CUDA: implement __hmax and __hmax2 for CUDA < 11.7

LostRuins · May 1, 2024 · 24ea3c6 · 24ea3c6
1 parent c4ec9c0
commit 24ea3c6
Showing 1 changed file with 17 additions and 4 deletions.
diff --git a/ggml-cuda/common.cuh b/ggml-cuda/common.cuh
@@ -137,7 +137,8 @@
 #define STRINGIZE(...) STRINGIZE_IMPL(__VA_ARGS__)
 
 #define WARP_SIZE 32
-#define CUDART_HMAX     11070 // CUDA 11.7, min. ver. for which __hmax and __hmax2 are known to work (may be higher than needed)
+#define CUDART_HMAX   11070 // CUDA 11.7, min. ver. for which __hmax and __hmax2 are known to work (may be higher than needed)
+#define CUDART_HMASK  12000 // CUDA 12.0, min. ver. for half2 -> uint mask comparisons
 
 #define CC_PASCAL     600
 #define MIN_CC_DP4A   610 // minimum compute capability for __dp4a, an intrinsic for byte-wise dot products
@@ -293,8 +294,20 @@ static __device__ __forceinline__ float warp_reduce_max(float x) {
     return x;
 }
 
+#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && CUDART_VERSION < CUDART_HMAX
+static __device__ __forceinline__ half __hmax(const half a, const half b) {
+    return __half2float(a) > __half2float(b) ? a : b;
+}
+static __device__ __forceinline__ half2 __hmax2(const half2 a, const half2 b) {
+    half2 ret;
+    reinterpret_cast<half&>(ret.x) =  __low2float(a) >  __low2float(b) ?  __low2half(a) :  __low2half(b);
+    reinterpret_cast<half&>(ret.y) = __high2float(a) > __high2float(b) ? __high2half(a) : __high2half(b);
+    return ret;
+}
+#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && CUDART_VERSION < CUDART_HMAX
+
 static __device__ __forceinline__ half2 warp_reduce_max(half2 x) {
-#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
+#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL
 #pragma unroll
    for (int mask = 16; mask > 0; mask >>= 1) {
        x = __hmax2(x, __shfl_xor_sync(0xffffffff, x, mask, 32));
@@ -303,10 +316,10 @@ static __device__ __forceinline__ half2 warp_reduce_max(half2 x) {
 #else
    GGML_UNUSED(x);
    NO_DEVICE_CODE;
-#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
+#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL
 }
 
-#if CUDART_VERSION < 12000
+#if CUDART_VERSION < CUDART_HMASK
 static __device__ __forceinline__ uint32_t __hgt2_mask(const half2 a, const half2 b) {
     const uint32_t mask_low  = 0x0000FFFF * (float( __low2half(a)) > float( __low2half(b)));
     const uint32_t mask_high = 0xFFFF0000 * (float(__high2half(a)) > float(__high2half(b)));