Merge branch 'master' into avx512

CMakeLists.txt

@@ -121,7 +121,8 @@ if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES
 else()
     message(STATUS "x86 detected")
     if (MSVC)
-        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /arch:AVX2 /D_CRT_SECURE_NO_WARNINGS=1")
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /arch:AVX2")
+        set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} /arch:AVX2")
     else()
         if (EMSCRIPTEN)
             # we require support for WASM SIMD 128-bit
@@ -148,7 +149,11 @@ target_include_directories(${TARGET} PUBLIC
     .
     )
 
+if (MSVC)
+  target_link_libraries(${TARGET} PRIVATE ${WHISPER_EXTRA_LIBS} ${CMAKE_THREAD_LIBS_INIT})
+else()
   target_link_libraries(${TARGET} PRIVATE m ${WHISPER_EXTRA_LIBS} ${CMAKE_THREAD_LIBS_INIT})
+endif()
 
 if (BUILD_SHARED_LIBS)
     target_link_libraries(${TARGET} PUBLIC
@@ -158,6 +163,10 @@ if (BUILD_SHARED_LIBS)
     target_compile_definitions(${TARGET} PUBLIC
         WHISPER_SHARED
         )
+
+    if (MSVC)
+        target_compile_definitions(${TARGET} PUBLIC __AVX2__ _CRT_SECURE_NO_WARNINGS)
+    endif()
 endif()
 
 target_compile_definitions(${TARGET} PUBLIC

examples/bench/bench.cpp

@@ -75,7 +75,7 @@ int main(int argc, char ** argv) {
     whisper_free(ctx);
 
     fprintf(stderr, "\n");
-    fprintf(stderr, "system_info: n_threads = %d | %s\n", params.n_threads, whisper_print_system_info());
+    fprintf(stderr, "system_info: n_threads = %d / %d | %s\n", params.n_threads, std::thread::hardware_concurrency(), whisper_print_system_info());
 
     fprintf(stderr, "\n");
     fprintf(stderr, "If you wish, you can submit these results here:\n");

examples/main/main.cpp

@@ -412,5 +412,8 @@ int main(int argc, char ** argv) {
     whisper_print_timings(ctx);
     whisper_free(ctx);
 
+    fprintf(stderr, "\n");
+    fprintf(stderr, "system_info: n_threads = %d / %d | %s\n", params.n_threads, std::thread::hardware_concurrency(), whisper_print_system_info());
+
     return 0;
 }

extra/bench-all.sh

@@ -0,0 +1,53 @@
+#!/bin/bash
+
+# Helper script to run the bench tool on all models and print the results in share-able format
+
+printf "Usage: ./bench.sh [n_threads]\n"
+
+if [ -z "$1" ]; then
+    n_threads=4
+else
+    n_threads=$1
+fi
+
+models=( "tiny" "base" "small" "medium" "large" )
+
+printf "\n"
+printf "Running benchmark for all models\n"
+printf "This can take a while!\n"
+printf "\n"
+
+printf "| CPU | OS | Config | Model | Threads | Load [ms] | Encode [ms] |\n"
+printf "| --- | -- | ------ | ----- | ------- | --------- | ----------- |\n"
+
+for model in "${models[@]}"; do
+    # run once to heat-up the cache
+    ./bench -m ./models/ggml-$model.bin -t $n_threads 2>/dev/null 1>/dev/null
+
+    # actual run
+    # store stderr output in a variable in order to parse it later
+    output=$(./bench -m ./models/ggml-$model.bin -t $n_threads 2>&1)
+
+    # parse the output:
+    load_time=$(echo "$output" | grep "load time" | awk '{print $5}')
+    encode_time=$(echo "$output" | grep "encode time" | awk '{print $5}')
+    system_info=$(echo "$output" | grep "system_info")
+    n_threads=$(echo "$output" | grep "system_info" | awk '{print $4}')
+
+    config=""
+
+    if [[ $system_info == *"AVX2 = 1"* ]]; then
+        config="$config AVX2"
+    fi
+
+    if [[ $system_info == *"NEON = 1"* ]]; then
+        config="$config NEON"
+    fi
+
+    if [[ $system_info == *"BLAS = 1"* ]]; then
+        config="$config BLAS"
+    fi
+
+    printf "| <todo> | <todo> | $config | $model | $n_threads | $load_time | $encode_time |\n"
+done
+