From 5ef588ba58c449d1fa22cddcdecb7fe7da2600b5 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Mon, 7 Apr 2025 13:18:07 +0300
Subject: [PATCH] test

---
 ggml/src/ggml-cuda/convert.cu     | 4 ++++
 ggml/src/ggml-cuda/vendors/hip.h  | 1 +
 ggml/src/ggml-cuda/vendors/musa.h | 1 +
 3 files changed, 6 insertions(+)

diff --git a/ggml/src/ggml-cuda/convert.cu b/ggml/src/ggml-cuda/convert.cu
index 2681c42eb..39164d650 100644
--- a/ggml/src/ggml-cuda/convert.cu
+++ b/ggml/src/ggml-cuda/convert.cu
@@ -580,7 +580,11 @@ static __global__ void convert_unary(const void * __restrict__ vx, dst_t * __res
     const src_t * x = (const src_t *) vx;
 
     if constexpr (std::is_same_v<src_t, nv_bfloat16>) {
+#if defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)
+        y[i] = (float)x[i];
+#else
         y[i] = __bfloat162float(x[i]);
+#endif
     } else if constexpr (std::is_same_v<dst_t, nv_bfloat16> && std::is_same_v<src_t, half>) {
         y[i] = (float)x[i];
     } else {
diff --git a/ggml/src/ggml-cuda/vendors/hip.h b/ggml/src/ggml-cuda/vendors/hip.h
index 3983ce5b4..120f4dc06 100644
--- a/ggml/src/ggml-cuda/vendors/hip.h
+++ b/ggml/src/ggml-cuda/vendors/hip.h
@@ -20,6 +20,7 @@
 #define CUBLAS_STATUS_SUCCESS HIPBLAS_STATUS_SUCCESS
 #define CUBLAS_TF32_TENSOR_OP_MATH 0
 #define CUDA_R_16F  HIPBLAS_R_16F
+#define CUDA_R_16BF HIPBLAS_R_16BF
 #define CUDA_R_32F  HIPBLAS_R_32F
 #define CU_DEVICE_ATTRIBUTE_VIRTUAL_MEMORY_MANAGEMENT_SUPPORTED hipDeviceAttributeVirtualMemoryManagementSupported
 #define CU_MEM_ALLOC_GRANULARITY_RECOMMENDED hipMemAllocationGranularityRecommended
diff --git a/ggml/src/ggml-cuda/vendors/musa.h b/ggml/src/ggml-cuda/vendors/musa.h
index f2d55796e..937779a90 100644
--- a/ggml/src/ggml-cuda/vendors/musa.h
+++ b/ggml/src/ggml-cuda/vendors/musa.h
@@ -15,6 +15,7 @@
 #define CUBLAS_STATUS_SUCCESS MUBLAS_STATUS_SUCCESS
 #define CUBLAS_TF32_TENSOR_OP_MATH MUBLAS_MATH_MODE_DEFAULT
 #define CUDA_R_16F  MUSA_R_16F
+#define CUDA_R_16BF MUSA_R_16BF
 #define CUDA_R_32F  MUSA_R_32F
 #define cublasComputeType_t cudaDataType_t
 #define cublasCreate mublasCreate