Wrong immediate generated by SPIR-V backend

[jdnk]

The SPIR-V backend generates -1 instead of 0xFFFFFFFF in OpVectorShuffle instruction.

Original C source:

// repr_minus_one.c

#include <stdint.h>

typedef union {
  uint32_t bits;
  float value;
} fp32bits;

float fp16mul(float *restrict s, uint16_t a, uint16_t b) {
  const fp32bits af = {.bits = (a << 16)};
  const fp32bits bf = {.bits = (b << 16)};

  return af.value * bf.value;
}

clang -S -emit-llvm -O3 repr_minus_one.c -o repr_minus_one.ll generates:

; repr_minus_one.ll

; ModuleID = 'repr_minus_one.c'
source_filename = "repr_minus_one.c"
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

; Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none) uwtable
define dso_local float @fp16mul(ptr noalias nocapture noundef readnone %s, i16 noundef zeroext %a, i16 noundef zeroext %b) local_unnamed_addr #0 {
entry:
  %0 = insertelement <2 x i16> poison, i16 %a, i64 0
  %1 = insertelement <2 x i16> %0, i16 %b, i64 1
  %2 = zext <2 x i16> %1 to <2 x i32>
  %3 = shl nuw <2 x i32> %2, <i32 16, i32 16>
  %4 = bitcast <2 x i32> %3 to <2 x float>
  %shift = shufflevector <2 x float> %4, <2 x float> poison, <2 x i32> <i32 1, i32 poison>
  %5 = fmul <2 x float> %shift, %4
  %mul = extractelement <2 x float> %5, i64 0
  ret float %mul
}

attributes #0 = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) uwtable "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cmov,+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }

!llvm.module.flags = !{!0, !1, !2, !3}
!llvm.ident = !{!4}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{i32 8, !"PIC Level", i32 2}
!2 = !{i32 7, !"PIE Level", i32 2}
!3 = !{i32 7, !"uwtable", i32 2}
!4 = !{!"clang version 19.1.6 (https://github.com/llvm/llvm-project.git e21dc4bd5474d04b8e62d7331362edcc5648d7e5)"}

llc -mtriple=spirv64 repr_minus_one.ll -o repr_minus_one.spt generates:

; repr_minus_one.spt

        OpCapability Kernel
        OpCapability Addresses
        OpCapability Int8
        OpCapability Int16
        OpCapability Linkage
        OpCapability Int64
        %1 = OpExtInstImport "OpenCL.std"
        OpMemoryModel Physical64 OpenCL
        OpSource OpenCL_CPP 100000
        OpName %16 "s"
        OpName %17 "a"
        OpName %18 "b"
        OpName %19 "fp16mul"
        OpName %25 "shift"
        OpName %27 "mul"
        OpDecorate %16 FuncParamAttr NoAlias
        OpDecorate %17 FuncParamAttr Zext
        OpDecorate %18 FuncParamAttr Zext
        OpDecorate %19 LinkageAttributes "fp16mul" Export
        %2 = OpTypeInt 8 0
        %3 = OpTypeInt 16 0
        %4 = OpTypeFloat 32
        %5 = OpTypePointer Function %2
        %6 = OpTypeFunction %4 %5 %3 %3
        %7 = OpTypeVector %4 2
        %8 = OpTypeInt 32 0
        %9 = OpTypeVector %8 2
        %10 = OpTypeVector %3 2
        %11 = OpTypeInt 64 0
        %12 = OpUndef %10
        %13 = OpConstant %8 16
        %14 = OpConstantComposite %9 %13 %13
        %15 = OpUndef %7
        %19 = OpFunction %4 Pure %6             ; -- Begin function fp16mul
        %16 = OpFunctionParameter %5
        %17 = OpFunctionParameter %3
        %18 = OpFunctionParameter %3
        %28 = OpLabel
        %20 = OpCompositeInsert %10 %17 %12 0
        %21 = OpCompositeInsert %10 %18 %20 1
        %22 = OpUConvert %9 %21
        %23 = OpShiftLeftLogical %9 %22 %14
        %24 = OpBitcast %7 %23
        %25 = OpVectorShuffle %7 %24 %15 1 -1  ;;; <-- here
        %26 = OpFMul %7 %25 %24
        %27 = OpCompositeExtract %4 %26 0
        OpReturnValue %27
        OpFunctionEnd
                                        ; -- End function

Towards the end, you can see %25 = OpVectorShuffle %7 %24 %15 1 -1, but the specification states that the Components literal of the OpVectorShuffle instruction must be an unsigned integer. I believe the constant 0xFFFFFFFF is supposed to be there instead of the -1.

Reproduced on LLVM 19.1.6, x86_64 (see repr_minus_one.ll for details).