optinize and clean code

huggingface · Jan 21, 2025 · 070878f · 070878f
1 parent 49703f9
commit 070878f
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 4 deletions.
diff --git a/candle-kernels/src/fill.cu b/candle-kernels/src/fill.cu
@@ -35,7 +35,7 @@ COPY2D_OP(double, copy2d_f64)
 COPY2D_OP(uint8_t, copy2d_u8)
 COPY2D_OP(uint32_t, copy2d_u32)
 COPY2D_OP(int64_t, copy2d_i64)
-  
+
 #if __CUDA_ARCH__ >= 530
 #include <cuda_bf16.h>
 extern "C" __global__ void fill_f16(__half *buf, __half value, const size_t numel) { fill_with(buf, value, numel); }

diff --git a/candle-kernels/src/reduce.cu b/candle-kernels/src/reduce.cu
@@ -579,9 +579,7 @@ ROPE_OP(__nv_bfloat16, rope_bf16, rope_i_bf16, rope_thd_bf16)
 FAST_OP(__nv_bfloat16, fast_min_bf16, fast_max_bf16, fast_argmin_bf16, fast_argmax_bf16, fast_sum_bf16)
 #endif
 
-#if __CUDA_ARCH__ >= 800
-SUM_OP(__nv_bfloat16, sum_bf16)
-#elif __CUDA_ARCH__ >= 530 &&  __CUDA_ARCH__  >= 750
+#if __CUDA_ARCH__ >= 750
 SUM_OP(__nv_bfloat16, sum_bf16)
 #elif __CUDA_ARCH__ >= 530 &&  __CUDA_ARCH__ < 750
 //The automatic fallback mechanism for these architectures: