vulkan: Workaround compiler bug in conv2d coopmat2 path (llama/24924)
* vulkan: Workaround compiler bug in conv2d coopmat2 path * apply same workaround to CONV_3D * Apply suggestion from @jeffbolznv
This commit is contained in:
parent
b564d5364a
commit
df03e266f1
|
|
@ -158,7 +158,7 @@ const uint32_t Csh_stride = BS_NPQ;
|
||||||
#ifdef COOPMAT
|
#ifdef COOPMAT
|
||||||
const uint32_t Csh_len = BS_K * Csh_stride;
|
const uint32_t Csh_len = BS_K * Csh_stride;
|
||||||
#else
|
#else
|
||||||
const uint32_t Csh_len = csh_store != 0 ? BS_K * Csh_stride : 1;
|
const uint32_t Csh_len = csh_store != 0 ? BS_K * Csh_stride : 8; // 8 to workaround compiler bug
|
||||||
#endif
|
#endif
|
||||||
shared SHMEM_TYPE Csh[Csh_len]; // K x NPQ
|
shared SHMEM_TYPE Csh[Csh_len]; // K x NPQ
|
||||||
#endif
|
#endif
|
||||||
|
|
|
||||||
|
|
@ -144,7 +144,7 @@ const uint32_t Csh_stride = BS_NPQ;
|
||||||
#ifdef COOPMAT
|
#ifdef COOPMAT
|
||||||
const uint32_t Csh_len = BS_K * Csh_stride;
|
const uint32_t Csh_len = BS_K * Csh_stride;
|
||||||
#else
|
#else
|
||||||
const uint32_t Csh_len = csh_store != 0 ? BS_K * Csh_stride : 1;
|
const uint32_t Csh_len = csh_store != 0 ? BS_K * Csh_stride : 8; // 8 to workaround compiler bug
|
||||||
#endif
|
#endif
|
||||||
shared SHMEM_TYPE Csh[Csh_len]; // K x NPQ
|
shared SHMEM_TYPE Csh[Csh_len]; // K x NPQ
|
||||||
#endif
|
#endif
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue