[llvm] 95661b9 - [X86] getTargetConstantBitsFromNode - support extracting fp data from ConstantDataSequential

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2023-05-30T11:38:31+01:00
New Revision: 95661b9c7545b56e6ec5a0cfec75587f37a7ca50

URL: https://github.com/llvm/llvm-project/commit/95661b9c7545b56e6ec5a0cfec75587f37a7ca50
DIFF: https://github.com/llvm/llvm-project/commit/95661b9c7545b56e6ec5a0cfec75587f37a7ca50.diff

LOG: [X86] getTargetConstantBitsFromNode - support extracting fp data from ConstantDataSequential

Fixes issue introduced by 0f8e0f4228805cbecce13dcfadef4c48a4f0f4cd where SimplifyDemandedBits could crash when trying to extract fp data from broadcasted constants

Added: 
    

Modified: 
    llvm/lib/Target/X86/X86ISelLowering.cpp
    llvm/test/CodeGen/X86/avx-vbroadcast.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 9b433280d5d9d..a89ab94c9e0d7 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -7453,9 +7453,19 @@ static bool getTargetConstantBitsFromNode(SDValue Op, unsigned EltSizeInBits,
     if (auto *CDS = dyn_cast<ConstantDataSequential>(Cst)) {
       Type *Ty = CDS->getType();
       Mask = APInt::getZero(Ty->getPrimitiveSizeInBits());
-      unsigned EltBits = CDS->getElementType()->getPrimitiveSizeInBits();
+      Type *EltTy = CDS->getElementType();
+      bool IsInteger = EltTy->isIntegerTy();
+      bool IsFP =
+          EltTy->isHalfTy() || EltTy->isFloatTy() || EltTy->isDoubleTy();
+      if (!IsInteger && !IsFP)
+        return false;
+      unsigned EltBits = EltTy->getPrimitiveSizeInBits();
       for (unsigned I = 0, E = CDS->getNumElements(); I != E; ++I)
-        Mask.insertBits(CDS->getElementAsAPInt(I), I * EltBits);
+        if (IsInteger)
+          Mask.insertBits(CDS->getElementAsAPInt(I), I * EltBits);
+        else
+          Mask.insertBits(CDS->getElementAsAPFloat(I).bitcastToAPInt(),
+                          I * EltBits);
       return true;
     }
     return false;

diff  --git a/llvm/test/CodeGen/X86/avx-vbroadcast.ll b/llvm/test/CodeGen/X86/avx-vbroadcast.ll
index 54bce767f1fcc..b442a6337e3b8 100644
--- a/llvm/test/CodeGen/X86/avx-vbroadcast.ll
+++ b/llvm/test/CodeGen/X86/avx-vbroadcast.ll
@@ -914,6 +914,48 @@ define double @broadcast_scale_xyz(ptr nocapture readonly, ptr nocapture readonl
   ret double %14
 }
 
+;
+; Broadcast v2f32 non-uniform constant via vmovddup
+;
+define void @fmul_by_v2f32_broadcast() nounwind {
+; X86-LABEL: fmul_by_v2f32_broadcast:
+; X86:       ## %bb.0:
+; X86-NEXT:    vmovddup {{.*#+}} xmm0 = [3.1E+1,0.0E+0,3.1E+1,0.0E+0]
+; X86-NEXT:    ## xmm0 = mem[0,0]
+; X86-NEXT:    ## implicit-def: $xmm1
+; X86-NEXT:    .p2align 4, 0x90
+; X86-NEXT:  LBB42_1: ## =>This Inner Loop Header: Depth=1
+; X86-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
+; X86-NEXT:    vmulps %xmm0, %xmm2, %xmm2
+; X86-NEXT:    vmovlps %xmm2, (%eax)
+; X86-NEXT:    vmulps %xmm0, %xmm1, %xmm1
+; X86-NEXT:    vmovlps %xmm1, (%eax)
+; X86-NEXT:    jmp LBB42_1
+;
+; X64-LABEL: fmul_by_v2f32_broadcast:
+; X64:       ## %bb.0:
+; X64-NEXT:    vmovddup {{.*#+}} xmm0 = [3.1E+1,0.0E+0,3.1E+1,0.0E+0]
+; X64-NEXT:    ## xmm0 = mem[0,0]
+; X64-NEXT:    ## implicit-def: $xmm1
+; X64-NEXT:    .p2align 4, 0x90
+; X64-NEXT:  LBB42_1: ## =>This Inner Loop Header: Depth=1
+; X64-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
+; X64-NEXT:    vmulps %xmm0, %xmm2, %xmm2
+; X64-NEXT:    vmovlps %xmm2, (%rax)
+; X64-NEXT:    vmulps %xmm0, %xmm1, %xmm1
+; X64-NEXT:    vmovlps %xmm1, (%rax)
+; X64-NEXT:    jmp LBB42_1
+  br label %1
+1:
+  %2 = phi <2 x float> [ undef, %0 ], [ %5, %1 ]
+  %3 = load <2 x float>, ptr poison, align 8
+  %4 = fmul <2 x float> %3, <float 3.100000e+01, float 0.000000e+00>
+  store <2 x float> %4, ptr poison, align 8
+  %5 = fmul <2 x float> %2, <float 3.100000e+01, float 0.000000e+00>
+  store <2 x float> %5, ptr poison, align 8
+  br label %1
+}
+
 ;
 ; When VBROADCAST replaces an existing load, ensure it still respects lifetime dependencies.
 ;