leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 25 Packages Projects Releases Wiki Activity

More updates for 3.1 (#958)

Browse Source 
* Updates for 3.1

* Minor change

* doc link fix

* Minor updates
This commit is contained in:
ANIKET SHIVAM
2023-05-24 10:17:16 -04:00
committed by GitHub
parent 13f413493a
commit f079619f5e
48 changed files with 1611 additions and 1858 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/library/src/reference/gemm.cu
-352
View File
@@ -345,358 +345,6 @@ void initialize_gemm_reference_operations(Manifest &manifest) {
complex<double>
>(manifest);
//
// FP8 GEMMs
//
//////////////////////////////////
/// ElementC: half_t
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e4m3_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
half_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e4m3_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float , // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e4m3_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e5m2_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float , // ElementAccumulator
half_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e5m2_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e5m2_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e4m3_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
half_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e4m3_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e4m3_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e5m2_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
half_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e5m2_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e5m2_t, // ElementB
half_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
/// ElementC: bfloat16_t
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e4m3_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
bfloat16_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e4m3_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e4m3_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e5m2_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
bfloat16_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e5m2_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e5m2_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e4m3_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
bfloat16_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e4m3_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e4m3_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e5m2_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
bfloat16_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e5m2_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e5m2_t, // ElementB
bfloat16_t, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
/// ElementC: float
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e4m3_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e4m3_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e4m3_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e5m2_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e5m2_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e4m3_t, // ElementA
float_e5m2_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e4m3_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e4m3_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e4m3_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
//////////////////////////////////
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e5m2_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e5m2_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e4m3_t // ElementD
>(manifest);
make_gemm_real_canonical_layouts<
float_e5m2_t, // ElementA
float_e5m2_t, // ElementB
float, // ElementC
float, // ElementScalar
float, // ElementAccumulator
float_e5m2_t // ElementD
>(manifest);
}
///////////////////////////////////////////////////////////////////////////////////////////////////