[PYTHON] Cleaned up legacy code; added simple standalone compilation …

…API (#22)

.clang-format

@@ -0,0 +1 @@
+BasedOnStyle: LLVM

.github/workflows/integration-tests.yml

@@ -4,8 +4,7 @@ on:
   workflow_dispatch:
   pull_request:
     branches:
-      - master
-      - v2.0
+      - main
 
 
 jobs:
@@ -21,7 +20,7 @@ jobs:
 
       - name: Clear cache
         run: |
-          rm -r /tmp/triton/
+          rm -r ~/.triton/
         continue-on-error: true
 
       - name: Install Triton

CMakeLists.txt

@@ -126,18 +126,10 @@ include_directories(${LLVM_INCLUDE_DIRS})
 # Python module
 if(BUILD_PYTHON_MODULE)
     message(STATUS "Adding Python module")
-    # Build CUTLASS python wrapper if requested
     set(PYTHON_SRC_PATH ${CMAKE_CURRENT_SOURCE_DIR}/python/src)
-    set(CUTLASS_INCLUDE_DIR "$ENV{CUTLASS_INCLUDE_DIR}")
-    set(CUTLASS_LIBRARY_DIR "$ENV{CUTLASS_LIBRARY_DIR}")
-    if(NOT("${CUTLASS_INCLUDE_DIR}" STREQUAL "") AND NOT("${CUTLASS_LIBRARY_DIR}" STREQUAL ""))
-        set(CUTLASS_SRC ${PYTHON_SRC_PATH}/cutlass.cc)
-        add_definitions(-DWITH_CUTLASS_BINDINGS)
-        set(CUTLASS_LIBRARIES "cutlass.a")
-    endif()
-    include_directories("." ${PYTHON_SRC_PATH} ${PYTHON_INCLUDE_DIRS} ${CUTLASS_INCLUDE_DIR})
-    link_directories(${PYTHON_LINK_DIRS} ${CUTLASS_LIBRARY_DIR})
-    set(PYTHON_SRC ${PYTHON_SRC_PATH}/main.cc ${PYTHON_SRC_PATH}/triton.cc  ${PYTHON_SRC_PATH}/superblock.cc ${CUTLASS_SRC})
+    include_directories("." ${PYTHON_SRC_PATH} ${PYTHON_INCLUDE_DIRS})
+    link_directories(${PYTHON_LINK_DIRS})
+    set(PYTHON_SRC ${PYTHON_SRC_PATH}/main.cc ${PYTHON_SRC_PATH}/triton.cc)
 endif()
 
 

include/triton/Dialect/TritonGPU/IR/TritonGPUAttrDefs.td

@@ -59,13 +59,13 @@ For example, a row-major coalesced layout may distribute a 64x16 tensor over 2 w
 
                           thread tile size 2
                         - - - - - - /\ - - - - - -
-block|  thread      || A_{0,  0}[T0]   A_{0,  1}[T0]   ... A_{0,  6}[T3]  A_{0,  7}[T3]      A_{0,  8}[T0]  A_{0,  9}[T0]  ... A_{0,  14}[T3]  A_{0,  15}[T3]
+warp |  thread      || A_{0,  0}[T0]   A_{0,  1}[T0]   ... A_{0,  6}[T3]  A_{0,  7}[T3]      A_{0,  8}[T0]  A_{0,  9}[T0]  ... A_{0,  14}[T3]  A_{0,  15}[T3]
 tile |  tile size 2 || A_{1,  0}[T0]   A_{1,  1}[T0]   ... A_{1,  6}[T3]  A_{1,  7}[T3]      A_{1,  8}[T0]  A_{1,  9}[T0]  ... A_{1,  14}[T3]  A_{1,  15}[T3]
 size }  ....
 32   |                 A_{30, 0}[T60]  A_{14, 1}[T60]  ... A_{14, 6}[T63] A_{14, 7}[T63]     A_{14, 8}[T60] A_{14, 9}[T60] ... A_{14, 14}[T63] A_{14, 15}[T63]
      |                 A_{31, 0}[T60]  A_{15, 1}[T60]  ... A_{15, 6}[T63] A_{15, 7}[T63]     A_{15, 8}[T60] A_{15, 9}[T60] ... A_{15, 14}[T63] A_{15, 15}[T63]
                       -----------------------------/\-----------------------------------
-                                            block tile size 8
+                                            warp tile size 8
 
 
                       A_{32, 0}[T0]  A_{32, 1}[T0]  ...  A_{32, 6}[T3]  A_{32, 7}[T3]        A_{32, 8}[T0]  A_{32, 9}[T0]  ... A_{32, 14}[T3]  A_{32, 15}[T3]