[Mesa-dev] [PATCH 5/5] ac: use correct LLVM opcodes for ordered comparisons

[Timothy Arceri]

On 15/02/18 04:39, Marek Olšák wrote:
> Reviewed-by: Marek Olšák <marek.olsak at amd.com>
> 
> Marek
> 
> On Wed, Feb 14, 2018 at 7:29 AM, Timothy Arceri <tarceri at itsqueeze.com> wrote:
>> Fixes glsl-1.30/execution/isinf-and-isnan* piglit tests for
>> radeonsi and should fix SPIRV errors when LLVM optimises away
>> the workarounds in vtn_handle_alu() for handling ordered
>> comparisons.
>>
>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104905
>> ---
>>   src/amd/common/ac_nir_to_llvm.c | 8 ++++----
>>   1 file changed, 4 insertions(+), 4 deletions(-)
>>
>> diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
>> index a0c5680205..e81f86bb08 100644
>> --- a/src/amd/common/ac_nir_to_llvm.c
>> +++ b/src/amd/common/ac_nir_to_llvm.c
>> @@ -1792,16 +1792,16 @@ static void visit_alu(struct ac_nir_context *ctx, const nir_alu_instr *instr)
>>                  result = emit_int_cmp(&ctx->ac, LLVMIntUGE, src[0], src[1]);
>>                  break;
>>          case nir_op_feq:
>> -               result = emit_float_cmp(&ctx->ac, LLVMRealUEQ, src[0], src[1]);
>> +               result = emit_float_cmp(&ctx->ac, LLVMRealOEQ, src[0], src[1]);
>>                  break;
>>          case nir_op_fne:
>> -               result = emit_float_cmp(&ctx->ac, LLVMRealUNE, src[0], src[1]);
>> +               result = emit_float_cmp(&ctx->ac, LLVMRealONE, src[0], src[1]);

It seems we need to leave this one as is to avoid regressions. This is 
also what radeonsi does.

>>                  break;
>>          case nir_op_flt:
>> -               result = emit_float_cmp(&ctx->ac, LLVMRealULT, src[0], src[1]);
>> +               result = emit_float_cmp(&ctx->ac, LLVMRealOLT, src[0], src[1]);
>>                  break;
>>          case nir_op_fge:
>> -               result = emit_float_cmp(&ctx->ac, LLVMRealUGE, src[0], src[1]);
>> +               result = emit_float_cmp(&ctx->ac, LLVMRealOGE, src[0], src[1]);
>>                  break;
>>          case nir_op_ufeq:
>>                  result = emit_float_cmp(&ctx->ac, LLVMRealUEQ, src[0], src[1]);
>> --
>> 2.14.3
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev