Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[DAG] isSplatValue - node is a splat if all demanded elts have the same whole constant value #74443

Merged
merged 1 commit into from
Dec 8, 2023
Merged

[DAG] isSplatValue - node is a splat if all demanded elts have the same whole constant value #74443

merged 1 commit into from
Dec 8, 2023

Conversation

RKSimon
Copy link
Collaborator

@RKSimon RKSimon commented Dec 5, 2023

No description provided.

@llvmbot
Copy link
Collaborator

llvmbot commented Dec 5, 2023 •
edited

@llvm/pr-subscribers-llvm-selectiondag
@llvm/pr-subscribers-backend-x86

@llvm/pr-subscribers-backend-arm

Author: Simon Pilgrim (RKSimon)

Changes

Patch is 326.97 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/74443.diff

13 Files Affected:

  • (modified) llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp (+6)
  • (modified) llvm/test/CodeGen/ARM/vector-store.ll (+7-10)
  • (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll (+5-5)
  • (modified) llvm/test/CodeGen/X86/fold-pcmpeqd-2.ll (+8-2)
  • (modified) llvm/test/CodeGen/X86/var-permute-256.ll (+90-76)
  • (modified) llvm/test/CodeGen/X86/vector-interleaved-load-i8-stride-4.ll (+705-713)
  • (modified) llvm/test/CodeGen/X86/vector-interleaved-load-i8-stride-6.ll (+1500-1529)
  • (modified) llvm/test/CodeGen/X86/vector-shift-ashr-128.ll (+1-1)
  • (modified) llvm/test/CodeGen/X86/vector-shift-ashr-256.ll (+1-1)
  • (modified) llvm/test/CodeGen/X86/vector-shift-lshr-128.ll (+1-1)
  • (modified) llvm/test/CodeGen/X86/vector-shift-lshr-256.ll (+1-1)
  • (modified) llvm/test/CodeGen/X86/vector-shift-shl-128.ll (+1-1)
  • (modified) llvm/test/CodeGen/X86/vector-shift-shl-256.ll (+1-1)
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 51ae8b703e50f..5be1892a44f6d 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -2881,6 +2881,12 @@ bool SelectionDAG::isSplatValue(SDValue V, const APInt &DemandedElts,
   }
   }
 
+  // Fallback - this is a splat if all demanded elts are the same constant.
+  if (computeKnownBits(V, DemandedElts, Depth).isConstant()) {
+    UndefElts = ~DemandedElts;
+    return true;
+  }
+
   return false;
 }
 
diff --git a/llvm/test/CodeGen/ARM/vector-store.ll b/llvm/test/CodeGen/ARM/vector-store.ll
index a8a1031637afc..9c8ea7a2c440c 100644
--- a/llvm/test/CodeGen/ARM/vector-store.ll
+++ b/llvm/test/CodeGen/ARM/vector-store.ll
@@ -403,17 +403,14 @@ define void @v3i8store(ptr %p) {
 ; CHECK-LABEL: v3i8store:
 ; CHECK:       @ %bb.0:
 ; CHECK-NEXT:    sub sp, #4
-; CHECK-NEXT:    vmov.i32 d16, #0xff
-; CHECK-NEXT:    mov r1, sp
-; CHECK-NEXT:    vmov.i32 d17, #0x0
-; CHECK-NEXT:    movs r2, #0
-; CHECK-NEXT:    vand d16, d17, d16
-; CHECK-NEXT:    vst1.32 {d16[0]}, [r1:32]
-; CHECK-NEXT:    vld1.32 {d16[0]}, [r1:32]
+; CHECK-NEXT:    movs r1, #0
+; CHECK-NEXT:    mov r2, sp
+; CHECK-NEXT:    str r1, [sp]
+; CHECK-NEXT:    vld1.32 {d16[0]}, [r2:32]
+; CHECK-NEXT:    strb r1, [r0, #2]
 ; CHECK-NEXT:    vmovl.u16 q8, d16
-; CHECK-NEXT:    strb r2, [r0, #2]
-; CHECK-NEXT:    vmov.32 r1, d16[0]
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vmov.32 r2, d16[0]
+; CHECK-NEXT:    strh r2, [r0]
 ; CHECK-NEXT:    add sp, #4
 ; CHECK-NEXT:    bx lr
   store <3 x i8> zeroinitializer, ptr %p, align 4
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
index a56a81f5f793b..769bfe8cd5ba9 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
@@ -244,14 +244,14 @@ define <8 x i64> @vrgather_shuffle_vx_v8i64(<8 x i64> %x) {
 ; RV32-NEXT:    addi a0, a0, %lo(.LCPI13_0)
 ; RV32-NEXT:    vsetivli zero, 8, e64, m4, ta, mu
 ; RV32-NEXT:    vle16.v v16, (a0)
-; RV32-NEXT:    vrgatherei16.vv v12, v8, v16
+; RV32-NEXT:    vmv.v.i v20, 5
 ; RV32-NEXT:    lui a0, %hi(.LCPI13_1)
 ; RV32-NEXT:    addi a0, a0, %lo(.LCPI13_1)
-; RV32-NEXT:    vle16.v v8, (a0)
-; RV32-NEXT:    li a0, 140
+; RV32-NEXT:    vle16.v v17, (a0)
+; RV32-NEXT:    li a0, 115
 ; RV32-NEXT:    vmv.s.x v0, a0
-; RV32-NEXT:    vmv.v.i v16, 5
-; RV32-NEXT:    vrgatherei16.vv v12, v16, v8, v0.t
+; RV32-NEXT:    vrgatherei16.vv v12, v20, v16
+; RV32-NEXT:    vrgatherei16.vv v12, v8, v17, v0.t
 ; RV32-NEXT:    vmv.v.v v8, v12
 ; RV32-NEXT:    ret
 ;
diff --git a/llvm/test/CodeGen/X86/fold-pcmpeqd-2.ll b/llvm/test/CodeGen/X86/fold-pcmpeqd-2.ll
index 88425ea87845d..5d563b141ed66 100644
--- a/llvm/test/CodeGen/X86/fold-pcmpeqd-2.ll
+++ b/llvm/test/CodeGen/X86/fold-pcmpeqd-2.ll
@@ -54,7 +54,10 @@ define void @program_1(ptr %dest, ptr %t0, <4 x float> %p0, <4 x float> %p1, <4
 ; X32-NEXT:    xorps %xmm0, %xmm0
 ; X32-NEXT:    movaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
 ; X32-NEXT:    movaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; X32-NEXT:    mulps %xmm0, %xmm0
+; X32-NEXT:    mulss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
+; X32-NEXT:    movaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; X32-NEXT:    movaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; X32-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; X32-NEXT:    movaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
 ; X32-NEXT:    movaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
 ; X32-NEXT:    mulps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
@@ -138,7 +141,10 @@ define void @program_1(ptr %dest, ptr %t0, <4 x float> %p0, <4 x float> %p1, <4
 ; X64-NEXT:    xorps %xmm0, %xmm0
 ; X64-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
 ; X64-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 ## 16-byte Reload
-; X64-NEXT:    mulps %xmm0, %xmm0
+; X64-NEXT:    mulss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; X64-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
+; X64-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 ## 16-byte Reload
+; X64-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; X64-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
 ; X64-NEXT:    movaps (%rsp), %xmm0 ## 16-byte Reload
 ; X64-NEXT:    mulps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
diff --git a/llvm/test/CodeGen/X86/var-permute-256.ll b/llvm/test/CodeGen/X86/var-permute-256.ll
index 6c07c4ca523f8..853d8a278960d 100644
--- a/llvm/test/CodeGen/X86/var-permute-256.ll
+++ b/llvm/test/CodeGen/X86/var-permute-256.ll
@@ -25,18 +25,20 @@ define <4 x i64> @var_shuffle_v4i64(<4 x i64> %v, <4 x i64> %indices) nounwind {
 ;
 ; AVX1-LABEL: var_shuffle_v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
-; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpaddq %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [2,2]
+; AVX1-NEXT:    # xmm3 = mem[0,0]
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm2, %xmm4
 ; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm4
-; AVX1-NEXT:    vpermilpd %ymm4, %ymm2, %ymm2
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm0[2,3,2,3]
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm4, %ymm2
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
-; AVX1-NEXT:    vpermilpd %ymm4, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
-; AVX1-NEXT:    vblendvpd %ymm1, %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vblendvpd %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_shuffle_v4i64:
@@ -88,15 +90,16 @@ define <8 x i32> @var_shuffle_v8i32(<8 x i32> %v, <8 x i32> %indices) nounwind {
 ;
 ; AVX1-LABEL: var_shuffle_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
-; AVX1-NEXT:    vpermilps %ymm1, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3,3,3,3]
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
+; AVX1-NEXT:    vpermilps %ymm1, %ymm3, %ymm3
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
-; AVX1-NEXT:    vblendvps %ymm1, %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vblendvps %ymm2, %ymm3, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; INT256-LABEL: var_shuffle_v8i32:
@@ -445,18 +448,20 @@ define <4 x double> @var_shuffle_v4f64(<4 x double> %v, <4 x i64> %indices) noun
 ;
 ; AVX1-LABEL: var_shuffle_v4f64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
-; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpaddq %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [2,2]
+; AVX1-NEXT:    # xmm3 = mem[0,0]
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm2, %xmm4
 ; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm4
-; AVX1-NEXT:    vpermilpd %ymm4, %ymm2, %ymm2
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm0[2,3,2,3]
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm4, %ymm2
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
-; AVX1-NEXT:    vpermilpd %ymm4, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
-; AVX1-NEXT:    vblendvpd %ymm1, %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vblendvpd %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_shuffle_v4f64:
@@ -508,15 +513,16 @@ define <8 x float> @var_shuffle_v8f32(<8 x float> %v, <8 x i32> %indices) nounwi
 ;
 ; AVX1-LABEL: var_shuffle_v8f32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
-; AVX1-NEXT:    vpermilps %ymm1, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3,3,3,3]
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
+; AVX1-NEXT:    vpermilps %ymm1, %ymm3, %ymm3
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
-; AVX1-NEXT:    vblendvps %ymm1, %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vblendvps %ymm2, %ymm3, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; INT256-LABEL: var_shuffle_v8f32:
@@ -569,17 +575,19 @@ define <4 x i64> @var_shuffle_v4i64_from_v2i64(<2 x i64> %v, <4 x i64> %indices)
 ; AVX1-LABEL: var_shuffle_v4i64_from_v2i64:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
-; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpaddq %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [2,2]
+; AVX1-NEXT:    # xmm3 = mem[0,0]
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm2, %xmm4
 ; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm3
-; AVX1-NEXT:    vpermilpd %ymm3, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; AVX1-NEXT:    vpermilpd %ymm3, %ymm0, %ymm2
-; AVX1-NEXT:    vblendvpd %ymm1, %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm1
+; AVX1-NEXT:    vblendvpd %ymm3, %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_shuffle_v4i64_from_v2i64:
@@ -633,14 +641,15 @@ define <8 x i32> @var_shuffle_v8i32_from_v4i32(<4 x i32> %v, <8 x i32> %indices)
 ; AVX1-LABEL: var_shuffle_v8i32_from_v4i32:
 ; AVX1:       # %bb.0: # %entry
 ; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3,3,3,3]
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
+; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm3
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
-; AVX1-NEXT:    vblendvps %ymm1, %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vblendvps %ymm2, %ymm3, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; INT256-LABEL: var_shuffle_v8i32_from_v4i32:
@@ -990,17 +999,19 @@ define <4 x double> @var_shuffle_v4f64_from_v2f64(<2 x double> %v, <4 x i64> %in
 ; AVX1-LABEL: var_shuffle_v4f64_from_v2f64:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
-; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpaddq %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [2,2]
+; AVX1-NEXT:    # xmm3 = mem[0,0]
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm2, %xmm4
 ; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm3
-; AVX1-NEXT:    vpermilpd %ymm3, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; AVX1-NEXT:    vpermilpd %ymm3, %ymm0, %ymm2
-; AVX1-NEXT:    vblendvpd %ymm1, %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm1
+; AVX1-NEXT:    vblendvpd %ymm3, %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_shuffle_v4f64_from_v2f64:
@@ -1054,14 +1065,15 @@ define <8 x float> @var_shuffle_v8f32_from_v4f32(<4 x float> %v, <8 x i32> %indi
 ; AVX1-LABEL: var_shuffle_v8f32_from_v4f32:
 ; AVX1:       # %bb.0: # %entry
 ; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3,3,3,3]
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
+; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm3
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
-; AVX1-NEXT:    vblendvps %ymm1, %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vblendvps %ymm2, %ymm3, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; INT256-LABEL: var_shuffle_v8f32_from_v4f32:
@@ -1271,20 +1283,22 @@ define <4 x i64> @var_shuffle_v4i64_with_v16i8_indices(<4 x i64> %v, <16 x i8> %
 ;
 ; AVX1-LABEL: var_shuffle_v4i64_with_v16i8_indices:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpsrld $16, %xmm1, %xmm2
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
 ; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [2,2]
+; AVX1-NEXT:    # xmm3 = mem[0,0]
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm1, %xmm4
 ; AVX1-NEXT:    vpaddq %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm4
-; AVX1-NEXT:    vpermilpd %ymm4, %ymm3, %ymm3
+; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm2, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm0[2,3,2,3]
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm4, %ymm2
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
-; AVX1-NEXT:    vpermilpd %ymm4, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtq {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX1-NEXT:    vblendvpd %ymm1, %ymm3, %ymm0, %ymm0
+; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vblendvpd %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_shuffle_v4i64_with_v16i8_indices:
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i8-stride-4.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i8-stride-4.ll
index c9cbb0994810f..e2195f1fc25a1 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i8-stride-4.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i8-stride-4.ll
@@ -156,64 +156,62 @@ define void @load_i8_stride4_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 define void @load_i8_stride4_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr %out.vec2, ptr %out.vec3) nounwind {
 ; SSE-LABEL: load_i8_stride4_vf8:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa (%rdi), %xmm1
+; SSE-NEXT:    movdqa (%rdi), %xmm3
 ; SSE-NEXT:    movdqa 16(%rdi), %xmm4
 ; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,255,0,255,0,255,0]
+; SSE-NEXT:    movdqa %xmm4, %xmm1
 ; SSE-NEXT:    movdqa %xmm4, %xmm2
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    packuswb %xmm2, %xmm0
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm4[3,1,2,3,4,5,6,7]
+; SSE-NEXT:    pand %xmm0, %xmm4
+; SSE-NEXT:    pand %xmm3, %xmm0
+; SSE-NEXT:    packuswb %xmm4, %xmm0
 ; SSE-NEXT:    packuswb %xmm0, %xmm0
-; SSE-NEXT:    pxor %xmm7, %xmm7
-; SSE-NEXT:    movdqa %xmm4, %xmm2
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm7[8],xmm2[9],xmm7[9],xmm2[10],xmm7[10],xmm2[11],xmm7[11],xmm2[12],xmm7[12],xmm2[13],xmm7[13],xmm2[14],xmm7[14],xmm2[15],xmm7[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,2,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm3[0,1,1,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm4, %xmm3
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm7[0],xmm3[1],xmm7[1],xmm3[2],xmm7[2],xmm3[3],xmm7[3],xmm3[4],xmm7[4],xmm3[5],xmm7[5],xmm3[6],xmm7[6],xmm3[7],xmm7[7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[0,2,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm6[0,1,1,3,4,5,6,7]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm8 = xmm8[0],xmm5[0],xmm8[1],xmm5[1]
-; SSE-NEXT:    movdqa %xmm1, %xmm5
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm7[8],xmm5[9],xmm7[9],xmm5[10],xmm7[10],xmm5[11],xmm7[11],xmm5[12],xmm7[12],xmm5[13],xmm7[13],xmm5[14],xmm7[14],xmm5[15],xmm7[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm5[0,2,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm6[1,3,2,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm1, %xmm6
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1],xmm6[2],xmm7[2],xmm6[3],xmm7[3],xmm6[4],xmm7[4],xmm6[5],xmm7[5],xmm6[6],xmm7[6],xmm6[7],xmm7[7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm6[0,2,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[1,3,2,3,4,5,6,7]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm9[0],xmm7[1],xmm9[1]
-; SSE-NEXT:    packuswb %xmm8, %xmm7
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,3,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm8, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,7,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,2,0]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,7,6,5,4]
-; SSE-NEXT:    pand %xmm8, %xmm1
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[1,0,3,2,4,5,6,7]
-; SSE-NEXT:    packuswb %xmm4, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,3,2,3]
+; SSE-NEXT:    pxor %xmm6, %xmm6
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm6[8],xmm1[9],xmm6[9],xmm1[10],xmm6[10],xmm1[11],xmm6[11],xmm1[12],xmm6[12],xmm1[13],xmm6[13]...
[truncated]

phoebewang
phoebewang reviewed Dec 5, 2023
View reviewed changes
llvm/test/CodeGen/X86/fold-pcmpeqd-2.ll Outdated Show resolved Hide resolved
llvm/test/CodeGen/X86/fold-pcmpeqd-2.ll Outdated Show resolved Hide resolved
llvm/test/CodeGen/X86/var-permute-256.ll Show resolved Hide resolved
@RKSimon RKSimon marked this pull request as draft December 6, 2023 17:22
@RKSimon RKSimon force-pushed the perf/splat-constant branch 2 times, most recently from 0b9a603 to ec63552 Compare December 7, 2023 15:24
@RKSimon RKSimon marked this pull request as ready for review December 7, 2023 15:24
phoebewang
phoebewang approved these changes Dec 8, 2023
View reviewed changes
Copy link
Contributor

@phoebewang phoebewang left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM.

@RKSimon RKSimon merged commit faecc73 into llvm:main Dec 8, 2023
3 of 4 checks passed
@RKSimon RKSimon deleted the perf/splat-constant branch December 8, 2023 10:53
RKSimon added a commit that referenced this pull request Jan 17, 2024
@RKSimon
Revert faecc73 #74443 [DAG] isSplatValue - node is a splat if all dem…
d92ce34 
…anded elts have the same whole constant value (#74443)

Relying on ComputeKnownBits to find a splat is causing miscompilations where a shift of zero is being assumed to give zero, but further simplification leads to a shift of zero by undef, resulting in an unexpected undef value.

Fixes #78109
ampandey-1995 pushed a commit to ampandey-1995/llvm-project that referenced this pull request Jan 19, 2024
@RKSimon @ampandey-1995
Revert faecc73 llvm#74443 [DAG] isSplatValue - node is a splat if all…
f50f825 
… demanded elts have the same whole constant value (llvm#74443)

Relying on ComputeKnownBits to find a splat is causing miscompilations where a shift of zero is being assumed to give zero, but further simplification leads to a shift of zero by undef, resulting in an unexpected undef value.

Fixes llvm#78109
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@phoebewang phoebewang phoebewang approved these changes

@preames preames Awaiting requested review from preames

@topperc topperc Awaiting requested review from topperc

@goldsteinn goldsteinn Awaiting requested review from goldsteinn

Assignees
No one assigned
Labels
backend:ARM backend:X86 llvm:SelectionDAG SelectionDAGISel as well
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
3 participants
@RKSimon @llvmbot @phoebewang