From 598ceb25d4a42b6a814b2acc5e7ce5c73d6b8fb7 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sun, 7 Feb 2021 11:41:43 +0000
Subject: [PATCH] [X86][AVX] Fold extract_subvector(splat, c) ->
 extract_subvector(splat, 0)

We already do this for VBROADCASTs, extend this for any splat that SelectionDAG::isSplatValue recognises as well.
---
 llvm/lib/Target/X86/X86ISelLowering.cpp      |  3 +-
 llvm/test/CodeGen/X86/vector-fshl-256.ll     | 48 ++++++++++++--------------
 llvm/test/CodeGen/X86/vector-fshr-256.ll     | 50 ++++++++++++----------------
 llvm/test/CodeGen/X86/vector-fshr-rot-256.ll |  8 ++---
 llvm/test/CodeGen/X86/vector-rotate-512.ll   |  4 ---
 5 files changed, 50 insertions(+), 63 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 6f8ef3e..bec7437 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -49852,7 +49852,8 @@ static SDValue combineExtractSubvector(SDNode *N, SelectionDAG &DAG,
   // extract the lowest subvector instead which should allow
   // SimplifyDemandedVectorElts do more simplifications.
   if (IdxVal != 0 && (InVec.getOpcode() == X86ISD::VBROADCAST ||
-                      InVec.getOpcode() == X86ISD::VBROADCAST_LOAD))
+                      InVec.getOpcode() == X86ISD::VBROADCAST_LOAD ||
+                      DAG.isSplatValue(InVec, /*AllowUndefs*/ false)))
     return extractSubVector(InVec, 0, DAG, SDLoc(N), SizeInBits);
 
   // If we're extracting a broadcasted subvector, just use the lowest subvector.
diff --git a/llvm/test/CodeGen/X86/vector-fshl-256.ll b/llvm/test/CodeGen/X86/vector-fshl-256.ll
index d673bde..6308a04 100644
--- a/llvm/test/CodeGen/X86/vector-fshl-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-256.ll
@@ -1521,39 +1521,35 @@ define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; XOPAVX1-NEXT:    vandps %ymm4, %ymm2, %ymm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm5, %xmm6
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vandnps %ymm4, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vpsubb %xmm4, %xmm3, %xmm4
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; XOPAVX1-NEXT:    vpcmpeqd %xmm6, %xmm6, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm5, %xmm4
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
-; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vandnps {{.*}}(%rip), %ymm4, %ymm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; XOPAVX1-NEXT:    vpsubb %xmm5, %xmm3, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
+; XOPAVX1-NEXT:    vpcmpeqd %xmm7, %xmm7, %xmm7
+; XOPAVX1-NEXT:    vpshlb %xmm7, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
+; XOPAVX1-NEXT:    vpsubb %xmm4, %xmm3, %xmm3
+; XOPAVX1-NEXT:    vpshlb %xmm7, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm3, %xmm3
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm6
-; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm4
+; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpandn {{.*}}(%rip), %ymm2, %ymm2
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
diff --git a/llvm/test/CodeGen/X86/vector-fshr-256.ll b/llvm/test/CodeGen/X86/vector-fshr-256.ll
index a1ab521..01e9af2 100644
--- a/llvm/test/CodeGen/X86/vector-fshr-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-256.ll
@@ -1465,45 +1465,39 @@ define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; XOPAVX1-NEXT:    vandnps %ymm4, %ymm2, %ymm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm5, %xmm6
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; XOPAVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm7
-; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm7, %xmm6
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vandnps {{.*}}(%rip), %ymm4, %ymm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; XOPAVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
 ; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vandps %ymm4, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vpsubb %xmm4, %xmm3, %xmm4
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm5, %xmm4
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
 ; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
 ; XOPAVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm6
-; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; XOPAVX2-NEXT:    vpandn {{.*}}(%rip), %ymm2, %ymm4
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
+; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
-; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
-; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm5, %xmm3
-; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm2
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm3
+; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
 ; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
diff --git a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
index 4114fba..f5a80d1 100644
--- a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
@@ -720,9 +720,9 @@ define <4 x i64> @splatvar_funnnel_v4i64(<4 x i64> %x, <4 x i64> %amt) nounwind
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX2:       # %bb.0:
+; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; XOPAVX2-NEXT:    vprotq %xmm1, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vprotq %xmm1, %xmm0, %xmm0
@@ -825,9 +825,9 @@ define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %amt) nounwind
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
+; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
-; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm0, %xmm0
@@ -963,9 +963,9 @@ define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %amt) nounw
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
-; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; XOPAVX2-NEXT:    vprotw %xmm1, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vprotw %xmm1, %xmm0, %xmm0
@@ -1155,9 +1155,9 @@ define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %amt) nounwind
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
-; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; XOPAVX2-NEXT:    vprotb %xmm1, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vprotb %xmm1, %xmm0, %xmm0
diff --git a/llvm/test/CodeGen/X86/vector-rotate-512.ll b/llvm/test/CodeGen/X86/vector-rotate-512.ll
index 13056a9..8eb66d2 100644
--- a/llvm/test/CodeGen/X86/vector-rotate-512.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-512.ll
@@ -430,7 +430,6 @@ define <32 x i16> @splatvar_rotate_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind
 ; AVX512F-LABEL: splatvar_rotate_v32i16:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
-; AVX512F-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX512F-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
@@ -447,7 +446,6 @@ define <32 x i16> @splatvar_rotate_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind
 ; AVX512VL-LABEL: splatvar_rotate_v32i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
-; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX512VL-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
@@ -506,7 +504,6 @@ define <64 x i8> @splatvar_rotate_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
 ; AVX512F-LABEL: splatvar_rotate_v64i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
@@ -532,7 +529,6 @@ define <64 x i8> @splatvar_rotate_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
 ; AVX512VL-LABEL: splatvar_rotate_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VL-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
-- 
2.7.4