From 4b7eccc7da88e2e6bb92308ca9d1663947658c2e Mon Sep 17 00:00:00 2001
From: 0cc4m <picard12@live.de>
Date: Wed, 1 Nov 2023 22:22:57 +0100
Subject: [PATCH] Add Vulkan to llama-bench

---
 examples/llama-bench/llama-bench.cpp | 11 ++++++++---
 ggml.c                               | 10 +++++++++-
 ggml.h                               |  1 +
 llama.cpp                            |  2 +-
 4 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/examples/llama-bench/llama-bench.cpp b/examples/llama-bench/llama-bench.cpp
index 20767d555..1c1ea0d17 100644
--- a/examples/llama-bench/llama-bench.cpp
+++ b/examples/llama-bench/llama-bench.cpp
@@ -479,6 +479,7 @@ struct test {
     static const int build_number;
     static const bool cuda;
     static const bool opencl;
+    static const bool vulkan;
     static const bool metal;
     static const bool gpu_blas;
     static const bool blas;
@@ -554,6 +555,9 @@ struct test {
         if (opencl) {
             return "OpenCL";
         }
+        if (vulkan) {
+            return "Vulkan";
+        }
         if (metal) {
             return "Metal";
         }
@@ -569,7 +573,7 @@ struct test {
     static const std::vector<std::string> & get_fields() {
         static const std::vector<std::string> fields = {
             "build_commit", "build_number",
-            "cuda", "opencl", "metal", "gpu_blas", "blas",
+            "cuda", "opencl", "vulkan", "metal", "gpu_blas", "blas",
             "cpu_info", "gpu_info",
             "model_filename", "model_type", "model_size", "model_n_params",
             "n_batch", "n_threads", "f16_kv",
@@ -591,7 +595,7 @@ struct test {
             field == "avg_ns" || field == "stddev_ns") {
             return INT;
         }
-        if (field == "cuda" || field == "opencl" || field == "metal" || field == "gpu_blas" || field == "blas" ||
+        if (field == "cuda" || field == "opencl" || field == "vulkan" || field == "metal" || field == "gpu_blas" || field == "blas" ||
             field == "f16_kv" || field == "mul_mat_q") {
             return BOOL;
         }
@@ -619,7 +623,7 @@ struct test {
         }
         std::vector<std::string> values = {
             build_commit, std::to_string(build_number),
-            std::to_string(cuda), std::to_string(opencl), std::to_string(metal), std::to_string(gpu_blas), std::to_string(blas),
+            std::to_string(cuda), std::to_string(opencl), std::to_string(vulkan), std::to_string(metal), std::to_string(gpu_blas), std::to_string(blas),
             cpu_info, gpu_info,
             model_filename, model_type, std::to_string(model_size), std::to_string(model_n_params),
             std::to_string(n_batch), std::to_string(n_threads), std::to_string(!f32_kv),
@@ -645,6 +649,7 @@ const std::string test::build_commit = BUILD_COMMIT;
 const int         test::build_number = BUILD_NUMBER;
 const bool        test::cuda         = !!ggml_cpu_has_cublas();
 const bool        test::opencl       = !!ggml_cpu_has_clblast();
+const bool        test::vulkan       = !!ggml_cpu_has_vulkan();
 const bool        test::metal        = !!ggml_cpu_has_metal();
 const bool        test::gpu_blas     = !!ggml_cpu_has_gpublas();
 const bool        test::blas         = !!ggml_cpu_has_blas();
diff --git a/ggml.c b/ggml.c
index d7d972ccb..fd5acbe38 100644
--- a/ggml.c
+++ b/ggml.c
@@ -22308,8 +22308,16 @@ int ggml_cpu_has_clblast(void) {
 #endif
 }
 
+int ggml_cpu_has_vulkan(void) {
+#if defined(GGML_USE_VULKAN)
+    return 1;
+#else
+    return 0;
+#endif
+}
+
 int ggml_cpu_has_gpublas(void) {
-    return ggml_cpu_has_cublas() || ggml_cpu_has_clblast();
+    return ggml_cpu_has_cublas() || ggml_cpu_has_clblast() || ggml_cpu_has_vulkan();
 }
 
 int ggml_cpu_has_sse3(void) {
diff --git a/ggml.h b/ggml.h
index 08bff5511..b75727607 100644
--- a/ggml.h
+++ b/ggml.h
@@ -2078,6 +2078,7 @@ extern "C" {
     GGML_API int ggml_cpu_has_blas       (void);
     GGML_API int ggml_cpu_has_cublas     (void);
     GGML_API int ggml_cpu_has_clblast    (void);
+    GGML_API int ggml_cpu_has_vulkan     (void);
     GGML_API int ggml_cpu_has_gpublas    (void);
     GGML_API int ggml_cpu_has_sse3       (void);
     GGML_API int ggml_cpu_has_ssse3      (void);
diff --git a/llama.cpp b/llama.cpp
index 01cc1fb8f..b3fab8a67 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -2495,7 +2495,7 @@ static void llm_load_tensors(
 #define LLAMA_BACKEND_OFFLOAD       GGML_BACKEND_GPU
 #define LLAMA_BACKEND_OFFLOAD_SPLIT GGML_BACKEND_GPU
 #elif defined(GGML_USE_VULKAN)
-    fprintf(stderr, "%s: using Vulkan for GPU acceleration\n", __func__);
+    LLAMA_LOG_INFO("%s: using Vulkan for GPU acceleration\n", __func__);
 #define LLAMA_BACKEND_OFFLOAD       GGML_BACKEND_GPU
 #define LLAMA_BACKEND_OFFLOAD_SPLIT GGML_BACKEND_GPU
 #else