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

flash: Comment out mvout to smem

Browse Source 
Verified up to O_before_PV; still stalls without DEBUG
This commit is contained in:
Hansung Kim
2024-09-09 15:21:49 -07:00
parent ecc800964a
commit 829af5d429
1 changed files with 8 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
tests/regression/flash_attention/kernel.gemmini.cpp
View File

@@ -82,6 +82,7 @@ void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
float *gmem_tmp_e3 = reinterpret_cast<float *>(0xe3000000UL); float *gmem_tmp_e3 = reinterpret_cast<float *>(0xe3000000UL);
// static shared memory allocation // static shared memory allocation
// these are in float elements, not bytes
constexpr uint32_t smem_Q_size = B_ROW * HEADDIM; constexpr uint32_t smem_Q_size = B_ROW * HEADDIM;
constexpr uint32_t smem_K_size = B_COL * HEADDIM; constexpr uint32_t smem_K_size = B_COL * HEADDIM;
constexpr uint32_t smem_QK_size = B_ROW * B_COL; constexpr uint32_t smem_QK_size = B_ROW * B_COL;
@@ -384,6 +385,8 @@ void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
// FIXME: perf: prevent GMEM->SMEM load for O tile // FIXME: perf: prevent GMEM->SMEM load for O tile
gemmini_fence(); gemmini_fence();
gemmini_fence(); gemmini_fence();
gemmini_fence();
gemmini_fence();
sp_tiled_matmul_full_spad_ws( sp_tiled_matmul_full_spad_ws(
spad_addr_P_consume, spad_addr_V_consume, spad_addr_P_consume, spad_addr_V_consume,
/*spad_D=*/spad_addr_O, /*spad_C=*/spad_addr_O, /*spad_D=*/spad_addr_O, /*spad_C=*/spad_addr_O,
@@ -446,7 +449,7 @@ void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
gemmini_fence(); gemmini_fence();
gemmini_fence(); gemmini_fence();
#if 1 #if 0
// mvout to SMEM // mvout to SMEM
// GEMMINI_CISC_CMD_I(9); // GEMMINI_CISC_CMD_I(9);
sp_tiled_matmul_full_spad_ws( sp_tiled_matmul_full_spad_ws(
@@ -493,10 +496,11 @@ void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
gemmini_fence(); gemmini_fence();
gemmini_fence(); gemmini_fence();
gemmini_fence(); gemmini_fence();
gemmini_fence();
// 0,2,.: opcode 0 (quartile 0/2, no accum) // 0,2,.: opcode 0 (quartile 0/2, no accum)
// 1,3,.: opcode 3 (quartile 1/3, no accum) // 1,3,.: opcode 3 (quartile 1/3, no accum)
const uint32_t opcode = 3 * (tile_k & 1); // const uint32_t opcode = 3 * (tile_k & 1);
//GEMMINI_CISC_CMD_I(opcode); //GEMMINI_CISC_CMD_I(opcode);
sp_tiled_matmul_full_spad_ws( sp_tiled_matmul_full_spad_ws(
spad_addr_Q, spad_addr_K_consume, spad_addr_Q, spad_addr_K_consume,
@@ -571,7 +575,7 @@ void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
gemmini_fence(); gemmini_fence();
gemmini_fence(); gemmini_fence();
#if 1 #if 0
// mvout to SMEM // mvout to SMEM
// GEMMINI_CISC_CMD_I(9); // GEMMINI_CISC_CMD_I(9);
sp_tiled_matmul_full_spad_ws( sp_tiled_matmul_full_spad_ws(
@@ -656,6 +660,7 @@ void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
gemmini_fence(); gemmini_fence();
gemmini_fence(); gemmini_fence();
gemmini_fence(); gemmini_fence();
gemmini_fence();
} }
threadblock_barrier(warpgroup_id_in_cluster, warps_per_warpgroup_per_core); threadblock_barrier(warpgroup_id_in_cluster, warps_per_warpgroup_per_core);