maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[CostModel][X86] Adjust vXi32 multiply costs if it can be performed using PMADDWD 

Update the costs to match the codegen from combineMulToPMADDWD - not only can we use PMADDWD is its zero-extended, but also if its a constant or sign-extended from a vXi16 (which can be replaced with a zero-extension).
This commit is contained in:
Simon Pilgrim 2021-09-25 16:28:48 +01:00
parent eb7c78c2c5
commit 8c83bd3bd4
4 changed files with 72 additions and 60 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
View File

@ -214,12 +214,24 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
unsigned Op2MinSize = BaseT::minRequiredElementSize(Args[1], Op2Signed);
unsigned OpMinSize = std::max(Op1MinSize, Op2MinSize);
// If both are representable as i15 and at least one is zero-extended,
// then we can treat this as PMADDWD which has the same costs
// as a vXi16 multiply..
if (OpMinSize <= 15 && (!Op1Signed || !Op2Signed) && !ST->isPMADDWDSlow())
LT.second =
MVT::getVectorVT(MVT::i16, 2 * LT.second.getVectorNumElements());
// If both are representable as i15 and at least one is constant,
// zero-extended, or sign-extended from vXi16 then we can treat this as
// PMADDWD which has the same costs as a vXi16 multiply.
if (OpMinSize <= 15 && !ST->isPMADDWDSlow()) {
bool Op1Constant =
isa<ConstantDataVector>(Args[0]) || isa<ConstantVector>(Args[0]);
bool Op2Constant =
isa<ConstantDataVector>(Args[1]) || isa<ConstantVector>(Args[1]);
bool Op1Sext16 = isa<SExtInst>(Args[0]) && Op1MinSize == 15;
bool Op2Sext16 = isa<SExtInst>(Args[1]) && Op2MinSize == 15;
bool IsZeroExtended = !Op1Signed || !Op2Signed;
bool IsConstant = Op1Constant || Op2Constant;
bool IsSext16 = Op1Sext16 || Op2Sext16;
if (IsConstant || IsZeroExtended || IsSext16)
LT.second =
MVT::getVectorVT(MVT::i16, 2 * LT.second.getVectorNumElements());
}
}
if ((ISD == ISD::SDIV || ISD == ISD::SREM || ISD == ISD::UDIV ||

70
llvm/test/Analysis/CostModel/X86/mul32.ll
View File

@ -525,11 +525,11 @@ define void @mul_sext_vXi16(<4 x i16> %a4, <4 x i16> %b4, <8 x i16> %a8, <8 x i1
; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %xb32 = sext <32 x i16> %b32 to <32 x i32>
; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %xa64 = sext <64 x i16> %a64 to <64 x i32>
; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %xb64 = sext <64 x i16> %b64 to <64 x i32>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSSE3-LABEL: 'mul_sext_vXi16'
@ -543,11 +543,11 @@ define void @mul_sext_vXi16(<4 x i16> %a4, <4 x i16> %b4, <8 x i16> %a8, <8 x i1
; SSSE3-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %xb32 = sext <32 x i16> %b32 to <32 x i32>
; SSSE3-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %xa64 = sext <64 x i16> %a64 to <64 x i32>
; SSSE3-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %xb64 = sext <64 x i16> %b64 to <64 x i32>
; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; SSSE3-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; SSSE3-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; SSSE3-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; SSSE3-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'mul_sext_vXi16'
@ -561,11 +561,11 @@ define void @mul_sext_vXi16(<4 x i16> %a4, <4 x i16> %b4, <8 x i16> %a8, <8 x i1
; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %xb32 = sext <32 x i16> %b32 to <32 x i32>
; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %xa64 = sext <64 x i16> %a64 to <64 x i32>
; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %xb64 = sext <64 x i16> %b64 to <64 x i32>
; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'mul_sext_vXi16'
@ -579,11 +579,11 @@ define void @mul_sext_vXi16(<4 x i16> %a4, <4 x i16> %b4, <8 x i16> %a8, <8 x i1
; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %xb32 = sext <32 x i16> %b32 to <32 x i32>
; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %xa64 = sext <64 x i16> %a64 to <64 x i32>
; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %xb64 = sext <64 x i16> %b64 to <64 x i32>
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX2-LABEL: 'mul_sext_vXi16'
@ -597,11 +597,11 @@ define void @mul_sext_vXi16(<4 x i16> %a4, <4 x i16> %b4, <8 x i16> %a8, <8 x i1
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %xb32 = sext <32 x i16> %b32 to <32 x i32>
; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %xa64 = sext <64 x i16> %a64 to <64 x i32>
; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %xb64 = sext <64 x i16> %b64 to <64 x i32>
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'mul_sext_vXi16'
@ -633,11 +633,11 @@ define void @mul_sext_vXi16(<4 x i16> %a4, <4 x i16> %b4, <8 x i16> %a8, <8 x i1
; SLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %xb32 = sext <32 x i16> %b32 to <32 x i32>
; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %xa64 = sext <64 x i16> %a64 to <64 x i32>
; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %xb64 = sext <64 x i16> %b64 to <64 x i32>
; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; SLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; SLM-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; SLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; SLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'mul_sext_vXi16'
@ -651,11 +651,11 @@ define void @mul_sext_vXi16(<4 x i16> %a4, <4 x i16> %b4, <8 x i16> %a8, <8 x i1
; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %xb32 = sext <32 x i16> %b32 to <32 x i32>
; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %xa64 = sext <64 x i16> %a64 to <64 x i32>
; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %xb64 = sext <64 x i16> %b64 to <64 x i32>
; GLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; GLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res4 = mul <4 x i32> %xa4, %xb4
; GLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res8 = mul <8 x i32> %xa8, %xb8
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %res16 = mul <16 x i32> %xa16, %xb16
; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %res32 = mul <32 x i32> %xa32, %xb32
; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %res64 = mul <64 x i32> %xa64, %xb64
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
%xa4 = sext <4 x i16> %a4 to <4 x i32>

16
llvm/test/Analysis/CostModel/X86/slm-arith-costs.ll
View File

@ -127,12 +127,12 @@ entry:
define <4 x i32> @slm-costs_8_v4_sext_mul(<4 x i8> %a) {
; SLM-LABEL: 'slm-costs_8_v4_sext_mul'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sext = sext <4 x i8> %a to <4 x i32>
; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -128, i32 127, i32 -128>
; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -128, i32 127, i32 -128>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
;
; GLM-LABEL: 'slm-costs_8_v4_sext_mul'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sext = sext <4 x i8> %a to <4 x i32>
; GLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -128, i32 127, i32 -128>
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -128, i32 127, i32 -128>
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
;
entry:
@ -144,12 +144,12 @@ entry:
define <4 x i32> @slm-costs_8_v4_sext_mul_fail(<4 x i8> %a) {
; SLM-LABEL: 'slm-costs_8_v4_sext_mul_fail'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sext = sext <4 x i8> %a to <4 x i32>
; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -128, i32 128, i32 -128>
; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -128, i32 128, i32 -128>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
;
; GLM-LABEL: 'slm-costs_8_v4_sext_mul_fail'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sext = sext <4 x i8> %a to <4 x i32>
; GLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -128, i32 128, i32 -128>
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -128, i32 128, i32 -128>
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
;
entry:
@ -161,12 +161,12 @@ entry:
define <4 x i32> @slm-costs_8_v4_sext_mul_fail_2(<4 x i8> %a) {
; SLM-LABEL: 'slm-costs_8_v4_sext_mul_fail_2'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sext = sext <4 x i8> %a to <4 x i32>
; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -129, i32 127, i32 -128>
; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -129, i32 127, i32 -128>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
;
; GLM-LABEL: 'slm-costs_8_v4_sext_mul_fail_2'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sext = sext <4 x i8> %a to <4 x i32>
; GLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -129, i32 127, i32 -128>
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res = mul nsw <4 x i32> %sext, <i32 127, i32 -129, i32 127, i32 -128>
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
;
entry:
@ -296,12 +296,12 @@ entry:
define <4 x i32> @slm-costs_16_v4_sext_mul(<4 x i16> %a) {
; SLM-LABEL: 'slm-costs_16_v4_sext_mul'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sext = sext <4 x i16> %a to <4 x i32>
; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %res = mul nsw <4 x i32> %sext, <i32 32767, i32 -32768, i32 32767, i32 -32768>
; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res = mul nsw <4 x i32> %sext, <i32 32767, i32 -32768, i32 32767, i32 -32768>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
;
; GLM-LABEL: 'slm-costs_16_v4_sext_mul'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sext = sext <4 x i16> %a to <4 x i32>
; GLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %res = mul nsw <4 x i32> %sext, <i32 32767, i32 -32768, i32 32767, i32 -32768>
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %res = mul nsw <4 x i32> %sext, <i32 32767, i32 -32768, i32 32767, i32 -32768>
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %res
;
entry:

22
llvm/test/Transforms/LoopVectorize/X86/mul_slm_16bit.ll
View File

@ -32,38 +32,38 @@ for.body: ; preds = %for.body.preheader,
%conv3 = sext i8 %1 to i32
; sources of the mul is sext\sext from i8
; use pmullw\sext seq.
; SLM: cost of 3 for VF 4 {{.*}} mul nsw i32 %conv3, %conv
; SLM: cost of 3 for VF 2 {{.*}} mul nsw i32 %conv3, %conv
%mul = mul nsw i32 %conv3, %conv
; sources of the mul is zext\sext from i8
; use pmulhw\pmullw\pshuf
; SLM: cost of 2 for VF 4 {{.*}} mul nsw i32 %conv4, %conv
; SLM: cost of 2 for VF 2 {{.*}} mul nsw i32 %conv4, %conv
%conv4 = zext i8 %1 to i32
%mul2 = mul nsw i32 %conv4, %conv
%sum0 = add i32 %mul, %mul2
; sources of the mul is zext\zext from i8
; use pmullw\zext
; SLM: cost of 2 for VF 4 {{.*}} mul nsw i32 %conv5, %conv4
; SLM: cost of 2 for VF 2 {{.*}} mul nsw i32 %conv5, %conv4
%conv5 = zext i8 %0 to i32
%mul3 = mul nsw i32 %conv5, %conv4
%sum1 = add i32 %sum0, %mul3
; sources of the mul is sext\-120
; use pmullw\sext
; SLM: cost of 3 for VF 4 {{.*}} mul nsw i32 -120, %conv3
; SLM: cost of 3 for VF 2 {{.*}} mul nsw i32 -120, %conv3
%mul4 = mul nsw i32 -120, %conv3
%sum2 = add i32 %sum1, %mul4
; sources of the mul is sext\250
; use pmulhw\pmullw\pshuf
; SLM: cost of 2 for VF 4 {{.*}} mul nsw i32 250, %conv3
; SLM: cost of 2 for VF 2 {{.*}} mul nsw i32 250, %conv3
%mul5 = mul nsw i32 250, %conv3
%sum3 = add i32 %sum2, %mul5
; sources of the mul is zext\-120
; use pmulhw\pmullw\pshuf
; SLM: cost of 2 for VF 4 {{.*}} mul nsw i32 -120, %conv4
; SLM: cost of 2 for VF 2 {{.*}} mul nsw i32 -120, %conv4
%mul6 = mul nsw i32 -120, %conv4
%sum4 = add i32 %sum3, %mul6
; sources of the mul is zext\250
; use pmullw\zext
; SLM: cost of 2 for VF 4 {{.*}} mul nsw i32 250, %conv4
; SLM: cost of 2 for VF 2 {{.*}} mul nsw i32 250, %conv4
%mul7 = mul nsw i32 250, %conv4
%sum5 = add i32 %sum4, %mul7
%add = add i32 %acc.013, 5
@ -101,23 +101,23 @@ for.body: ; preds = %for.body.preheader,
%conv3 = sext i16 %1 to i32
; sources of the mul is sext\sext from i16
; use pmulhw\pmullw\pshuf seq.
; SLM: cost of 5 for VF 4 {{.*}} mul nsw i32 %conv3, %conv
; SLM: cost of 3 for VF 4 {{.*}} mul nsw i32 %conv3, %conv
%mul = mul nsw i32 %conv3, %conv
; sources of the mul is zext\sext from i16
; use pmulld
; SLM: cost of 11 for VF 4 {{.*}} mul nsw i32 %conv4, %conv
; SLM: cost of 2 for VF 4 {{.*}} mul nsw i32 %conv4, %conv
%conv4 = zext i16 %1 to i32
%mul2 = mul nsw i32 %conv4, %conv
%sum0 = add i32 %mul, %mul2
; sources of the mul is zext\zext from i16
; use pmulhw\pmullw\zext
; SLM: cost of 5 for VF 4 {{.*}} mul nsw i32 %conv5, %conv4
; SLM: cost of 2 for VF 4 {{.*}} mul nsw i32 %conv5, %conv4
%conv5 = zext i16 %0 to i32
%mul3 = mul nsw i32 %conv5, %conv4
%sum1 = add i32 %sum0, %mul3
; sources of the mul is sext\-32000
; use pmulhw\pmullw\sext
; SLM: cost of 5 for VF 4 {{.*}} mul nsw i32 -32000, %conv3
; SLM: cost of 2 for VF 4 {{.*}} mul nsw i32 -32000, %conv3
%mul4 = mul nsw i32 -32000, %conv3
%sum2 = add i32 %sum1, %mul4
; sources of the mul is sext\64000