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

sgemm_tcore: Use old opcodes to match frozen rtl

Browse Source
This commit is contained in:
Hansung Kim
2024-06-15 00:26:57 -07:00
parent 095ccfd79a
commit 9d6ff196b3
2 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/regression/sgemm_tcore/kernel.cpp
View File

@@ -334,7 +334,7 @@ inline void thread_block_gemm(kernel_arg_t *__UNIFORM__ arg,
GEMMINI_CISC_CMD_R((dim_n << 16) | (dim_k << 8) | 8);
gemmini_fence();
GEMMINI_CISC_CMD_I(12);
GEMMINI_CISC_CMD_I(10);
gemmini_fence();
#if 0
@@ -380,9 +380,9 @@ inline void thread_block_gemm(kernel_arg_t *__UNIFORM__ arg,
GEMMINI_CISC_CMD_R((dim_n << 16) | (dim_k << 8) | 8);
// gemmini_fence();
// block_k is even: opcode 13 (write to local_a_buf)
// block_k is odd: opcode 12 (write to local_a)
const uint32_t opcode = 13 - (block_k & 1);
// block_k is even: opcode 11 (write to local_a_buf)
// block_k is odd: opcode 10 (write to local_a)
const uint32_t opcode = 11 - (block_k & 1);
GEMMINI_CISC_CMD_R(opcode);
// // TODO: branch is probably slow
// if (block_k & 1) {