[llvm] b7795eb - [X86] Building constant vector which element type is half will cause assertion fail.

[via llvm-commits]

Author: Liu, Chen3
Date: 2021-08-24T14:34:30+08:00
New Revision: b7795eb6469cfb98c4e331dac8087ac0b2eef9cc

URL: https://github.com/llvm/llvm-project/commit/b7795eb6469cfb98c4e331dac8087ac0b2eef9cc
DIFF: https://github.com/llvm/llvm-project/commit/b7795eb6469cfb98c4e331dac8087ac0b2eef9cc.diff

LOG: [X86] Building constant vector which element type is half will cause assertion fail.

Fix assertion fail when building con constant vector which element type is half.

Differential Revision: https://reviews.llvm.org/D108612

Added: 
    llvm/test/CodeGen/X86/build_fp16_constant_vector.ll

Modified: 
    llvm/lib/Target/X86/X86ISelLowering.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 9f8e76c786f9a..5096d45ad10df 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -9006,7 +9006,9 @@ static Constant *getConstantVector(MVT VT, const APInt &SplatValue,
     APInt Val = SplatValue.extractBits(ScalarSize, ScalarSize * i);
     Constant *Const;
     if (VT.isFloatingPoint()) {
-      if (ScalarSize == 32) {
+      if (ScalarSize == 16) {
+        Const = ConstantFP::get(C, APFloat(APFloat::IEEEhalf(), Val));
+      } else if (ScalarSize == 32) {
         Const = ConstantFP::get(C, APFloat(APFloat::IEEEsingle(), Val));
       } else {
         assert(ScalarSize == 64 && "Unsupported floating point scalar size");

diff  --git a/llvm/test/CodeGen/X86/build_fp16_constant_vector.ll b/llvm/test/CodeGen/X86/build_fp16_constant_vector.ll
new file mode 100644
index 0000000000000..6cb449822145a
--- /dev/null
+++ b/llvm/test/CodeGen/X86/build_fp16_constant_vector.ll
@@ -0,0 +1,21 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512fp16 | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512fp16 | FileCheck %s
+
+define dso_local <32 x half> @foo(<32 x half> %a, <32 x half> %b, <32 x half> %c) {
+; CHECK-LABEL: foo:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vbroadcastf64x4 {{.*#+}} zmm3 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0,-1.0E+0]
+; CHECK-NEXT:    # zmm3 = mem[0,1,2,3,0,1,2,3]
+; CHECK-NEXT:    vfmadd213ph %zmm2, %zmm3, %zmm0
+; CHECK-NEXT:    vfmadd213ph %zmm2, %zmm3, %zmm1
+; CHECK-NEXT:    vaddph %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    ret{{[l|q]}}
+entry:
+  %0 = tail call fast <32 x half> @llvm.fma.v32f16(<32 x half> %a, <32 x half> <half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00>, <32 x half> %c)
+  %1 = tail call fast <32 x half> @llvm.fma.v32f16(<32 x half> %b, <32 x half> <half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xH3C00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00, half 0xHBC00>, <32 x half> %c)
+  %2 = fadd <32 x half> %0, %1
+  ret <32 x half> %2
+}
+
+declare <32 x half> @llvm.fma.v32f16(<32 x half>, <32 x half>, <32 x half>)