disabling 128 on non preshufflequant

This commit is contained in:
khuagarw
2025-12-19 01:49:22 +00:00
parent 5b4a67ec6d
commit 82fcd3c0bd
2 changed files with 32 additions and 34 deletions

View File

@@ -49,10 +49,10 @@ void bquant_quantgrouped_fp8_instance_factory(
using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 64, 128>>;
return RUN_GEMM_EXAMPLE_PREC_TYPE;
};
lut[hash_multiple_strings(
{"fp8", "bquant", "non-preshuffleb", "non-preshufflequant", "1x128x128"})] =
[](const ck_tile::ArgParser& arg_parser) {
using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 128, 128>>;
return RUN_GEMM_EXAMPLE_PREC_TYPE;
};
// lut[hash_multiple_strings(
// {"fp8", "bquant", "non-preshuffleb", "non-preshufflequant", "1x128x128"})] =
// [](const ck_tile::ArgParser& arg_parser) {
// using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 128, 128>>;
// return RUN_GEMM_EXAMPLE_PREC_TYPE;
// };
}

View File

@@ -4,7 +4,7 @@
#include "run_gemm_quant_example.inc"
template <typename T>
using GemmConfig = GemmConfigPreshuffleQuantDecode<T>;
using GemmConfig = GemmConfigPreshuffleBQuantPrefill<T>; // GemmConfigPreshuffleQuantDecode<T>;
// //GemmConfigPreshuffleBQuantPrefill<T>;
void bquant_quantgrouped_preshufflequant_instance_factory(
@@ -24,33 +24,31 @@ void bquant_quantgrouped_preshufflequant_instance_factory(
ck_tile::QuantType::BQuantGrouped>(arg_parser);
};
// lut[hash_multiple_strings({"fp8", "bquant", "non-preshuffleb", "preshufflequant",
// "1x8x128"})] =
// [](const ck_tile::ArgParser& arg_parser) {
// using TypeConfig = decltype(GemmQuantTypeConfig<ck_tile::fp8_t,
// ck_tile::fp8_t,
// ck_tile::half_t,
// float>{});
// using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 8, 128>>;
// return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
// TypeConfig,
// QuantGroupSize,
// ck_tile::QuantType::BQuantGrouped>(arg_parser);
// };
// lut[hash_multiple_strings({"fp8",
// "bquant",
// "non-preshuffleb",
// "preshufflequant",
// "1x16x128"})] = [](const ck_tile::ArgParser& arg_parser) {
// using TypeConfig =
// decltype(GemmQuantTypeConfig<ck_tile::fp8_t, ck_tile::fp8_t, ck_tile::half_t,
// float>{});
// using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 16, 128>>;
// return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
// TypeConfig,
// QuantGroupSize,
// ck_tile::QuantType::BQuantGrouped>(arg_parser);
// };
lut[hash_multiple_strings({"fp8", "bquant", "non-preshuffleb", "preshufflequant", "1x8x128"})] =
[](const ck_tile::ArgParser& arg_parser) {
using TypeConfig = decltype(GemmQuantTypeConfig<ck_tile::fp8_t,
ck_tile::fp8_t,
ck_tile::half_t,
float>{});
using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 8, 128>>;
return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
TypeConfig,
QuantGroupSize,
ck_tile::QuantType::BQuantGrouped>(arg_parser);
};
lut[hash_multiple_strings({"fp8",
"bquant",
"non-preshuffleb",
"preshufflequant",
"1x16x128"})] = [](const ck_tile::ArgParser& arg_parser) {
using TypeConfig =
decltype(GemmQuantTypeConfig<ck_tile::fp8_t, ck_tile::fp8_t, ck_tile::half_t, float>{});
using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 16, 128>>;
return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
TypeConfig,
QuantGroupSize,
ck_tile::QuantType::BQuantGrouped>(arg_parser);
};
lut[hash_multiple_strings({"fp8",
"bquant",
"non-preshuffleb",