mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-03 13:11:25 +00:00
Add input fp8 and output bf16 attention (#2726)
* change host using fp16 to check * fp8 to fp8 compare * rewrite input parameters * add not squant * remove some output code * for scale = 1 * format * saturates only for fp8 * add fp8bf16 data type * add fp8bf16 data type * fix test fp8 code * add run_fp8bf16_tests * change fmha fwd example parameter(adding fp8bf16) * Support fp8bf16 for Aiter * Support aiter fp8bf16 in c++ * fix comment about fp8 in readme.md * add fp8fp32 * add fp8fp32 test * remove range_q etc. * format * fix test parameters about squant and fmha example input fp8bf16 fp8fp32 data type * add fp8bf16 to data_type function * change colmajor to rowmajor in test_ck_tile_fmha_fwd_fp8 * format * reset atol for fp8 * fix bug for atol --------- Co-authored-by: rocking <ChunYu.Lai@amd.com> Co-authored-by: asleepzzz <hanwen.chang@amd.com>
This commit is contained in:
@@ -1446,29 +1446,35 @@ struct FmhaFwdKernel
|
||||
auto o_acc_tile = [&]() {
|
||||
if constexpr(kDoFp8StaticQuant)
|
||||
{
|
||||
return FmhaPipeline{}(
|
||||
q_dram_window,
|
||||
identity{}, // q_element_func
|
||||
k_dram_window,
|
||||
identity{}, // k_element_func
|
||||
v_dram_window,
|
||||
identity{}, // v_element_func
|
||||
bias_dram_window,
|
||||
identity{}, // bias_element_func
|
||||
randval_dram_window,
|
||||
lse_dram_window,
|
||||
identity{}, // lse_element_func
|
||||
identity{}, // s_acc_element_func
|
||||
scales{kargs.scale_p}, // p_compute_element_func
|
||||
composes(saturates<fp8_t>{}, scales{kargs.scale_o}), // o_acc_element_func
|
||||
mask,
|
||||
position_encoding,
|
||||
kargs.scale_s,
|
||||
variant,
|
||||
variant_params,
|
||||
block_indices,
|
||||
smem_ptr,
|
||||
dropout);
|
||||
auto o_acc_element_func = [&]() {
|
||||
if constexpr(std::is_same_v<ODataType, ck_tile::fp8_t>)
|
||||
return ck_tile::composes(ck_tile::saturates<ck_tile::fp8_t>{},
|
||||
ck_tile::scales{kargs.scale_o});
|
||||
else
|
||||
return ck_tile::scales{kargs.scale_o};
|
||||
}();
|
||||
return FmhaPipeline{}(q_dram_window,
|
||||
identity{}, // q_element_func
|
||||
k_dram_window,
|
||||
identity{}, // k_element_func
|
||||
v_dram_window,
|
||||
identity{}, // v_element_func
|
||||
bias_dram_window,
|
||||
identity{}, // bias_element_func
|
||||
randval_dram_window,
|
||||
lse_dram_window,
|
||||
identity{}, // lse_element_func
|
||||
identity{}, // s_acc_element_func
|
||||
scales{kargs.scale_p}, // p_compute_element_func
|
||||
o_acc_element_func, // o_acc_element_func
|
||||
mask,
|
||||
position_encoding,
|
||||
kargs.scale_s,
|
||||
variant,
|
||||
variant_params,
|
||||
block_indices,
|
||||
smem_ptr,
|
||||
dropout);
|
||||
}
|
||||
else
|
||||
{
|
||||
|
||||
Reference in New Issue
Block a user