wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

sgemm_impl: Add skeleton wgmma routine for single_tile

Browse Source
This commit is contained in:
Hansung Kim
2024-10-28 12:47:20 -07:00
parent e8a943e893
commit 36eb50060f
1 changed files with 62 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
tests/regression/sgemm_tcore/sgemm_impl.hpp
View File

@@ -233,6 +233,18 @@ inline void vx_wmma(const int dest_reg) {
} }
} }
inline void vx_wgmma() {
// .insn r opcode6, func3, func7, rd, rs1, rs2
// https://www.rowleydownload.co.uk/arm/documentation/gnu/as/RISC_002dV_002dFormats.html#RISC_002dV_002dFormats
asm volatile (".insn r %0, 0, 0, x0, x0, x0" :: "i"(RISCV_CUSTOM3));
}
inline void vx_wgmma_wait() {
// .insn r opcode6, func3, func7, rd, rs1, rs2
// func3 == 1 encodes wait
asm volatile (".insn r %0, 1, 0, x0, x0, x0" :: "i"(RISCV_CUSTOM3));
}
// Remap logical row/col coordinate of a matrix element to a memory index that // Remap logical row/col coordinate of a matrix element to a memory index that
// follows the 2-level block-row-major layout that Gemmini DMA uses // follows the 2-level block-row-major layout that Gemmini DMA uses
template <bool use_dma, uint32_t dim_col> template <bool use_dma, uint32_t dim_col>
@@ -804,9 +816,19 @@ __attribute__((always_inline)) inline void thread_block_gemm_single_tile(
const uint32_t warps_per_threadblock_per_core = const uint32_t warps_per_threadblock_per_core =
NUM_WARPS / threadblocks_per_cluster; NUM_WARPS / threadblocks_per_cluster;
// TODO: it would be useful if this bit is split out into a function, so that if constexpr (TENSOR_HOPPER) {
// preloading accumulation tile can be used for full GEMMs at the start of #pragma GCC unroll 1
// the K-loop. for (int i = 0; i < BK_LOOP; i++) {
#pragma GCC unroll 4
for (uint32_t local_k = 0; local_k < tile_dim_k; local_k += TCK) {
// FIXME: use local_a and local_b here
vx_wgmma();
}
}
} else {
// TODO: it would be useful if this bit is split out into a function, so
// that preloading accumulation tile can be used for full GEMMs at the start
// of the K-loop.
if constexpr (load_accum) { if constexpr (load_accum) {
#pragma GCC unroll #pragma GCC unroll
for (int wm_iter = 0; wm_iter < WMITER; wm_iter++) { for (int wm_iter = 0; wm_iter < WMITER; wm_iter++) {
@@ -837,8 +859,8 @@ __attribute__((always_inline)) inline void thread_block_gemm_single_tile(
wmma_load_a<T, layout_a, tile_dim_m, tile_dim_n, tile_dim_k>( wmma_load_a<T, layout_a, tile_dim_m, tile_dim_n, tile_dim_k>(
local_a, local_k, warp_row, wm_iter, tid_in_warp); local_a, local_k, warp_row, wm_iter, tid_in_warp);
} else { } else {
wmma_load_a<T, layout_a, leading_dim_a>(local_a, local_k, warp_row, wmma_load_a<T, layout_a, leading_dim_a>(
wm_iter, tid_in_warp); local_a, local_k, warp_row, wm_iter, tid_in_warp);
} }
// perform mma // perform mma
vx_wmma(wm_iter); vx_wmma(wm_iter);
@@ -846,6 +868,7 @@ __attribute__((always_inline)) inline void thread_block_gemm_single_tile(
} }
} }
} }
}
if constexpr (GEMMINI_DMA) { if constexpr (GEMMINI_DMA) {
// Call gemmini fence at the end of the loop to overlap dma & wmma. // Call gemmini fence at the end of the loop to overlap dma & wmma.