Skip to content

GitLab

"...resnet50_tensorflow.git" did not exist on "8dace71ec50b435217bff5b99f0823ea7ae03ec3"
Commit 62ebdfde authored by Jing Zhang's avatar Jing Zhang
Browse files

clean xdlops_gemm

parent cb35d6fc
Hide whitespace changes
Inline Side-by-side
Showing with 266 additions and 281 deletions
composable_kernel/include/tensor_operation/blockwise_gemm_xdlops.hpp
View file @ 62ebdfde
......@@ -32,7 +32,6 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
static constexpr index_t K0 = BK0NK1BlockDesc{}.GetLength(I0);
static constexpr index_t KPerBlock = K0;
static constexpr index_t KPack = K1;
static constexpr auto xdlops_gemm = XdlopsGemm<FloatAB, MPerXDL, NPerXDL, K1>{};
......@@ -66,21 +65,10 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
const auto wave_idx = GetWaveIdx();
const auto waveId_m = wave_idx[I0];
const auto laneId = wave_idx[I2];
const auto blk_idx = xdlops_gemm.GetBlkIdx();
const auto xdlops_a_idx = xdlops_gemm.CalculateAThreadOriginDataIndex();
const auto blk_id = blk_idx[I0];
const auto blk_td = blk_idx[I1];
if constexpr(xdlops_gemm.IsKReduction)
{
return make_tuple(blk_id, 0, waveId_m, blk_td, 0);
}
else
{
return make_tuple(0, 0, waveId_m, laneId, 0);
}
return make_tuple(xdlops_a_idx[I0], 0, waveId_m, xdlops_a_idx[I1], 0);
}
__device__ static auto CalculateBThreadOriginDataIndex()
......@@ -88,21 +76,10 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
const auto wave_idx = GetWaveIdx();
const auto waveId_n = wave_idx[I1];
const auto laneId = wave_idx[I2];
const auto blk_idx = xdlops_gemm.GetBlkIdx();
const auto blk_id = blk_idx[I0];
const auto blk_td = blk_idx[I1];
const auto xdlops_b_idx = xdlops_gemm.CalculateBThreadOriginDataIndex();
if constexpr(xdlops_gemm.IsKReduction)
{
return make_tuple(blk_id, 0, waveId_n, blk_td, 0);
}
else
{
return make_tuple(0, 0, waveId_n, laneId, 0);
}
return make_tuple(xdlops_b_idx[I0], 0, waveId_n, xdlops_b_idx[I1], 0);
}
template <index_t m0, index_t n0, index_t xdlops_i, index_t blk_i>
......@@ -145,10 +122,6 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
static_assert(BlockSize == MWaves * NWaves * WaveSize,
"BlockSize != MWaves * NWaves * WaveSize\n");
static_assert(KPerBlock % xdlops_gemm.KPerXdlops == 0, "KPerBlock is wrong!");
static_assert(K1 % xdlops_gemm.mfma_type.k_base == 0, "K1 is wrong!");
static_assert(MPerBlock % (MPerXDL * MRepeat) == 0 && NPerBlock % (NPerXDL * NRepeat) == 0,
"wrong!");
......@@ -234,10 +207,10 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
vector_type<FloatAB, K1> b_thread_vec;
static_for<0, KPerBlock, xdlops_gemm.KPerXdlops>{}([&](auto k) {
static_for<0, KPerBlock, xdlops_gemm.KPerXdlops>{}([&](auto k0) {
// read A
a_thread_copy_.Run(a_k0_m0_m1_m2_k1_block_desc,
make_tuple(k, I0, I0, I0, I0),
make_tuple(k0, I0, I0, I0, I0),
a_block_buf,
a_thread_desc_,
make_tuple(I0, I0, I0, I0, I0),
......@@ -245,14 +218,14 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
// read B
b_thread_copy_.Run(b_k0_n0_n1_n2_k1_block_desc,
make_tuple(k, I0, I0, I0, I0),
make_tuple(k0, I0, I0, I0, I0),
b_block_buf,
b_thread_desc_,
make_tuple(I0, I0, I0, I0, I0),
b_thread_buf);
using mfma_input_type =
typename vector_type<FloatAB, xdlops_gemm.mfma_type.k_base>::type;
typename vector_type<FloatAB, xdlops_gemm.mfma_type.k_per_blk>::type;
static_for<0, MRepeat, 1>{}([&](auto m0) {
static_for<0, NRepeat, 1>{}([&](auto n0) {
......
composable_kernel/include/tensor_operation/xdlops_gemm.hpp
View file @ 62ebdfde
......@@ -9,19 +9,16 @@ namespace ck {
enum struct mfma_instr
{
/// fp32
mfma_f32_32x32x1xf32 = 0,
mfma_f32_16x16x1xf32,
mfma_f32_4x4x1xf32,
mfma_f32_32x32x2xf32, // k reduction
mfma_f32_16x16x4xf32, // k reduction
/// fp16
mfma_f32_32x32x4f16,
mfma_f32_16x16x4f16,
mfma_f32_4x4x4f16,
mfma_f32_32x32x8f16, // k reduction
mfma_f32_16x16x16f16, // k reduction
/// bfp16
mfma_f32_32x32x2bf16,
mfma_f32_16x16x2bf16,
mfma_f32_4x4x2bf16,
......@@ -35,19 +32,17 @@ struct mfma_info;
template <>
struct mfma_info<mfma_instr::mfma_f32_32x32x1xf32>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 4;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 2;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 32;
static constexpr index_t n = 32;
static constexpr index_t k = 1;
static constexpr index_t cycles = 64;
static constexpr index_t k_base = 1;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 4;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 2;
static constexpr index_t m_per_blk = 32;
static constexpr index_t n_per_blk = 32;
static constexpr index_t k_per_blk = 1;
static constexpr bool is_k_reduction = false;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -64,19 +59,17 @@ struct mfma_info<mfma_instr::mfma_f32_32x32x1xf32>
template <>
struct mfma_info<mfma_instr::mfma_f32_32x32x2xf32>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 4;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 32;
static constexpr index_t n = 32;
static constexpr index_t k = 2;
static constexpr index_t cycles = 64;
static constexpr index_t k_base = 1;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 4;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t m_per_blk = 32;
static constexpr index_t n_per_blk = 32;
static constexpr index_t k_per_blk = 1;
static constexpr bool is_k_reduction = true;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -93,19 +86,17 @@ struct mfma_info<mfma_instr::mfma_f32_32x32x2xf32>
template <>
struct mfma_info<mfma_instr::mfma_f32_16x16x4xf32>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 1;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 16;
static constexpr index_t n = 16;
static constexpr index_t k = 4;
static constexpr index_t cycles = 32;
static constexpr index_t k_base = 1;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 1;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t m_per_blk = 16;
static constexpr index_t n_per_blk = 16;
static constexpr index_t k_per_blk = 1;
static constexpr bool is_k_reduction = true;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -122,19 +113,17 @@ struct mfma_info<mfma_instr::mfma_f32_16x16x4xf32>
template <>
struct mfma_info<mfma_instr::mfma_f32_16x16x1xf32>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 1;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 4;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 16;
static constexpr index_t n = 16;
static constexpr index_t k = 1;
static constexpr index_t cycles = 32;
static constexpr index_t k_base = 1;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 1;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 4;
static constexpr index_t m_per_blk = 16;
static constexpr index_t n_per_blk = 16;
static constexpr index_t k_per_blk = 1;
static constexpr bool is_k_reduction = false;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -152,19 +141,17 @@ struct mfma_info<mfma_instr::mfma_f32_16x16x1xf32>
template <>
struct mfma_info<mfma_instr::mfma_f32_4x4x1xf32>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 1;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 64;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = 1;
static constexpr index_t num_output_blks = 1;
static constexpr index_t num_regs_xdlops = 4;
static constexpr index_t m = 4;
static constexpr index_t n = 64;
static constexpr index_t k = 1;
static constexpr index_t cycles = 8;
static constexpr index_t k_base = 1;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 1;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 64;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = 1;
static constexpr index_t num_output_blks = 1;
static constexpr index_t m_per_blk = 4;
static constexpr index_t n_per_blk = 64;
static constexpr index_t k_per_blk = 1;
static constexpr bool is_k_reduction = false;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -181,19 +168,17 @@ struct mfma_info<mfma_instr::mfma_f32_4x4x1xf32>
template <>
struct mfma_info<mfma_instr::mfma_f32_32x32x4f16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 4;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 2;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 32;
static constexpr index_t n = 32;
static constexpr index_t k = 4;
static constexpr index_t cycles = 64;
static constexpr index_t k_base = 4;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 4;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 2;
static constexpr index_t m_per_blk = 32;
static constexpr index_t n_per_blk = 32;
static constexpr index_t k_per_blk = 4;
static constexpr bool is_k_reduction = false;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -210,19 +195,17 @@ struct mfma_info<mfma_instr::mfma_f32_32x32x4f16>
template <>
struct mfma_info<mfma_instr::mfma_f32_32x32x8f16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 4;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 32;
static constexpr index_t n = 32;
static constexpr index_t k = 8;
static constexpr index_t cycles = 64;
static constexpr index_t k_base = 4;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 4;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t m_per_blk = 32;
static constexpr index_t n_per_blk = 32;
static constexpr index_t k_per_blk = 4;
static constexpr bool is_k_reduction = true;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -239,19 +222,17 @@ struct mfma_info<mfma_instr::mfma_f32_32x32x8f16>
template <>
struct mfma_info<mfma_instr::mfma_f32_16x16x16f16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 1;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 16;
static constexpr index_t n = 16;
static constexpr index_t k = 16;
static constexpr index_t cycles = 32;
static constexpr index_t k_base = 4;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 1;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t m_per_blk = 16;
static constexpr index_t n_per_blk = 16;
static constexpr index_t k_per_blk = 4;
static constexpr bool is_k_reduction = true;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -268,19 +249,17 @@ struct mfma_info<mfma_instr::mfma_f32_16x16x16f16>
template <>
struct mfma_info<mfma_instr::mfma_f32_16x16x4f16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 1;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 4;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 16;
static constexpr index_t n = 16;
static constexpr index_t k = 4;
static constexpr index_t cycles = 32;
static constexpr index_t k_base = 4;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 1;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 4;
static constexpr index_t m_per_blk = 16;
static constexpr index_t n_per_blk = 16;
static constexpr index_t k_per_blk = 4;
static constexpr bool is_k_reduction = false;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -297,19 +276,17 @@ struct mfma_info<mfma_instr::mfma_f32_16x16x4f16>
template <>
struct mfma_info<mfma_instr::mfma_f32_4x4x4f16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 1;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 64;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = 1;
static constexpr index_t num_output_blks = 1;
static constexpr index_t num_regs_xdlops = 4;
static constexpr index_t m = 4;
static constexpr index_t n = 64;
static constexpr index_t k = 4;
static constexpr index_t cycles = 8;
static constexpr index_t k_base = 4;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 1;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 64;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = 1;
static constexpr index_t num_output_blks = 1;
static constexpr index_t m_per_blk = 4;
static constexpr index_t n_per_blk = 64;
static constexpr index_t k_per_blk = 4;
static constexpr bool is_k_reduction = false;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -327,19 +304,17 @@ struct mfma_info<mfma_instr::mfma_f32_4x4x4f16>
template <>
struct mfma_info<mfma_instr::mfma_f32_32x32x2bf16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 4;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 2;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 32;
static constexpr index_t n = 32;
static constexpr index_t k = 2;
static constexpr index_t cycles = 64;
static constexpr index_t k_base = 2;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 4;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 2;
static constexpr index_t m_per_blk = 32;
static constexpr index_t n_per_blk = 32;
static constexpr index_t k_per_blk = 2;
static constexpr bool is_k_reduction = false;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -361,19 +336,17 @@ struct mfma_info<mfma_instr::mfma_f32_32x32x2bf16>
template <>
struct mfma_info<mfma_instr::mfma_f32_32x32x4bf16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 4;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 32;
static constexpr index_t n = 32;
static constexpr index_t k = 4;
static constexpr index_t cycles = 64;
static constexpr index_t k_base = 2;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 4;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 32;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t m_per_blk = 32;
static constexpr index_t n_per_blk = 32;
static constexpr index_t k_per_blk = 2;
static constexpr bool is_k_reduction = true;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -394,19 +367,17 @@ struct mfma_info<mfma_instr::mfma_f32_32x32x4bf16>
template <>
struct mfma_info<mfma_instr::mfma_f32_16x16x8bf16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 1;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 16;
static constexpr index_t n = 16;
static constexpr index_t k = 8;
static constexpr index_t cycles = 32;
static constexpr index_t k_base = 2;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 1;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 1;
static constexpr index_t m_per_blk = 16;
static constexpr index_t n_per_blk = 16;
static constexpr index_t k_per_blk = 2;
static constexpr bool is_k_reduction = true;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -427,19 +398,17 @@ struct mfma_info<mfma_instr::mfma_f32_16x16x8bf16>
template <>
struct mfma_info<mfma_instr::mfma_f32_16x16x2bf16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 1;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_blk;
static constexpr index_t num_output_blks = 4;
static constexpr index_t num_regs_xdlops = num_regs_blk * num_output_blks;
static constexpr index_t m = 16;
static constexpr index_t n = 16;
static constexpr index_t k = 2;
static constexpr index_t cycles = 32;
static constexpr index_t k_base = 2;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 1;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 16;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = wave_size / num_threads_per_blk;
static constexpr index_t num_output_blks = 4;
static constexpr index_t m_per_blk = 16;
static constexpr index_t n_per_blk = 16;
static constexpr index_t k_per_blk = 2;
static constexpr bool is_k_reduction = false;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -460,19 +429,17 @@ struct mfma_info<mfma_instr::mfma_f32_16x16x2bf16>
template <>
struct mfma_info<mfma_instr::mfma_f32_4x4x2bf16>
{
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_blk = 1;
static constexpr index_t num_regs_blk = group_size * num_groups_blk;
static constexpr index_t num_threads_blk = 64;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = 1;
static constexpr index_t num_output_blks = 1;
static constexpr index_t num_regs_xdlops = 4;
static constexpr index_t m = 4;
static constexpr index_t n = 64;
static constexpr index_t k = 2;
static constexpr index_t cycles = 8;
static constexpr index_t k_base = 2;
static constexpr index_t group_size = 4;
static constexpr index_t num_groups_per_blk = 1;
static constexpr index_t num_regs_per_blk = group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 64;
static constexpr index_t wave_size = 64;
static constexpr index_t num_input_blks = 1;
static constexpr index_t num_output_blks = 1;
static constexpr index_t m_per_blk = 4;
static constexpr index_t n_per_blk = 64;
static constexpr index_t k_per_blk = 2;
static constexpr bool is_k_reduction = false;
template <index_t MPerXdlops,
index_t NPerXdlops,
......@@ -505,14 +472,9 @@ struct xdlops_info
return true;
}
static constexpr bool IsKReduction()
{
return (mfma_type.num_output_blks == 1) && (mfma_type.num_input_blks > 1);
}
static constexpr index_t GetKPerXdlops()
{
return IsKReduction() ? mfma_type.num_input_blks : 1;
return mfma_type.is_k_reduction ? mfma_type.num_input_blks : 1;
}
static constexpr index_t GetNumCRegs() { return MPerXdlops * NPerXdlops / mfma_type.wave_size; }
......@@ -521,6 +483,13 @@ struct xdlops_info
template <class base_type, index_t MPerWave, index_t NPerWave, index_t KPack>
struct XdlopsGemm
{
static constexpr auto I0 = Number<0>{};
static constexpr auto I1 = Number<1>{};
static constexpr auto I2 = Number<2>{};
static constexpr auto I3 = Number<3>{};
static constexpr auto I4 = Number<4>{};
static constexpr auto I5 = Number<5>{};
template <class base_type_ = base_type,
index_t MPerWave_ = MPerWave,
index_t NPerWave_ = NPerWave>
......@@ -684,7 +653,30 @@ struct XdlopsGemm
__device__ static constexpr index_t GetNumXdlops()
{
return MPerXdlops * NPerXdlops / (mfma_type.m * mfma_type.n * mfma_type.num_output_blks);
return MPerXdlops * NPerXdlops /
(mfma_type.m_per_blk * mfma_type.n_per_blk * mfma_type.num_output_blks);
}
__host__ __device__ static void mfma_info_check()
{
static_assert(mfma_type.num_threads_per_blk == mfma_type.n_per_blk,
"n_per_blk != num_threads_per_blk");
static_assert(mfma_type.num_regs_per_blk * mfma_type.num_input_blks == mfma_type.m_per_blk,
"m_per_blk != num_input_blks * num_regs_per_blk");
static_assert(mfma_type.num_output_blks == mfma_type.num_input_blks ||
mfma_type.num_output_blks == 1,
"incorrect num_output_blks");
static_assert(mfma_type.num_regs_per_blk * mfma_type.wave_size ==
mfma_type.m_per_blk * mfma_type.n_per_blk,
"num_regs_per_blk incorrect");
static_assert(mfma_type.is_k_reduction ||
(mfma_type.num_input_blks == mfma_type.num_output_blks),
"is_k_reduction wrong!");
}
__host__ __device__ constexpr XdlopsGemm()
......@@ -696,30 +688,12 @@ struct XdlopsGemm
static_assert(MPerXdlops == 4 || MPerXdlops == 8 || MPerXdlops == 16 || MPerXdlops == 32 ||
MPerXdlops == 64,
"Only support GemmMPerXdlops == 4, 8, 16, 32 or 64 for xdlops");
static_assert(mfma_type.num_threads_blk == mfma_type.n, "n != num_threads_blk");
static_assert(mfma_type.num_regs_blk * mfma_type.num_input_blks == mfma_type.m,
"m != num_input_blks * num_regs_blk");
static_assert(mfma_type.num_output_blks == mfma_type.num_input_blks ||
mfma_type.num_output_blks == 1,
"incorrect num_output_blks");
static_assert(mfma_type.num_regs_blk * mfma_type.wave_size == mfma_type.m * mfma_type.n,
"num_regs_blk incorrect");
static_assert(mfma_type.k % mfma_type.k_base == 0, "k % kbase != 0!");
}
template <typename CM0N0M1N1M2N2GridDesc>
__host__ __device__ static constexpr auto
MakeCM0N0M1N1M2M3M4N2GridDescriptor(const CM0N0M1N1M2N2GridDesc& c_m0_n0_m1_n1_m2_n2_grid_desc)
{
constexpr auto I0 = Number<0>{};
constexpr auto I1 = Number<1>{};
constexpr auto I2 = Number<2>{};
constexpr auto I3 = Number<3>{};
constexpr auto I4 = Number<4>{};
constexpr auto I5 = Number<5>{};
constexpr auto M0 = c_m0_n0_m1_n1_m2_n2_grid_desc.GetLength(I0);
constexpr auto N0 = c_m0_n0_m1_n1_m2_n2_grid_desc.GetLength(I1);
constexpr auto M1 = c_m0_n0_m1_n1_m2_n2_grid_desc.GetLength(I2);
......@@ -727,9 +701,9 @@ struct XdlopsGemm
constexpr auto M2 = c_m0_n0_m1_n1_m2_n2_grid_desc.GetLength(I4);
constexpr auto N2 = c_m0_n0_m1_n1_m2_n2_grid_desc.GetLength(I5);
static_assert(N2 == mfma_type.num_threads_blk, "");
static_assert(N2 == mfma_type.num_threads_per_blk, "");
static_assert(
M2 == (mfma_type.num_groups_blk * mfma_type.num_output_blks * mfma_type.group_size),
M2 == (mfma_type.num_groups_per_blk * mfma_type.num_output_blks * mfma_type.group_size),
"");
return transform_dynamic_tensor_descriptor(
......@@ -738,10 +712,10 @@ struct XdlopsGemm
make_pass_through_transform(N0),
make_pass_through_transform(M1),
make_pass_through_transform(N1),
make_unmerge_transform(make_tuple(mfma_type.num_groups_blk,
make_unmerge_transform(make_tuple(mfma_type.num_groups_per_blk,
mfma_type.num_input_blks,
mfma_type.group_size)),
make_pass_through_transform(mfma_type.num_threads_blk)),
make_pass_through_transform(mfma_type.num_threads_per_blk)),
make_tuple(Sequence<0>{},
Sequence<1>{},
Sequence<2>{},
......@@ -768,40 +742,79 @@ struct XdlopsGemm
is_same<base_type, ushort>::value,
"base base_type must be float, half, ushort!");
static_assert(KPack % mfma_type.k_base == 0, "KPack cannot be divided by k_base");
static_assert(KPack % mfma_type.k_per_blk == 0, "KPack cannot be divided by k_per_blk");
static_for<0, KPack / mfma_type.k_base, 1>{}([&](auto k) {
static_for<0, KPack / mfma_type.k_per_blk, 1>{}([&](auto k) {
mfma_type.template run<MPerXdlops, NPerXdlops, c_offset>(
p_a_wave[k], p_b_wave[k], p_c_thread);
});
}
static constexpr auto GetBlkIdx()
__device__ static auto GetLaneId() { return get_thread_local_1d_id() % mfma_type.wave_size; }
__device__ static auto GetBlkIdx(const index_t laneId)
{
const auto threadidx_to_blk_idx_adaptor = make_single_stage_tensor_adaptor(
make_tuple(make_merge_transform(
make_tuple(1, mfma_type.num_input_blks, mfma_type.num_threads_blk))),
make_tuple(1, mfma_type.num_input_blks, mfma_type.num_threads_per_blk))),
make_tuple(Sequence<0, 1, 2>{}),
make_tuple(Sequence<0>{}));
const auto blk_idx = threadidx_to_blk_idx_adaptor.CalculateBottomIndex(
make_multi_index(get_thread_local_1d_id()));
const auto blk_idx =
threadidx_to_blk_idx_adaptor.CalculateBottomIndex(make_multi_index(laneId));
const auto blk_id = blk_idx[Number<1>{}];
const auto blk_td = blk_idx[Number<2>{}];
const auto blk_id = blk_idx[I1];
const auto blk_td = blk_idx[I2];
return make_tuple(blk_id, blk_td);
}
__host__ __device__ static auto CalculateAThreadOriginDataIndex()
{
const auto laneId = GetLaneId();
const auto blk_idx = GetBlkIdx(laneId);
const auto blk_id = blk_idx[I0];
const auto blk_td = blk_idx[I1];
if constexpr(mfma_type.is_k_reduction)
{
return make_tuple(blk_id, blk_td);
}
else
{
return make_tuple(0, laneId);
}
}
__host__ __device__ static auto CalculateBThreadOriginDataIndex()
{
const auto laneId = GetLaneId();
const auto blk_idx = GetBlkIdx(laneId);
const auto blk_id = blk_idx[I0];
const auto blk_td = blk_idx[I1];
if constexpr(mfma_type.is_k_reduction)
{
return make_tuple(blk_id, blk_td);
}
else
{
return make_tuple(0, laneId);
}
}
__device__ static CIndex GetBeginOfThreadBlk(index_t xdlops_i, index_t blk_i)
{
const auto blk_idx = GetBlkIdx();
const auto laneId = GetLaneId();
const auto blk_idx = GetBlkIdx(laneId);
const auto blk_id = blk_idx[Number<0>{}];
const auto blk_td = blk_idx[Number<1>{}];
const auto blk_id = blk_idx[I0];
const auto blk_td = blk_idx[I1];
index_t n_offset = blk_i * mfma_type.n + blk_td;
index_t m_offset = xdlops_i * mfma_type.m + blk_id * mfma_type.group_size;
index_t n_offset = blk_i * mfma_type.n_per_blk + blk_td;
index_t m_offset = xdlops_i * mfma_type.m_per_blk + blk_id * mfma_type.group_size;
return CIndex{m_offset, n_offset};
}
......@@ -810,26 +823,25 @@ struct XdlopsGemm
static constexpr index_t NPerXdlops = GetXdlopsInfo().NPerXdlops;
static constexpr index_t KPerXdlops = GetXdlopsInfo().GetKPerXdlops();
static constexpr bool IsKReduction = GetXdlopsInfo().IsKReduction();
static constexpr bool IsABroadcast = GetXdlopsInfo().IsABroadcast();
static constexpr auto mfma_type = GetXdlopsInfo().mfma_type;
struct CLayout
{
__host__ __device__ static constexpr index_t M1() { return mfma_type.num_groups_blk; }
__host__ __device__ static constexpr index_t M1() { return mfma_type.num_groups_per_blk; }
__host__ __device__ static constexpr index_t M0() { return mfma_type.group_size; }
__host__ __device__ static constexpr index_t N1() { return mfma_type.num_input_blks; }
__host__ __device__ static constexpr index_t N0() { return mfma_type.num_threads_blk; }
__host__ __device__ static constexpr index_t N0() { return mfma_type.num_threads_per_blk; }
__device__ static constexpr index_t GetBlkSize() { return mfma_type.num_regs_blk; }
__device__ static constexpr index_t GetBlkSize() { return mfma_type.num_regs_per_blk; }
__device__ static constexpr index_t GetNumBlks() { return mfma_type.num_output_blks; }
__device__ static constexpr index_t GetNumXdlops()
{
return MPerXdlops * NPerXdlops /
(mfma_type.m * mfma_type.n * mfma_type.num_output_blks);
(mfma_type.m_per_blk * mfma_type.n_per_blk * mfma_type.num_output_blks);
}
};
......
host/driver_offline/src/conv_fwd_driver_offline.cpp
View file @ 62ebdfde
......@@ -3,7 +3,7 @@
#include <initializer_list>
#include <cstdlib>
#include <stdlib.h>
#include <half.hpp>
//#include <half.hpp>
#include "config.hpp"
#include "print.hpp"
#include "device.hpp"
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment