From: Georgi Gerganov <redacted>
Date: Thu, 8 Aug 2024 11:46:24 +0000 (+0300)
Subject: sync : vulkan (llama/0)
X-Git-Tag: upstream/0.0.1642~454
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=bc97237c9ecf5905776bcfe70d71aa0d71a74fee;p=pkg%2Fggml%2Fsources%2Fggml

sync : vulkan (llama/0)
---

diff --git a/src/vulkan-shaders/CMakeLists.txt b/src/vulkan-shaders/CMakeLists.txt
index 41551e00..10075db3 100644
--- a/src/vulkan-shaders/CMakeLists.txt
+++ b/src/vulkan-shaders/CMakeLists.txt
@@ -1,5 +1,7 @@
+find_package (Threads REQUIRED)
 
 set(TARGET vulkan-shaders-gen)
 add_executable(${TARGET} vulkan-shaders-gen.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_link_libraries(vulkan-shaders-gen PUBLIC Threads::Threads)
diff --git a/src/vulkan-shaders/mul_mat_vec.comp b/src/vulkan-shaders/mul_mat_vec.comp
index 15d2a806..46a6369b 100644
--- a/src/vulkan-shaders/mul_mat_vec.comp
+++ b/src/vulkan-shaders/mul_mat_vec.comp
@@ -16,6 +16,13 @@ void main() {
     const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
     const uint tid = gl_LocalInvocationID.x;
 
+    // There are not enough cols to use all threads
+    if (tid >= p.ncols) {
+        return;
+    }
+
+    const uint block_size = min(p.ncols, BLOCK_SIZE);
+
     uint a_offset, b_offset, d_offset;
     get_offsets(a_offset, b_offset, d_offset);
 
@@ -23,8 +30,8 @@ void main() {
 
     tmp[tid] = FLOAT_TYPE(0.0f);
 
-    [[unroll]] for (uint i = 0; i < p.ncols/BLOCK_SIZE; i += 2) {
-        const uint col = i*BLOCK_SIZE + 2*tid;
+    [[unroll]] for (uint i = 0; i < p.ncols/block_size; i += 2) {
+        const uint col = i*block_size + 2*tid;
         const uint ib = (row*p.ncols + col)/QUANT_K; // block index
         const uint iqs = (col%QUANT_K)/QUANT_R; // quant index
         const uint iybs = col - col%QUANT_K; // y block start index
@@ -38,7 +45,7 @@ void main() {
 
     // sum up partial sums and write back result
     barrier();
-    [[unroll]] for (uint s = BLOCK_SIZE/2; s > 0; s >>= 1) {
+    [[unroll]] for (uint s = block_size/2; s > 0; s >>= 1) {
         if (tid < s) {
             tmp[tid] += tmp[tid + s];
         }
diff --git a/src/vulkan-shaders/vulkan-shaders-gen.cpp b/src/vulkan-shaders/vulkan-shaders-gen.cpp
index 258a1933..a792e203 100644
--- a/src/vulkan-shaders/vulkan-shaders-gen.cpp
+++ b/src/vulkan-shaders/vulkan-shaders-gen.cpp
@@ -22,6 +22,7 @@
 #ifdef _WIN32
     #include <windows.h>
     #include <direct.h> // For _mkdir on Windows
+    #include <algorithm> // For std::replace on w64devkit
 #else
     #include <unistd.h>
     #include <sys/wait.h>
@@ -179,11 +180,7 @@ bool string_ends_with(const std::string& str, const std::string& suffix) {
     return std::equal(suffix.rbegin(), suffix.rend(), str.rbegin());
 }
 
-#ifdef _WIN32
-    static const char path_separator = '\\';
-#else
-    static const char path_separator = '/';
-#endif
+static const char path_separator = '/';
 
 std::string join_paths(const std::string& path1, const std::string& path2) {
     return path1 + path_separator + path2;
@@ -198,7 +195,11 @@ void string_to_spv(const std::string& _name, const std::string& in_fname, const
     std::string out_fname = join_paths(output_dir, name + ".spv");
     std::string in_path = join_paths(input_dir, in_fname);
 
-    std::vector<std::string> cmd = {GLSLC, "-fshader-stage=compute", "--target-env=vulkan1.2", "-O", in_path, "-o", out_fname};
+    #ifdef _WIN32
+        std::vector<std::string> cmd = {GLSLC, "-fshader-stage=compute", "--target-env=vulkan1.2", "-O", "\"" + in_path + "\"", "-o", "\"" + out_fname + "\""};
+    #else
+        std::vector<std::string> cmd = {GLSLC, "-fshader-stage=compute", "--target-env=vulkan1.2", "-O", in_path, "-o",  out_fname};
+    #endif
     for (const auto& define : defines) {
         cmd.push_back("-D" + define.first + "=" + define.second);
     }
@@ -482,10 +483,16 @@ void write_output_files() {
 
     for (const auto& pair : shader_fnames) {
         const std::string& name = pair.first;
-        const std::string& path = pair.second;
+        #ifdef _WIN32
+            std::string path = pair.second;
+            std::replace(path.begin(), path.end(), '/', '\\' );
+        #else
+            const std::string& path = pair.second;
+        #endif
+
         FILE* spv = fopen(path.c_str(), "rb");
         if (!spv) {
-            std::cerr << "Error opening SPIR-V file: " << path << "\n";
+            std::cerr << "Error opening SPIR-V file: " << path << " (" << strerror(errno) << ")\n";
             continue;
         }
 
@@ -497,7 +504,7 @@ void write_output_files() {
         size_t read_size = fread(data.data(), 1, size, spv);
         fclose(spv);
         if (read_size != size) {
-            std::cerr << "Error reading SPIR-V file: " << path << "\n";
+            std::cerr << "Error reading SPIR-V file: " << path << " (" << strerror(errno) << ")\n";
             continue;
         }