diff options
| author | JRPan <[email protected]> | 2025-09-03 16:01:54 -0400 |
|---|---|---|
| committer | GitHub <[email protected]> | 2025-09-03 13:01:54 -0700 |
| commit | 987a23f84f8705c0566a47aeda2132fa33a82618 (patch) | |
| tree | 67cc731de018ab6573a1dd1926cc005b3e5980db /src/gpgpu-sim/shader.h | |
| parent | 21b18be3832b80b00d4890f041818bac431866e7 (diff) | |
Fix correlation on A100 CUTLASS GEMM Kernels (#127)
* 1. Add a queue to promote fairness CTA scheduling
2. Add a Perfect xbar interconnect option
* remove merge queue
* Should not change this
* rename pending cta to allocated_ctas
* run formatter
Diffstat (limited to 'src/gpgpu-sim/shader.h')
| -rw-r--r-- | src/gpgpu-sim/shader.h | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h index 5922d24..6bf020b 100644 --- a/src/gpgpu-sim/shader.h +++ b/src/gpgpu-sim/shader.h @@ -2564,6 +2564,7 @@ class shader_core_ctx : public core_t { // Jin: concurrent kernels on a sm public: + std::deque<kernel_info_t *> pending_ctas; bool can_issue_1block(kernel_info_t &kernel); bool occupy_shader_resource_1block(kernel_info_t &kernel, bool occupy); void release_shader_resource_1block(unsigned hw_ctaid, kernel_info_t &kernel); |