[CINN] Integrate CUTLASS into CINN [Part1] (#58079)

ZzSean · web-flow · commit 341afb72f9d8 · 2023-11-15T00:48:38.000-06:00
* [CINN] Integrate CUTLASS into CINN [Part1]

* fix cmake
diff --git a/cmake/cinn.cmake b/cmake/cinn.cmake
@@ -70,6 +70,10 @@ if(WITH_GPU)
     message(STATUS "Enable CINN CUDNN")
     add_definitions(-DCINN_WITH_CUDNN)
   endif()
+  if(WITH_CUTLASS)
+    message(STATUS "Enable CINN CUTLASS")
+    add_definitions(-DCINN_WITH_CUTLASS)
+  endif()
   enable_language(CUDA)
   find_package(CUDA REQUIRED)
   include_directories(${CUDA_INCLUDE_DIRS})
@@ -199,6 +203,11 @@ if(WITH_GPU)
   endif()
 endif()
 
+if(WITH_CUTLASS)
+  target_link_libraries(cinnapi cutlass)
+  add_dependencies(cinnapi cutlass)
+endif()
+
 function(gen_cinncore LINKTYPE)
   set(CINNCORE_TARGET cinncore)
   if(${LINKTYPE} STREQUAL "STATIC")
@@ -258,6 +267,11 @@ function(gen_cinncore LINKTYPE)
       target_link_libraries(${CINNCORE_TARGET} ${CUDA_NVTX_LIB})
     endif()
   endif()
+
+  if(WITH_CUTLASS)
+    target_link_libraries(cinnapi cutlass)
+    add_dependencies(cinnapi cutlass)
+  endif()
 endfunction()
 
 gen_cinncore(STATIC)
diff --git a/cmake/third_party.cmake b/cmake/third_party.cmake
@@ -272,6 +272,17 @@ include(external/gflags) # download, build, install gflags
 include(external/glog) # download, build, install glog
 
 ########################### include third_party according to flags ###############################
+if(WITH_GPU
+   AND NOT WITH_ARM
+   AND NOT WIN32
+   AND NOT APPLE)
+  if(${CMAKE_CUDA_COMPILER_VERSION} GREATER_EQUAL 11.0)
+    include(external/cutlass) # download, build, install cusparselt
+    list(APPEND third_party_deps extern_cutlass)
+    set(WITH_CUTLASS ON)
+  endif()
+endif()
+
 if(WITH_CINN)
   if(WITH_MKL)
     add_definitions(-DCINN_WITH_MKL_CBLAS)
@@ -555,11 +566,6 @@ if(WITH_GPU
    AND NOT WITH_ARM
    AND NOT WIN32
    AND NOT APPLE)
-  if(${CMAKE_CUDA_COMPILER_VERSION} GREATER_EQUAL 11.0)
-    include(external/cutlass) # download, build, install cusparselt
-    list(APPEND third_party_deps extern_cutlass)
-    set(WITH_CUTLASS ON)
-  endif()
   if(${CMAKE_CUDA_COMPILER_VERSION} GREATER_EQUAL 11.4)
     foreach(arch ${NVCC_ARCH_BIN})
       if(${arch} GREATER_EQUAL 80)
diff --git a/paddle/cinn/frontend/optimize.cc b/paddle/cinn/frontend/optimize.cc
@@ -38,6 +38,7 @@ PD_DECLARE_bool(use_reduce_split_pass);
 PD_DECLARE_bool(cinn_use_dense_merge_pass);
 PD_DECLARE_string(cinn_custom_call_deny_ops);
 PD_DECLARE_bool(general_fusion_merge_pass);
+PD_DECLARE_bool(cinn_use_cutlass);
 
 namespace cinn {
 namespace frontend {
@@ -58,6 +59,7 @@ OptimizeOptions DefaultTrainingOptimizeOptions() {
     return FLAGS_cinn_custom_call_deny_ops.find(op) != std::string::npos;
   };
   bool is_gemm_use_cublas = FLAGS_cinn_use_custom_call &&
+                            !FLAGS_cinn_use_cutlass &&
                             !can_find_custom_call_deny_op("matmul") &&
                             !can_find_custom_call_deny_op("cublas_gemm") &&
                             !can_find_custom_call_deny_op("cublas_matmul");
diff --git a/paddle/cinn/hlir/pass/custom_call_pass.cc b/paddle/cinn/hlir/pass/custom_call_pass.cc
@@ -18,6 +18,7 @@
 #include "paddle/cinn/utils/string.h"
 
 PD_DECLARE_string(cinn_custom_call_deny_ops);
+PD_DECLARE_bool(cinn_use_cutlass);
 
 namespace cinn {
 namespace hlir {
@@ -72,8 +73,10 @@ class GraphAlterHelper {
         }
       }
 
-      node->attrs.attr_store["original_op"] = node->op()->name;
-      node->attrs.op = framework::Operator::Get("custom_call");
+      if (!FLAGS_cinn_use_cutlass || node->op()->name != "matmul") {
+        node->attrs.attr_store["original_op"] = node->op()->name;
+        node->attrs.op = framework::Operator::Get("custom_call");
+      }
     }
   }
 
diff --git a/paddle/cinn/runtime/flags.cc b/paddle/cinn/runtime/flags.cc
@@ -220,6 +220,10 @@ PD_DEFINE_double(cinn_infer_model_version,
                  "Paddle has different model format in inference model. We use "
                  "a flag to load different versions.");
 
+PD_DEFINE_bool(cinn_use_cutlass,
+               BoolFromEnv("FLAGS_cinn_use_cutlass", false),
+               "Whether to use cutlass kernels");
+
 namespace cinn {
 namespace runtime {
 

Original file line number	Diff line number	Diff line change
`@@ -18,6 +18,7 @@`
`18`	`18`	`#include "paddle/cinn/utils/string.h"`
`19`	`19`
`20`	`20`	`PD_DECLARE_string(cinn_custom_call_deny_ops);`
	`21`	`+PD_DECLARE_bool(cinn_use_cutlass);`
`21`	`22`
`22`	`23`	`namespace cinn {`
`23`	`24`	`namespace hlir {`
`@@ -72,8 +73,10 @@ class GraphAlterHelper {`
`72`	`73`	`}`
`73`	`74`	`}`
`74`	`75`
`75`		`- node->attrs.attr_store["original_op"] = node->op()->name;`
`76`		`- node->attrs.op = framework::Operator::Get("custom_call");`
	`76`	`+ if (!FLAGS_cinn_use_cutlass \|\| node->op()->name != "matmul") {`
	`77`	`+ node->attrs.attr_store["original_op"] = node->op()->name;`
	`78`	`+ node->attrs.op = framework::Operator::Get("custom_call");`
	`79`	`+ }`
`77`	`80`	`}`
`78`	`81`	`}`
`79`	`82`