From a7a4e238fbbc2671eabc267c74b748da1405b767 Mon Sep 17 00:00:00 2001
From: Concedo <39025047+LostRuins@users.noreply.github.com>
Date: Wed, 26 Jun 2024 13:38:59 +0800
Subject: [PATCH] i hate doing this

---
 CMakeLists.txt   | 7 ++++---
 ggml-cuda/mmq.cu | 5 -----
 2 files changed, 4 insertions(+), 8 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 6ede7ca40..44bf32571 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -126,14 +126,15 @@ if (LLAMA_CUBLAS)
         # 60 == f16 CUDA intrinsics
         # 61 == integer CUDA intrinsics
         # 70 == (assumed) compute capability at which unrolling a loop in mul_mat_q kernels is faster
+        # 75 == int8 tensor cores
         if (LLAMA_CUDA_F16 OR LLAMA_CUDA_DMMV_F16)
-            set(CMAKE_CUDA_ARCHITECTURES "60;61;70") # needed for f16 CUDA intrinsics
+            set(CMAKE_CUDA_ARCHITECTURES "60;61;70;75") # needed for f16 CUDA intrinsics
         else()
             message("CUDA Toolkit Version: ${CUDAToolkit_VERSION}")
             if(CUDAToolkit_VERSION VERSION_GREATER 12)
-                 set(CMAKE_CUDA_ARCHITECTURES "52;61;70") # lowest CUDA 12 standard + lowest for integer intrinsics
+                 set(CMAKE_CUDA_ARCHITECTURES "52;61;70;75") # lowest CUDA 12 standard + lowest for integer intrinsics
             else()
-                 set(CMAKE_CUDA_ARCHITECTURES "37;52;61;70") # lowest CUDA 12 standard + lowest for integer intrinsics
+                 set(CMAKE_CUDA_ARCHITECTURES "37;52;61;70;75") # lowest CUDA 12 standard + lowest for integer intrinsics
             endif()
         endif()
     endif()
diff --git a/ggml-cuda/mmq.cu b/ggml-cuda/mmq.cu
index b45d1a6b2..7d951035a 100644
--- a/ggml-cuda/mmq.cu
+++ b/ggml-cuda/mmq.cu
@@ -108,11 +108,6 @@ bool ggml_cuda_should_use_mmq(enum ggml_type type, int cc, int64_t ne11) {
         return false;
     }
 
-    if(g_mul_mat_q)
-    {
-        return true;
-    }
-
     if (cc < CC_OFFSET_AMD) {
         return cc < CC_VOLTA || ne11 < MMQ_DP4A_MAX_BATCH_SIZE;
     }