[SPIRV] Avoid OpQuantizeToF16 in SPIR-V kernel test

[YixingZhang007]

This PR resolves the current failure in the integer-casts.ll SPIR-V test during CI runs in llvm-project.
The failure occurs because the SPIR-V instruction OpQuantizeToF16 requires the Capability::Shader. However, the function in integer-casts.ll is written as a kernel function and executed in a kernel environment. Therefore, Capability::Kernel is emitted instead of Capability::Shader. To fix this, we remove the QuantizeToF16 test frominteger-casts.ll in this PR.

[llvmbot]

@llvm/pr-subscribers-backend-spir-v

Author: None (YixingZhang007)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/158086.diff

1 Files Affected:

• (modified) llvm/test/CodeGen/SPIRV/instructions/integer-casts.ll (-3)

diff --git a/llvm/test/CodeGen/SPIRV/instructions/integer-casts.ll b/llvm/test/CodeGen/SPIRV/instructions/integer-casts.ll
index 6a4b4f593bf3b..667c8adfdba4b 100644
--- a/llvm/test/CodeGen/SPIRV/instructions/integer-casts.ll
+++ b/llvm/test/CodeGen/SPIRV/instructions/integer-casts.ll
@@ -258,7 +258,6 @@ define <4 x i32>  @u16tou32v4(<4 x i16> %a) {
 ; CHECK: %[[#]] = OpUConvert [[U32]] %[[#]]
 ; CHECK: %[[#]] = OpSConvert [[U32]] %[[#]]
 ; CHECK: %[[#]] = OpFConvert [[F16]] %[[#]]
-; CHECK: %[[#]] = OpQuantizeToF16 [[F32]] %[[#]]
 ; CHECK: %[[#]] = OpSatConvertSToU [[U64]] %[[#]]
 ; CHECK: %[[#]] = OpSatConvertUToS [[U64]] %[[#]]
 ; CHECK: %[[#]] = OpConvertPtrToU [[U64]] [[Arg1]]
@@ -281,7 +280,6 @@ define dso_local spir_kernel void @test_wrappers(ptr addrspace(4) %arg, i64 %arg
   %r5 = call spir_func i32 @__spirv_UConvert(i64 1)
   %r6 = call spir_func i32 @__spirv_SConvert(i64 1)
   %r7 = call spir_func half @__spirv_FConvert(float 0.000000e+00)
-  %r8 = call spir_func float @__spirv_QuantizeToF16(float 0.000000e+00)
   %r9 = call spir_func i64 @__spirv_SatConvertSToU(i64 1)
   %r10 = call spir_func i64 @__spirv_SatConvertUToS(i64 1)
   %r11 = call spir_func i64 @__spirv_ConvertPtrToU(ptr addrspace(4) %arg)
@@ -305,7 +303,6 @@ declare dso_local spir_func float @__spirv_ConvertUToF(i32)
 declare dso_local spir_func i32 @__spirv_UConvert(i64)
 declare dso_local spir_func i32 @__spirv_SConvert(i64)
 declare dso_local spir_func half @__spirv_FConvert(float)
-declare dso_local spir_func float @__spirv_QuantizeToF16(float)
 declare dso_local spir_func i64 @__spirv_SatConvertSToU(i64)
 declare dso_local spir_func i64 @__spirv_SatConvertUToS(i64)
 declare dso_local spir_func i64 @__spirv_ConvertPtrToU(ptr addrspace(4))

[YixingZhang007]

After this change, we no longer have test coverage for QuantizeToF16, since there is not yet sufficient support for running it in a Shader environment. Tests for QuantizeToF16 will be added once Shader support for this feature is implemented.

[maarquitos14]

LGTM. Can we spin off the OpQuantizeToF16 test to a new test and mark as xfail/unsupported, so we don't reduce test coverage?

[YixingZhang007]

LGTM. Can we spin off the OpQuantizeToF16 test to a new test and mark as xfail/unsupported, so we don't reduce test coverage?

Thanks for the suggestion! I have added the test for OpQuantizeToF16 in quantizeto16.ll and marked it as XFAIL.