[llvm] [SPIRV] Added Support for the constrained conversion intrinsics (PR #157437)

[Subash B via llvm-commits]

================
@@ -0,0 +1,41 @@
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv64-unknown-unknown %s -o - | FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv64-unknown-unknown %s -o - -filetype=obj | spirv-val %}
+
+; CHECK-DAG: OpName %[[#sf:]] "conv"
+; CHECK-DAG: OpName %[[#uf:]] "conv1"
+; CHECK-DAG: OpName %[[#fs:]] "conv2"
+; CHECK-DAG: OpName %[[#fu:]] "conv3"
+; CHECK-DAG: OpName %[[#fe:]] "conv4"
+; CHECK-DAG: OpName %[[#ft:]] "conv5"
+
+; CHECK-DAG: OpDecorate %[[#sf]] FPRoundingMode RTE
+; CHECK-DAG: OpDecorate %[[#uf]] FPRoundingMode RTZ
+; CHECK-DAG: OpDecorate %[[#ft]] FPRoundingMode RTP
+; CHECK-NOT: OpDecorate %[[#fs]] FPRoundingMode
+; CHECK-NOT: OpDecorate %[[#fu]] FPRoundingMode
+; CHECK-NOT: OpDecorate %[[#fe]] FPRoundingMode
+
+; CHECK: %[[#sf]] = OpConvertSToF
+; CHECK: %[[#uf]] = OpConvertUToF
+; CHECK: %[[#fs]] = OpConvertFToS
+; CHECK: %[[#fu]] = OpConvertFToU
+; CHECK: %[[#fe]] = OpFConvert
+; CHECK: %[[#ft]] = OpFConvert
+
+define dso_local spir_kernel void @test(float %a, i32 %in, i32 %ui) {
+entry:
+  %conv = tail call float @llvm.experimental.constrained.sitofp.f32.i32(i32 %in, metadata !"round.tonearest", metadata !"fpexcept.strict") 
----------------
SubashBoopathi wrote:

I’ve updated the test as requested:
1. Split the test into one case per function.
2. Added coverage for different type combinations and vector variants.
3. Simplified the RUN lines by removing -verify-machineinstrs and -O0.

https://github.com/llvm/llvm-project/pull/157437