[clang] [llvm] [HLSL] Implement `WaveReadLaneAt` intrinsic (PR #111010)

[Steven Perron via cfe-commits]

================
@@ -0,0 +1,28 @@
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv32v1.3-vulkan-unknown %s -o - | FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv-vulkan-unknown %s -o - -filetype=obj | spirv-val %}
+
+; Test lowering to spir-v backend
+
+; CHECK-DAG:   %[[#uint:]] = OpTypeInt 32 0
+; CHECK-DAG:   %[[#scope:]] = OpConstant %[[#uint]] 3
+; CHECK-DAG:   %[[#f32:]] = OpTypeFloat 32
+; CHECK-DAG:   %[[#expr:]] = OpFunctionParameter %[[#f32]]
+; CHECK-DAG:   %[[#idx:]] = OpFunctionParameter %[[#uint]]
+
+define spir_func void @test_1(float %expr, i32 %idx) #0 {
+entry:
+  %0 = call token @llvm.experimental.convergence.entry()
+; CHECK:   %[[#ret:]] = OpGroupNonUniformShuffle %[[#f32]] %[[#expr]] %[[#idx]] %[[#scope]]
+  %1 = call float @llvm.spv.wave.read.lane.at(float %expr, i32 %idx) [ "convergencectrl"(token %0) ]
----------------
s-perron wrote:

Could you add tests that check a few different expression/return types? It would be good to make sure that ti does not always return a float. It does not have to be as extensive as the DX test. Maybe just an 32-bit int and a vector of some sort.

https://github.com/llvm/llvm-project/pull/111010