Commit 35625464 authored by Craig Topper's avatar Craig Topper
Browse files

[X86] Fix the cost model for v16i16->v16i32 zero_extend/sign_extend with AVX2 

We seem to be inheriting the cost from sse4.1. But if we have 256-bit registers we should be able to do this with just one extract to split the 16i16 and two v8i16->v8i32 operations so our cost should be 3 not 4.

Differential Revision: https://reviews.llvm.org/D73646
parent 228ea1a4

llvm/lib/Target/X86/X86TargetTransformInfo.cpp

+2 −0
Original line number Diff line number Diff line
@@ -1424,6 +1424,8 @@ int X86TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src, 
    { ISD::ZERO_EXTEND, MVT::v8i32,  MVT::v8i16,  1 },

    { ISD::SIGN_EXTEND, MVT::v4i64,  MVT::v4i32,  1 },

    { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i32,  1 },

    { ISD::ZERO_EXTEND, MVT::v16i32, MVT::v16i16, 3 },

    { ISD::SIGN_EXTEND, MVT::v16i32, MVT::v16i16, 3 },



    { ISD::TRUNCATE,    MVT::v4i8,   MVT::v4i64,  2 },

    { ISD::TRUNCATE,    MVT::v4i16,  MVT::v4i64,  2 },

llvm/test/Analysis/CostModel/X86/arith-fix.ll

+4 −4
Original line number Diff line number Diff line
@@ -100,8 +100,8 @@ define i32 @smul(i32 %arg) { 
; AVX2-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %V16I32 = call <16 x i32> @llvm.smul.fix.v16i32(<16 x i32> undef, <16 x i32> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call i16 @llvm.smul.fix.i16(i16 undef, i16 undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V8I16 = call <8 x i16> @llvm.smul.fix.v8i16(<8 x i16> undef, <8 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V16I16 = call <16 x i16> @llvm.smul.fix.v16i16(<16 x i16> undef, <16 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V32I16 = call <32 x i16> @llvm.smul.fix.v32i16(<32 x i16> undef, <32 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V16I16 = call <16 x i16> @llvm.smul.fix.v16i16(<16 x i16> undef, <16 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V32I16 = call <32 x i16> @llvm.smul.fix.v32i16(<32 x i16> undef, <32 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I8 = call i8 @llvm.smul.fix.i8(i8 undef, i8 undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I8 = call <16 x i8> @llvm.smul.fix.v16i8(<16 x i8> undef, <16 x i8> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V32I8 = call <32 x i8> @llvm.smul.fix.v32i8(<32 x i8> undef, <32 x i8> undef, i32 3)
@@ -334,8 +334,8 @@ define i32 @umul(i32 %arg) { 
; AVX2-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %V16I32 = call <16 x i32> @llvm.umul.fix.v16i32(<16 x i32> undef, <16 x i32> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call i16 @llvm.umul.fix.i16(i16 undef, i16 undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V8I16 = call <8 x i16> @llvm.umul.fix.v8i16(<8 x i16> undef, <8 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V16I16 = call <16 x i16> @llvm.umul.fix.v16i16(<16 x i16> undef, <16 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V32I16 = call <32 x i16> @llvm.umul.fix.v32i16(<32 x i16> undef, <32 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V16I16 = call <16 x i16> @llvm.umul.fix.v16i16(<16 x i16> undef, <16 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V32I16 = call <32 x i16> @llvm.umul.fix.v32i16(<32 x i16> undef, <32 x i16> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I8 = call i8 @llvm.umul.fix.i8(i8 undef, i8 undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I8 = call <16 x i8> @llvm.umul.fix.v16i8(<16 x i8> undef, <16 x i8> undef, i32 3)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V32I8 = call <32 x i8> @llvm.umul.fix.v32i8(<32 x i8> undef, <32 x i8> undef, i32 3)

llvm/test/Analysis/CostModel/X86/arith-overflow.ll

+4 −4
Original line number Diff line number Diff line
@@ -1056,8 +1056,8 @@ define i32 @smul(i32 %arg) { 
; AVX2-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call { i16, i1 } @llvm.smul.with.overflow.i16(i16 undef, i16 undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.smul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.smul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.smul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I8 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 undef, i8 undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.smul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.smul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
@@ -1294,8 +1294,8 @@ define i32 @umul(i32 %arg) { 
; AVX2-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 undef, i16 undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.umul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.umul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.umul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 undef, i8 undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.umul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)

; AVX2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.umul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)

llvm/test/Analysis/CostModel/X86/cast.ll

+1 −1
Original line number Diff line number Diff line
@@ -177,7 +177,7 @@ define i32 @zext_sext(<8 x i1> %in) { 
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %D = zext <4 x i32> undef to <4 x i64>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %D1 = zext <8 x i32> undef to <8 x i64>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %D2 = sext <8 x i32> undef to <8 x i64>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %D3 = zext <16 x i16> undef to <16 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %D3 = zext <16 x i16> undef to <16 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %D4 = zext <16 x i8> undef to <16 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %D5 = zext <16 x i1> undef to <16 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %E = trunc <4 x i64> undef to <4 x i32>

llvm/test/Analysis/CostModel/X86/extend.ll

+2 −2
Original line number Diff line number Diff line
@@ -127,7 +127,7 @@ define i32 @zext_vXi16() { 
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = zext <2 x i16> undef to <2 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = zext <4 x i16> undef to <4 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = zext <8 x i16> undef to <8 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = zext <16 x i16> undef to <16 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V16i32 = zext <16 x i16> undef to <16 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef

;

; AVX512-LABEL: 'zext_vXi16'
@@ -630,7 +630,7 @@ define i32 @sext_vXi16() { 
; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i32 = sext <2 x i16> undef to <2 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = sext <4 x i16> undef to <4 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = sext <8 x i16> undef to <8 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = sext <16 x i16> undef to <16 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V16i32 = sext <16 x i16> undef to <16 x i32>

; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef

;

; AVX512-LABEL: 'sext_vXi16'