maybe this time it will work

harrisonvanderbyl · Jan 8, 2024 · 1676d30 · 1676d30
1 parent 903636f
commit 1676d30
Show file tree

Hide file tree

Showing 3 changed files with 4 additions and 13 deletions.
diff --git a/.github/workflows/mac.yml b/.github/workflows/mac.yml
@@ -70,7 +70,7 @@ jobs:
 
       - name: Compilation
         env:
-          SCONS_CACHE: ${{ steps.godot-cache.outputs.scons-cache }}
+          SCONS_CACHE: ${{github.workspace}}/.scons_cache/
         run: |
           scons arch=arm64 target=${{ matrix.target }} ${{ matrix.sconsflags }} -j4
 

diff --git a/SCsub b/SCsub
@@ -40,20 +40,11 @@ if(env["PLATFORM"] != "win32"):
         if env["arch"] == "arm32":
             print("arm32")
 
-
-            # env_rwkv.Append(CCFLAGS=['-march=armv8-a+simd'])
-            # clang, dont use march
-            # env_rwkv.Append(CCFLAGS=['-mpfu=neon-fp-armv7'])
 
         else:
             print("arm64")
-
-            # env_rwkv.Append(CCFLAGS=['-march=armv8.2-a+simd+crypto'])
-
-            # armv8
-            # env_rwkv.Append(CCFLAGS=['-mfpu=neon-fp-armv8'])
-
-    # -fopenmp  -flto  -fopenmp -funroll-loops -D_GLIBCXX_PARALLEL
+            env_rwkv.Append(CCFLAGS=['-march=armv8.2-a+bf16'])
+
 
 
 

diff --git a/rwkv.cuh b/rwkv.cuh
+63 −20		include/tensor/intrinsics/intrinsics.h
+3 −3		include/tensor/operators/gather/gather.h
+29 −29		include/tensor/tensor.h