clean up

Signed-off-by: maleksan85 <[email protected]>

CMakeLists.txt

@@ -37,7 +37,7 @@ set(PYTHON_SUPPORTED_VERSIONS "3.9" "3.10" "3.11" "3.12")
 set(CUDA_SUPPORTED_ARCHS "7.0;7.5;8.0;8.6;8.9;9.0")
 
 # Supported AMD GPU architectures.
-set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100")
+set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101")
 
 #
 # Supported/expected torch versions for CUDA/ROCm.
@@ -187,7 +187,8 @@ message(STATUS "FetchContent base directory: ${FETCHCONTENT_BASE_DIR}")
 
 set(VLLM_EXT_SRC
   "csrc/cache_kernels.cu"
-  "csrc/attention/attention_kernels.cu"
+  "csrc/attention/paged_attention_v1.cu"
+  "csrc/attention/paged_attention_v2.cu"
   "csrc/pos_encoding_kernels.cu"
   "csrc/activation_kernels.cu"
   "csrc/layernorm_kernels.cu"

csrc/attention/paged_attention_v1.cu

@@ -17,15 +17,18 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
-#include <algorithm>
-
-#include "attention_dtypes.h"
-#include "attention_utils.cuh"
 #include "attention_kernels.cuh"
 
+#ifndef USE_ROCM
+  #define WARP_SIZE 32
+#else
+  #define WARP_SIZE warpSize
+#endif
+
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
+#define DIVIDE_ROUND_UP(a, b) (((a) + (b) - 1) / (b))
+
 #define LAUNCH_PAGED_ATTENTION_V1(HEAD_SIZE)                                \
   VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize(                     \
       ((void*)vllm::paged_attention_v1_kernel<T, CACHE_T, HEAD_SIZE,        \
@@ -187,4 +190,4 @@ void paged_attention_v1(
 #undef WARP_SIZE
 #undef MAX
 #undef MIN
-#undef DIVIDE_ROUND_UP
+#undef DIVIDE_ROUND_UP

csrc/attention/paged_attention_v2.cu

@@ -17,15 +17,17 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
-#include <algorithm>
-
-#include "attention_dtypes.h"
-#include "attention_utils.cuh"
 #include "attention_kernels.cuh"
 
+#ifndef USE_ROCM
+  #define WARP_SIZE 32
+#else
+  #define WARP_SIZE warpSize
+#endif
+
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
+#define DIVIDE_ROUND_UP(a, b) (((a) + (b) - 1) / (b))
 
 #define LAUNCH_PAGED_ATTENTION_V2(HEAD_SIZE)                                   \
   vllm::paged_attention_v2_kernel<T, CACHE_T, HEAD_SIZE, BLOCK_SIZE,           \
@@ -198,4 +200,4 @@ void paged_attention_v2(
 #undef WARP_SIZE
 #undef MAX
 #undef MIN
-#undef DIVIDE_ROUND_UP
+#undef DIVIDE_ROUND_UP