//
}
let Predicates = [HasAVX] in {
+def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
+ (memopv4f64 addr:$src2), (i8 imm:$imm))),
+ (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
+}
+
+let Predicates = [HasAVX1Only] in {
def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
-def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
- (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
-def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
- (memopv8f32 addr:$src2), (i8 imm:$imm))),
- (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
(bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
(memopv4i64 addr:$src2), (i8 imm:$imm))),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
-def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
- (memopv4f64 addr:$src2), (i8 imm:$imm))),
- (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
(bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
}
// AVX1 broadcast patterns
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX1Only] in {
def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
(VBROADCASTSSYrm addr:$src)>;
def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
(VBROADCASTSDYrm addr:$src)>;
+def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
+ (VBROADCASTSSrm addr:$src)>;
+}
+
+let Predicates = [HasAVX] in {
def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
(VBROADCASTSSYrm addr:$src)>;
def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
(VBROADCASTSDYrm addr:$src)>;
def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
(VBROADCASTSSrm addr:$src)>;
-def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
- (VBROADCASTSSrm addr:$src)>;
// Provide fallback in case the load node that is used in the patterns above
// is used by additional users, which prevents the pattern selection.
//===----------------------------------------------------------------------===//
// VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
//
-let AddedComplexity = 1 in {
def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2, i8imm:$src3),
"vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
"vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
(i8 imm:$src3)))]>, VEX_4V;
-}
-let Predicates = [HasAVX2], AddedComplexity = 1 in {
+let Predicates = [HasAVX2] in {
def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
[]>, VEX_4V;
}
-let Predicates = [HasAVX2], AddedComplexity = 1 in {
+let Predicates = [HasAVX2] in {
def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
(i32 imm)),
(VINSERTI128rr VR256:$src1, VR128:$src2,
(i32 imm)),
(VINSERTI128rr VR256:$src1, VR128:$src2,
(INSERT_get_vinsertf128_imm VR256:$ins))>;
+
+def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
+ (i32 imm)),
+ (VINSERTI128rm VR256:$src1, addr:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
+ (bc_v4i32 (memopv2i64 addr:$src2)),
+ (i32 imm)),
+ (VINSERTI128rm VR256:$src1, addr:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
+ (bc_v16i8 (memopv2i64 addr:$src2)),
+ (i32 imm)),
+ (VINSERTI128rm VR256:$src1, addr:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
+ (bc_v8i16 (memopv2i64 addr:$src2)),
+ (i32 imm)),
+ (VINSERTI128rm VR256:$src1, addr:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
}
//===----------------------------------------------------------------------===//
(ins i128mem:$dst, VR256:$src1, i8imm:$src2),
"vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
-let Predicates = [HasAVX2], AddedComplexity = 1 in {
+let Predicates = [HasAVX2] in {
def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
(v2i64 (VEXTRACTI128rr
(v4i64 VR256:$src1),
(v16i8 (VEXTRACTI128rr
(v32i8 VR256:$src1),
(EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+
+def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTI128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTI128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTI128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
+ (i32 imm))), addr:$dst),
+ (VEXTRACTI128mr addr:$dst, VR256:$src1,
+ (EXTRACT_get_vextractf128_imm VR128:$ext))>;
}
//===----------------------------------------------------------------------===//