diff --git a/ggml/src/ggml-cuda/mma.cuh b/ggml/src/ggml-cuda/mma.cuh index 1cec98b06..c72f12131 100644 --- a/ggml/src/ggml-cuda/mma.cuh +++ b/ggml/src/ggml-cuda/mma.cuh @@ -1409,14 +1409,14 @@ namespace ggml_cuda_mma { tile<16, 16, float, dl_d> & D, const tile<16, 8, nv_bfloat162, dl_ab> & A, const tile<16, 8, nv_bfloat162, dl_ab> & B) { #if defined(AMD_WMMA_AVAILABLE) #if defined(RDNA4) - using bf16x8_t = __attribute__((ext_vector_type(8))) __bf16; + using bf16x8_t = __attribute__((ext_vector_type(8))) short; using floatx8_t = __attribute__((ext_vector_type(8))) float; floatx8_t& acc_frag = reinterpret_cast(D.x[0]); const bf16x8_t& a_frag = reinterpret_cast(A.x[0]); const bf16x8_t& b_frag = reinterpret_cast(B.x[0]); acc_frag = __builtin_amdgcn_wmma_f32_16x16x16_bf16_w32_gfx12(a_frag, b_frag, acc_frag); #elif defined(RDNA3) - using bf16x16_t = __attribute__((ext_vector_type(16))) __bf16; + using bf16x16_t = __attribute__((ext_vector_type(16))) short; using floatx8_t = __attribute__((ext_vector_type(8))) float; floatx8_t& acc_frag = reinterpret_cast(D.x[0]); const bf16x16_t& a_frag = reinterpret_cast(A.x[0]);