From 987a23f84f8705c0566a47aeda2132fa33a82618 Mon Sep 17 00:00:00 2001
From: JRPan <25518778+JRPan@users.noreply.github.com>
Date: Wed, 3 Sep 2025 16:01:54 -0400
Subject: 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
---
 src/abstract_hardware_model.cc | 2 ++
 1 file changed, 2 insertions(+)

(limited to 'src/abstract_hardware_model.cc')

diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc
index e8ddf95..7123c24 100644
--- a/src/abstract_hardware_model.cc
+++ b/src/abstract_hardware_model.cc
@@ -782,6 +782,7 @@ kernel_info_t::kernel_info_t(dim3 gridDim, dim3 blockDim,
       num_blocks() * entry->gpgpu_ctx->device_runtime->g_TB_launch_latency;
 
   cache_config_set = false;
+  allocated_ctas = 0;
 }
 
 /*A snapshot of the texture mappings needs to be stored in the kernel's info as
@@ -815,6 +816,7 @@ kernel_info_t::kernel_info_t(
   cache_config_set = false;
   m_NameToCudaArray = nameToCudaArray;
   m_NameToTextureInfo = nameToTextureInfo;
+  allocated_ctas = 0;
 }
 
 kernel_info_t::~kernel_info_t() {
-- 
cgit v1.3