AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

radv: fix OpQuantizeToF16 for NaN on GFX6-7

Browse Source 
Do not flush NaN to 0.

Fixes
dEQP-VK.spirv_assembly.instruction.compute.opquantize.propagated_nans

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This commit is contained in:
Samuel Pitoiset
2019-10-24 18:48:54 +02:00
parent d82dfca872
commit 5912792501
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/llvm/ac_nir_to_llvm.c
+2 -2
View File
@@ -356,10 +356,10 @@ static LLVMValueRef emit_f2f16(struct ac_llvm_context *ctx,
*/
LLVMValueRef temp, cond2;
temp = emit_intrin_1f_param(ctx, "llvm.fabs", ctx->f32, result);
cond = LLVMBuildFCmp(ctx->builder, LLVMRealUGT,
cond = LLVMBuildFCmp(ctx->builder, LLVMRealOGT,
LLVMBuildBitCast(ctx->builder, LLVMConstInt(ctx->i32, 0x38800000, false), ctx->f32, ""),
temp, "");
cond2 = LLVMBuildFCmp(ctx->builder, LLVMRealUNE,
cond2 = LLVMBuildFCmp(ctx->builder, LLVMRealONE,
temp, ctx->f32_0, "");
cond = LLVMBuildAnd(ctx->builder, cond, cond2, "");
result = LLVMBuildSelect(ctx->builder, cond, ctx->f32_0, result, "");