Clip fp8 to +/-240 on all targets. (#1172)

* clip fp8 to +/-240 on all targets

* if inputs to fp8 conversion are +/-inf, they remain unaltered

* increase tolerance for test_elementwise_layernorm to prevent false errors

* change the input values for gemm examples to floats

* reduce gemm example float input values to prevent errors

* increase the tolerance for gemm examples

[ROCm/composable_kernel commit: d0c7b45150]
This commit is contained in:
Illia Silin
2024-02-27 12:31:05 -08:00
committed by GitHub
parent d2750747d4
commit f0f3b65e2a
4 changed files with 16 additions and 13 deletions

View File

@@ -49,7 +49,7 @@ struct ProblemSizeStreamK final
struct ExecutionConfig final
{
bool do_verification = true;
int init_method = 1;
int init_method = 2;
bool time_kernel = false;
};

View File

@@ -69,8 +69,8 @@ bool run_gemm(const ProblemType& problem_size, const ExecutionConfig& config)
ck::utils::FillUniformDistributionIntegerValue<BDataType>{-5.f, 5.f}(b_k_n);
break;
default:
ck::utils::FillUniformDistribution<ADataType>{-1.f, 1.f}(a_m_k);
ck::utils::FillUniformDistribution<BDataType>{-1.f, 1.f}(b_k_n);
ck::utils::FillUniformDistribution<ADataType>{-0.1f, 0.1f}(a_m_k);
ck::utils::FillUniformDistribution<BDataType>{-0.1f, 0.1f}(b_k_n);
}
Tensor<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
@@ -240,7 +240,8 @@ bool run_gemm(const ProblemType& problem_size, const ExecutionConfig& config)
#else
c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
return ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
return ck::utils::check_err(
c_m_n_device_result, c_m_n_host_result, "Error: Incorrect results!", 1e-1, 1e-1);
#endif
}