[X86] combineConcatVectorOps - add support for AVX1 and/or/xor/andnp concatenation #130645
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…concatenation Limit AVX1-only targets to cases where at least one operand is beneficial to concatenate, but avoid unnecessary 256-bit NOT patterns unless it allows further folds. There are still a large number of regressions when limiting AVX2+ targets to beneficial concatenation - so added a TODO for now.
|
@llvm/pr-subscribers-backend-x86 Author: Simon Pilgrim (RKSimon) ChangesLimit AVX1-only targets to cases where at least one operand is beneficial to concatenate, but avoid unnecessary 256-bit NOT patterns unless it allows further folds. There are still a large number of regressions when limiting AVX2+ targets to beneficial concatenation - so added a TODO for now. Patch is 507.55 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/130645.diff 21 Files Affected:
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index aec071f038dc6..b27d9da922b22 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -58304,11 +58304,21 @@ static SDValue combineConcatVectorOps(const SDLoc &DL, MVT VT,
case ISD::OR:
case ISD::XOR:
case X86ISD::ANDNP:
- if (!IsSplat && ((VT.is256BitVector() && Subtarget.hasInt256()) ||
+ // TODO: AVX2+ targets should only use CombineSubOperand like AVX1.
+ if (!IsSplat && (VT.is256BitVector() ||
(VT.is512BitVector() && Subtarget.useAVX512Regs()))) {
- return DAG.getNode(Op0.getOpcode(), DL, VT,
- ConcatSubOperand(VT, Ops, 0),
- ConcatSubOperand(VT, Ops, 1));
+ // Don't concatenate root AVX1 NOT patterns.
+ if (Op0.getOpcode() == ISD::XOR && Depth == 0 &&
+ !Subtarget.hasInt256() && llvm::all_of(Ops, [](SDValue X) {
+ return ISD::isBuildVectorAllOnes(X.getOperand(1).getNode());
+ }))
+ break;
+ SDValue Concat0 = CombineSubOperand(VT, Ops, 0);
+ SDValue Concat1 = CombineSubOperand(VT, Ops, 1);
+ if (Concat0 || Concat1 || Subtarget.hasInt256())
+ return DAG.getNode(Op0.getOpcode(), DL, VT,
+ Concat0 ? Concat0 : ConcatSubOperand(VT, Ops, 0),
+ Concat1 ? Concat1 : ConcatSubOperand(VT, Ops, 1));
}
break;
case X86ISD::PCMPEQ:
diff --git a/llvm/test/CodeGen/X86/avx-shift.ll b/llvm/test/CodeGen/X86/avx-shift.ll
index 1da78ebd75b14..c9c09d7d10346 100644
--- a/llvm/test/CodeGen/X86/avx-shift.ll
+++ b/llvm/test/CodeGen/X86/avx-shift.ll
@@ -136,13 +136,11 @@ define <32 x i8> @vshift10(<32 x i8> %a) {
define <32 x i8> @vshift11(<32 x i8> %a) {
; CHECK-LABEL: vshift11:
; CHECK: # %bb.0:
-; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT: vpsrlw $2, %xmm1, %xmm1
-; CHECK-NEXT: vbroadcastss {{.*#+}} xmm2 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; CHECK-NEXT: vpand %xmm2, %xmm1, %xmm1
+; CHECK-NEXT: vpsrlw $2, %xmm0, %xmm1
+; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm0
; CHECK-NEXT: vpsrlw $2, %xmm0, %xmm0
-; CHECK-NEXT: vpand %xmm2, %xmm0, %xmm0
-; CHECK-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; CHECK-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; CHECK-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; CHECK-NEXT: retq
%s = lshr <32 x i8> %a, <i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2>
ret <32 x i8> %s
@@ -151,13 +149,11 @@ define <32 x i8> @vshift11(<32 x i8> %a) {
define <32 x i8> @vshift12(<32 x i8> %a) {
; CHECK-LABEL: vshift12:
; CHECK: # %bb.0:
-; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm1
-; CHECK-NEXT: vpsllw $2, %xmm1, %xmm1
-; CHECK-NEXT: vbroadcastss {{.*#+}} xmm2 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; CHECK-NEXT: vpand %xmm2, %xmm1, %xmm1
+; CHECK-NEXT: vpsllw $2, %xmm0, %xmm1
+; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm0
; CHECK-NEXT: vpsllw $2, %xmm0, %xmm0
-; CHECK-NEXT: vpand %xmm2, %xmm0, %xmm0
-; CHECK-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; CHECK-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; CHECK-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; CHECK-NEXT: retq
%s = shl <32 x i8> %a, <i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2>
ret <32 x i8> %s
diff --git a/llvm/test/CodeGen/X86/gfni-shifts.ll b/llvm/test/CodeGen/X86/gfni-shifts.ll
index 38f6cabd21fba..5cd1a2c76762e 100644
--- a/llvm/test/CodeGen/X86/gfni-shifts.ll
+++ b/llvm/test/CodeGen/X86/gfni-shifts.ll
@@ -1232,21 +1232,19 @@ define <32 x i8> @constant_shl_v32i8(<32 x i8> %a) nounwind {
;
; GFNIAVX1-LABEL: constant_shl_v32i8:
; GFNIAVX1: # %bb.0:
-; GFNIAVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; GFNIAVX1-NEXT: vpmovzxbw {{.*#+}} xmm2 = [1,4,16,64,128,32,8,2]
-; GFNIAVX1-NEXT: vpmaddubsw %xmm2, %xmm1, %xmm3
-; GFNIAVX1-NEXT: vbroadcastss {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; GFNIAVX1-NEXT: vpand %xmm4, %xmm3, %xmm3
-; GFNIAVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [0,2,0,8,0,32,0,128,0,64,0,16,0,4,0,1]
-; GFNIAVX1-NEXT: vpmaddubsw %xmm5, %xmm1, %xmm1
+; GFNIAVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [0,2,0,8,0,32,0,128,0,64,0,16,0,4,0,1]
+; GFNIAVX1-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm2
+; GFNIAVX1-NEXT: vpsllw $8, %xmm2, %xmm2
+; GFNIAVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
+; GFNIAVX1-NEXT: vpmaddubsw %xmm1, %xmm3, %xmm1
; GFNIAVX1-NEXT: vpsllw $8, %xmm1, %xmm1
-; GFNIAVX1-NEXT: vpor %xmm1, %xmm3, %xmm1
-; GFNIAVX1-NEXT: vpmaddubsw %xmm2, %xmm0, %xmm2
-; GFNIAVX1-NEXT: vpand %xmm4, %xmm2, %xmm2
-; GFNIAVX1-NEXT: vpmaddubsw %xmm5, %xmm0, %xmm0
-; GFNIAVX1-NEXT: vpsllw $8, %xmm0, %xmm0
-; GFNIAVX1-NEXT: vpor %xmm0, %xmm2, %xmm0
-; GFNIAVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; GFNIAVX1-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm1
+; GFNIAVX1-NEXT: vpmovzxbw {{.*#+}} xmm2 = [1,4,16,64,128,32,8,2]
+; GFNIAVX1-NEXT: vpmaddubsw %xmm2, %xmm3, %xmm3
+; GFNIAVX1-NEXT: vpmaddubsw %xmm2, %xmm0, %xmm0
+; GFNIAVX1-NEXT: vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; GFNIAVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; GFNIAVX1-NEXT: vorps %ymm1, %ymm0, %ymm0
; GFNIAVX1-NEXT: retq
;
; GFNIAVX2-LABEL: constant_shl_v32i8:
@@ -2575,33 +2573,31 @@ define <64 x i8> @constant_shl_v64i8(<64 x i8> %a) nounwind {
;
; GFNIAVX1-LABEL: constant_shl_v64i8:
; GFNIAVX1: # %bb.0:
-; GFNIAVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT: vpmovzxbw {{.*#+}} xmm3 = [1,4,16,64,128,32,8,2]
-; GFNIAVX1-NEXT: vpmaddubsw %xmm3, %xmm2, %xmm4
-; GFNIAVX1-NEXT: vbroadcastss {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
-; GFNIAVX1-NEXT: vpand %xmm5, %xmm4, %xmm4
-; GFNIAVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,2,0,8,0,32,0,128,0,64,0,16,0,4,0,1]
-; GFNIAVX1-NEXT: vpmaddubsw %xmm6, %xmm2, %xmm2
-; GFNIAVX1-NEXT: vpsllw $8, %xmm2, %xmm2
-; GFNIAVX1-NEXT: vpor %xmm2, %xmm4, %xmm2
-; GFNIAVX1-NEXT: vpmaddubsw %xmm3, %xmm0, %xmm4
-; GFNIAVX1-NEXT: vpand %xmm5, %xmm4, %xmm4
-; GFNIAVX1-NEXT: vpmaddubsw %xmm6, %xmm0, %xmm0
-; GFNIAVX1-NEXT: vpsllw $8, %xmm0, %xmm0
-; GFNIAVX1-NEXT: vpor %xmm0, %xmm4, %xmm0
-; GFNIAVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT: vpmaddubsw %xmm3, %xmm2, %xmm4
-; GFNIAVX1-NEXT: vpand %xmm5, %xmm4, %xmm4
-; GFNIAVX1-NEXT: vpmaddubsw %xmm6, %xmm2, %xmm2
+; GFNIAVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [0,2,0,8,0,32,0,128,0,64,0,16,0,4,0,1]
+; GFNIAVX1-NEXT: vpmaddubsw %xmm2, %xmm0, %xmm3
+; GFNIAVX1-NEXT: vpsllw $8, %xmm3, %xmm3
+; GFNIAVX1-NEXT: vextractf128 $1, %ymm0, %xmm4
+; GFNIAVX1-NEXT: vpmaddubsw %xmm2, %xmm4, %xmm5
+; GFNIAVX1-NEXT: vpsllw $8, %xmm5, %xmm5
+; GFNIAVX1-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3
+; GFNIAVX1-NEXT: vpmovzxbw {{.*#+}} xmm5 = [1,4,16,64,128,32,8,2]
+; GFNIAVX1-NEXT: vpmaddubsw %xmm5, %xmm4, %xmm4
+; GFNIAVX1-NEXT: vpmaddubsw %xmm5, %xmm0, %xmm0
+; GFNIAVX1-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0
+; GFNIAVX1-NEXT: vbroadcastss {{.*#+}} ymm4 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
+; GFNIAVX1-NEXT: vandps %ymm4, %ymm0, %ymm0
+; GFNIAVX1-NEXT: vorps %ymm3, %ymm0, %ymm0
+; GFNIAVX1-NEXT: vpmaddubsw %xmm2, %xmm1, %xmm3
+; GFNIAVX1-NEXT: vpsllw $8, %xmm3, %xmm3
+; GFNIAVX1-NEXT: vextractf128 $1, %ymm1, %xmm6
+; GFNIAVX1-NEXT: vpmaddubsw %xmm2, %xmm6, %xmm2
; GFNIAVX1-NEXT: vpsllw $8, %xmm2, %xmm2
-; GFNIAVX1-NEXT: vpor %xmm2, %xmm4, %xmm2
-; GFNIAVX1-NEXT: vpmaddubsw %xmm3, %xmm1, %xmm3
-; GFNIAVX1-NEXT: vpand %xmm5, %xmm3, %xmm3
-; GFNIAVX1-NEXT: vpmaddubsw %xmm6, %xmm1, %xmm1
-; GFNIAVX1-NEXT: vpsllw $8, %xmm1, %xmm1
-; GFNIAVX1-NEXT: vpor %xmm1, %xmm3, %xmm1
-; GFNIAVX1-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; GFNIAVX1-NEXT: vinsertf128 $1, %xmm2, %ymm3, %ymm2
+; GFNIAVX1-NEXT: vpmaddubsw %xmm5, %xmm6, %xmm3
+; GFNIAVX1-NEXT: vpmaddubsw %xmm5, %xmm1, %xmm1
+; GFNIAVX1-NEXT: vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; GFNIAVX1-NEXT: vandps %ymm4, %ymm1, %ymm1
+; GFNIAVX1-NEXT: vorps %ymm2, %ymm1, %ymm1
; GFNIAVX1-NEXT: retq
;
; GFNIAVX2-LABEL: constant_shl_v64i8:
diff --git a/llvm/test/CodeGen/X86/known-bits-vector.ll b/llvm/test/CodeGen/X86/known-bits-vector.ll
index 2eef32eb61414..dbf3d6635fb92 100644
--- a/llvm/test/CodeGen/X86/known-bits-vector.ll
+++ b/llvm/test/CodeGen/X86/known-bits-vector.ll
@@ -384,23 +384,19 @@ declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
define <8 x float> @knownbits_mask_concat_uitofp(<4 x i32> %a0, <4 x i32> %a1) nounwind {
; X86-LABEL: knownbits_mask_concat_uitofp:
; X86: # %bb.0:
-; X86-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,3,1,3]
-; X86-NEXT: vbroadcastss {{.*#+}} xmm2 = [131071,131071,131071,131071]
-; X86-NEXT: vandps %xmm2, %xmm1, %xmm1
; X86-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2,0,2]
-; X86-NEXT: vandps %xmm2, %xmm0, %xmm0
+; X86-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,3,1,3]
; X86-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; X86-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vcvtdq2ps %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: knownbits_mask_concat_uitofp:
; X64: # %bb.0:
-; X64-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,3,1,3]
-; X64-NEXT: vbroadcastss {{.*#+}} xmm2 = [131071,131071,131071,131071]
-; X64-NEXT: vandps %xmm2, %xmm1, %xmm1
; X64-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2,0,2]
-; X64-NEXT: vandps %xmm2, %xmm0, %xmm0
+; X64-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,3,1,3]
; X64-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; X64-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vcvtdq2ps %ymm0, %ymm0
; X64-NEXT: retq
%1 = and <4 x i32> %a0, <i32 131071, i32 -1, i32 131071, i32 -1>
diff --git a/llvm/test/CodeGen/X86/pmul.ll b/llvm/test/CodeGen/X86/pmul.ll
index fe8a4fa163129..c7cc2acaf2627 100644
--- a/llvm/test/CodeGen/X86/pmul.ll
+++ b/llvm/test/CodeGen/X86/pmul.ll
@@ -959,7 +959,7 @@ define <64 x i8> @mul_v64i8(<64 x i8> %i, <64 x i8> %j) nounwind {
;
; AVX512F-LABEL: mul_v64i8:
; AVX512F: # %bb.0: # %entry
-; AVX512F-NEXT: vpbroadcastd {{.*#+}} zmm2 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
+; AVX512F-NEXT: vpbroadcastd {{.*#+}} zmm2 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm3
; AVX512F-NEXT: vpand %ymm2, %ymm3, %ymm4
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm5
diff --git a/llvm/test/CodeGen/X86/vector-fshl-256.ll b/llvm/test/CodeGen/X86/vector-fshl-256.ll
index 6fb74643e4bb7..fdd0d68b89003 100644
--- a/llvm/test/CodeGen/X86/vector-fshl-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-256.ll
@@ -2270,20 +2270,16 @@ define <16 x i16> @splatconstant_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y) no
define <32 x i8> @splatconstant_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y) nounwind {
; AVX1-LABEL: splatconstant_funnnel_v32i8:
; AVX1: # %bb.0:
-; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT: vpsrlw $4, %xmm2, %xmm2
-; AVX1-NEXT: vbroadcastss {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm2
+; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT: vpsllw $4, %xmm2, %xmm2
-; AVX1-NEXT: vbroadcastss {{.*#+}} xmm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm1
+; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
+; AVX1-NEXT: vpsllw $4, %xmm0, %xmm2
+; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT: vpsllw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; AVX1-NEXT: vorps %ymm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
diff --git a/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll b/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
index 9806d63257180..f0f3d6028c27a 100644
--- a/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
@@ -1623,19 +1623,17 @@ define <16 x i16> @splatconstant_funnnel_v16i16(<16 x i16> %x) nounwind {
define <32 x i8> @splatconstant_funnnel_v32i8(<32 x i8> %x) nounwind {
; AVX1-LABEL: splatconstant_funnnel_v32i8:
; AVX1: # %bb.0:
-; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm2
-; AVX1-NEXT: vbroadcastss {{.*#+}} xmm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX1-NEXT: vpandn %xmm2, %xmm3, %xmm2
-; AVX1-NEXT: vpsllw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vpor %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm2
-; AVX1-NEXT: vpandn %xmm2, %xmm3, %xmm2
+; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm1
+; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT: vpsrlw $4, %xmm2, %xmm3
+; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT: vbroadcastss {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX1-NEXT: vandnps %ymm1, %ymm3, %ymm1
; AVX1-NEXT: vpsllw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpor %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT: vpsllw $4, %xmm2, %xmm2
+; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT: vandps %ymm3, %ymm0, %ymm0
+; AVX1-NEXT: vorps %ymm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: splatconstant_funnnel_v32i8:
diff --git a/llvm/test/CodeGen/X86/vector-fshr-256.ll b/llvm/test/CodeGen/X86/vector-fshr-256.ll
index 1741ae213d126..6d7d279a3637a 100644
--- a/llvm/test/CodeGen/X86/vector-fshr-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-256.ll
@@ -1625,17 +1625,6 @@ define <32 x i8> @constant_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y) nounwind {
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm2
; AVX1-NEXT: vpsrlw $8, %xmm2, %xmm2
; AVX1-NEXT: vpackuswb %xmm4, %xmm2, %xmm2
-; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4
-; AVX1-NEXT: vpaddb %xmm4, %xmm4, %xmm4
-; AVX1-NEXT: vpmovzxbw {{.*#+}} xmm7 = [128,32,8,2,128,2,8,32]
-; AVX1-NEXT: vpmaddubsw %xmm7, %xmm4, %xmm8
-; AVX1-NEXT: vbroadcastss {{.*#+}} xmm9 = [255,255,255,255,255,255,255,255]
-; AVX1-NEXT: vpand %xmm9, %xmm8, %xmm8
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm10 = [0,64,0,16,0,4,0,1,0,1,0,4,0,16,0,64]
-; AVX1-NEXT: vpmaddubsw %xmm10, %xmm4, %xmm4
-; AVX1-NEXT: vpsllw $8, %xmm4, %xmm4
-; AVX1-NEXT: vpor %xmm4, %xmm8, %xmm4
-; AVX1-NEXT: vpor %xmm2, %xmm4, %xmm2
; AVX1-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm3[8],xmm1[9],xmm3[9],xmm1[10],xmm3[10],xmm1[11],xmm3[11],xmm1[12],xmm3[12],xmm1[13],xmm3[13],xmm1[14],xmm3[14],xmm1[15],xmm3[15]
; AVX1-NEXT: vpmullw %xmm5, %xmm3, %xmm3
; AVX1-NEXT: vpsrlw $8, %xmm3, %xmm3
@@ -1643,14 +1632,23 @@ define <32 x i8> @constant_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y) nounwind {
; AVX1-NEXT: vpmullw %xmm6, %xmm1, %xmm1
; AVX1-NEXT: vpsrlw $8, %xmm1, %xmm1
; AVX1-NEXT: vpackuswb %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX1-NEXT: vpaddb %xmm0, %xmm0, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [0,64,0,16,0,4,0,1,0,1,0,4,0,16,0,64]
+; AVX1-NEXT: vpmaddubsw %xmm3, %xmm2, %xmm4
+; AVX1-NEXT: vpsllw $8, %xmm4, %xmm4
+; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT: vpaddb %xmm0, %xmm0, %xmm0
-; AVX1-NEXT: vpmaddubsw %xmm7, %xmm0, %xmm3
-; AVX1-NEXT: vpand %xmm3, %xmm9, %xmm3
-; AVX1-NEXT: vpmaddubsw %xmm10, %xmm0, %xmm0
-; AVX1-NEXT: vpsllw $8, %xmm0, %xmm0
-; AVX1-NEXT: vpor %xmm0, %xmm3, %xmm0
-; AVX1-NEXT: vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT: vpmaddubsw %xmm3, %xmm0, %xmm3
+; AVX1-NEXT: vpsllw $8, %xmm3, %xmm3
+; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm4, %ymm3
+; AVX1-NEXT: vorps %ymm1, %ymm3, %ymm1
+; AVX1-NEXT: vpmovzxbw {{.*#+}} xmm3 = [128,32,8,2,128,2,8,32]
+; AVX1-NEXT: vpmaddubsw %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpmaddubsw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX1-NEXT: vorps %ymm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: constant_funnnel_v32i8:
@@ -2058,20 +2056,16 @@ define <16 x i16> @splatconstant_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y) no
define <32 x i8> @splatconstant_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y) nounwind {
; AVX1-LABEL: splatconstant_funnnel_v32i8:
; AVX1: # %bb.0:
-; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT: vpsrlw $4, %xmm2, %xmm2
-; AVX1-NEXT: vbroadcastss {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm2
+; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT: vpsllw $4, %xmm2, %xmm2
-; AVX1-NEXT: vbroadcastss {{.*#+}} xmm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm1
+; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
+; AVX1-NEXT: vpsllw $4, %xmm0, %xmm2
+; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT: vpsllw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; AVX1-NEXT: vorps %ymm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
diff --git a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
index 3055663786740..c66a3f709c5ca 100644
--- a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
@@ -1674,19 +1674,17 @@ define <16 x i16> @splatconstant_funnnel_v16i16(<16 x i16> %x) nounwind {
define <32 x i8> @splatconstant_funnnel_v32i8(<32 x i8> %x) nounwind {
; AVX1-LABEL: splatconstant_funnnel_v32i8:
; AVX1: # %bb.0:
-; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm2
-; AVX1-NEXT: vbroadcastss {{.*#+}} xmm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX1-NEXT: vpandn %xmm2, %xmm3, %xmm2
-; AVX1-NEXT: vpsllw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vpor %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm2
-; AVX1-NEXT: vpandn %xmm2, %xmm3, %xmm2
+; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm1
+; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT: vpsrlw $4, %xmm2, %xmm3
+; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT: vbroadcastss {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX1-NEXT: vandnps %ymm1, %ymm3, %ymm1
; AVX1-NEXT: vpsllw $4, %xm...
[truncated]
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Limit AVX1-only targets to cases where at least one operand is beneficial to concatenate, but avoid unnecessary 256-bit NOT patterns unless it allows further folds.
There are still a large number of regressions when limiting AVX2+ targets to beneficial concatenation - so added a TODO for now.