Allow dynamic ops per buffer based on dispatches and memory (#1864)

* Allow dynamic ops per buffer based on dispatches and memory

* add initial arch values
This commit is contained in:
Awni Hannun
2025-02-13 19:18:22 -08:00
committed by GitHub
parent 9733e16496
commit 7aea5b1895
5 changed files with 62 additions and 21 deletions

View File

@@ -109,7 +109,7 @@ std::tuple<bool, int64_t, array> check_transpose(
///////////////////////////////////////////////////////////////////////////////
#define GEMM_TPARAM_MACRO(devc) \
if (devc == 'g') { /* Small device */ \
if (devc == 'g' || devc == 'p') { /* Small device */ \
if (!transpose_a && transpose_b) { /* nt */ \
bm = 64; \
bn = 32; \