diff --git a/ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp b/ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp index 49759c5937782..477355c29304b 100644 --- a/ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp +++ b/ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp @@ -16,6 +16,7 @@ #include #include #include +#include #include #include @@ -92,11 +93,11 @@ void execute_command(const std::string& command, std::string& stdout_str, std::s std::array buffer; DWORD bytes_read; - while (ReadFile(stdout_read, buffer.data(), buffer.size(), &bytes_read, NULL) && bytes_read > 0) { + while (ReadFile(stdout_read, buffer.data(), (DWORD)buffer.size(), &bytes_read, NULL) && bytes_read > 0) { stdout_str.append(buffer.data(), bytes_read); } - while (ReadFile(stderr_read, buffer.data(), buffer.size(), &bytes_read, NULL) && bytes_read > 0) { + while (ReadFile(stderr_read, buffer.data(), (DWORD)buffer.size(), &bytes_read, NULL) && bytes_read > 0) { stderr_str.append(buffer.data(), bytes_read); } @@ -190,7 +191,12 @@ std::string basename(const std::string &path) { return path.substr(path.find_last_of("/\\") + 1); } -void string_to_spv(const std::string& _name, const std::string& in_fname, const std::map& defines, bool fp16 = true) { +// variables to track number of compiles in progress +static uint32_t compile_count = 0; +static std::mutex compile_count_mutex; +static std::condition_variable compile_count_cond; + +void string_to_spv_func(const std::string& _name, const std::string& in_fname, const std::map& defines, bool fp16 = true) { std::string name = _name + (fp16 ? "" : "_fp32"); std::string out_fname = join_paths(output_dir, name + ".spv"); std::string in_path = join_paths(input_dir, in_fname); @@ -233,6 +239,12 @@ void string_to_spv(const std::string& _name, const std::string& in_fname, const } catch (const std::exception& e) { std::cerr << "Error executing command for " << name << ": " << e.what() << std::endl; } + { + std::lock_guard guard(compile_count_mutex); + assert(compile_count > 0); + compile_count--; + } + compile_count_cond.notify_all(); } std::map merge_maps(const std::map& a, const std::map& b) { @@ -241,7 +253,22 @@ std::map merge_maps(const std::map>& tasks, bool fp16, bool matmul_id) { +static std::vector> compiles; +void string_to_spv(const std::string& _name, const std::string& in_fname, const std::map& defines, bool fp16 = true) { + { + // wait until fewer than N compiles are in progress. + // 16 is an arbitrary limit, the goal is to avoid "failed to create pipe" errors. + uint32_t N = 16; + std::unique_lock guard(compile_count_mutex); + while (compile_count >= N) { + compile_count_cond.wait(guard); + } + compile_count++; + } + compiles.push_back(std::async(string_to_spv_func, _name, in_fname, defines, fp16)); +} + +void matmul_shaders(bool fp16, bool matmul_id) { std::string load_vec = fp16 ? "8" : "4"; std::string aligned_b_type_f32 = fp16 ? "mat2x4" : "vec4"; std::string aligned_b_type_f16 = fp16 ? "f16mat2x4" : "f16vec4"; @@ -259,19 +286,11 @@ void matmul_shaders(std::vector>& tasks, bool fp16, bool matmu } // Shaders with f16 B_TYPE - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv(shader_name + "_f32_f16", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F32", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}), fp16); - })); - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv(shader_name + "_f32_f16_aligned", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F32", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}}), fp16); - })); - - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv(shader_name + "_f16", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F16", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}), fp16); - })); - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv(shader_name + "_f16_aligned", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F16", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}}), fp16); - })); + string_to_spv(shader_name + "_f32_f16", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F32", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}), fp16); + string_to_spv(shader_name + "_f32_f16_aligned", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F32", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}}), fp16); + + string_to_spv(shader_name + "_f16", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F16", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}), fp16); + string_to_spv(shader_name + "_f16_aligned", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F16", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}}), fp16); for (const auto& tname : type_names) { std::string data_a_key = "DATA_A_" + to_uppercase(tname); @@ -279,22 +298,18 @@ void matmul_shaders(std::vector>& tasks, bool fp16, bool matmu std::string load_vec_a_unaligned = (tname == "f32" || tname == "f16") ? "1" : "2"; // For aligned matmul loads std::string load_vec_a = (tname == "f32" || tname == "f16") ? load_vec : "2"; - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv(shader_name + "_" + tname + "_f32", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}), fp16); - })); - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv(shader_name + "_" + tname + "_f32_aligned", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}}), fp16); - })); + string_to_spv(shader_name + "_" + tname + "_f32", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}), fp16); + string_to_spv(shader_name + "_" + tname + "_f32_aligned", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}}), fp16); } } -void process_shaders(std::vector>& tasks) { +void process_shaders() { std::cout << "ggml_vulkan: Generating and compiling shaders to SPIR-V" << std::endl; std::map base_dict = {{"FLOAT_TYPE", "float"}}; for (const auto& fp16 : {false, true}) { - matmul_shaders(tasks, fp16, false); - matmul_shaders(tasks, fp16, true); + matmul_shaders(fp16, false); + matmul_shaders(fp16, true); } for (const auto& tname : type_names) { @@ -302,201 +317,103 @@ void process_shaders(std::vector>& tasks) { std::string data_a_key = "DATA_A_" + to_uppercase(tname); std::string shader = (string_ends_with(tname, "_k")) ? "mul_mat_vec_" + tname + ".comp" : "mul_mat_vec.comp"; - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("mul_mat_vec_" + tname + "_f32_f32", shader, merge_maps(base_dict, {{data_a_key, "1"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}})); - })); - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("mul_mat_vec_" + tname + "_f16_f32", shader, merge_maps(base_dict, {{data_a_key, "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}})); - })); + string_to_spv("mul_mat_vec_" + tname + "_f32_f32", shader, merge_maps(base_dict, {{data_a_key, "1"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}})); + string_to_spv("mul_mat_vec_" + tname + "_f16_f32", shader, merge_maps(base_dict, {{data_a_key, "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}})); - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("mul_mat_vec_id_" + tname + "_f32", shader, merge_maps(base_dict, {{"MUL_MAT_ID", "1"}, {data_a_key, "1"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}})); - })); + string_to_spv("mul_mat_vec_id_" + tname + "_f32", shader, merge_maps(base_dict, {{"MUL_MAT_ID", "1"}, {data_a_key, "1"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}})); // Dequant shaders if (tname != "f16") { - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("dequant_" + tname, "dequant_" + tname + ".comp", merge_maps(base_dict, {{data_a_key, "1"}, {"D_TYPE", "float16_t"}})); - })); + string_to_spv("dequant_" + tname, "dequant_" + tname + ".comp", merge_maps(base_dict, {{data_a_key, "1"}, {"D_TYPE", "float16_t"}})); } if (!string_ends_with(tname, "_k")) { shader = (tname == "f32" || tname == "f16") ? "get_rows.comp" : "get_rows_quant.comp"; if (tname == "f16") { - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("get_rows_" + tname, shader, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}}); - })); + string_to_spv("get_rows_" + tname, shader, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}}); } else { - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("get_rows_" + tname, shader, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float16_t"}}); - })); + string_to_spv("get_rows_" + tname, shader, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float16_t"}}); } - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("get_rows_" + tname + "_f32", shader, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float"}}); - })); + string_to_spv("get_rows_" + tname + "_f32", shader, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float"}}); } } - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("mul_mat_vec_p021_f16_f32", "mul_mat_vec_p021.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("mul_mat_vec_nc_f16_f32", "mul_mat_vec_nc.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}); - })); + string_to_spv("mul_mat_vec_p021_f16_f32", "mul_mat_vec_p021.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("mul_mat_vec_nc_f16_f32", "mul_mat_vec_nc.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}); // Norms - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("norm_f32", "norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); - })); - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("group_norm_f32", "group_norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); - })); - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("rms_norm_f32", "rms_norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("cpy_f32_f32", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("cpy_f32_f16", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("cpy_f16_f16", "copy.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("add_f32", "add.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("add_f16_f32_f16", "add.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float16_t"}, {"FLOAT_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("acc_f32", "acc.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("split_k_reduce", "mul_mat_split_k_reduce.comp", {}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("mul_f32", "mul.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("div_f32", "div.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("repeat_f32", "repeat.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("scale_f32", "scale.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("sqr_f32", "square.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("sin_f32", "sin.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("cos_f32", "cos.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("clamp_f32", "clamp.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("pad_f32", "pad.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("concat_f32", "concat.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("concat_f16", "concat.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("concat_i32", "concat.comp", {{"A_TYPE", "int"}, {"B_TYPE", "int"}, {"D_TYPE", "int"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("upscale_f32", "upscale.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("gelu_f32", "gelu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("gelu_quick_f32", "gelu_quick.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("silu_f32", "silu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("relu_f32", "relu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("leaky_relu_f32", "leaky_relu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("tanh_f32", "tanh.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("diag_mask_inf_f32", "diag_mask_inf.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("soft_max_f32", "soft_max.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}})); - })); - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("soft_max_f32_f16", "soft_max.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}})); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("rope_norm_f32", "rope_norm.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("rope_norm_f16", "rope_norm.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("rope_neox_f32", "rope_neox.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); - })); - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("rope_neox_f16", "rope_neox.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}}); - })); - - tasks.push_back(std::async(std::launch::async, [] { - string_to_spv("argsort_f32", "argsort.comp", {{"A_TYPE", "float"}}); - })); - - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("sum_rows_f32", "sum_rows.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); - })); - - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("im2col_f32", "im2col.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); - })); - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("im2col_f32_f16", "im2col.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}})); - })); - - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("timestep_embedding_f32", "timestep_embedding.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); - })); - - tasks.push_back(std::async(std::launch::async, [=] { - string_to_spv("pool2d_f32", "pool2d.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); - })); + string_to_spv("norm_f32", "norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); + string_to_spv("group_norm_f32", "group_norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); + string_to_spv("rms_norm_f32", "rms_norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); + + string_to_spv("cpy_f32_f32", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("cpy_f32_f16", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}}); + string_to_spv("cpy_f16_f16", "copy.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}}); + + string_to_spv("add_f32", "add.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); + string_to_spv("add_f16_f32_f16", "add.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float16_t"}, {"FLOAT_TYPE", "float"}}); + + string_to_spv("acc_f32", "acc.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); + + string_to_spv("split_k_reduce", "mul_mat_split_k_reduce.comp", {}); + + string_to_spv("mul_f32", "mul.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); + + string_to_spv("div_f32", "div.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); + + string_to_spv("repeat_f32", "repeat.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + + string_to_spv("scale_f32", "scale.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); + + string_to_spv("sqr_f32", "square.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); + + string_to_spv("sin_f32", "sin.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); + + string_to_spv("cos_f32", "cos.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); + + string_to_spv("clamp_f32", "clamp.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}}); + + string_to_spv("pad_f32", "pad.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + + string_to_spv("concat_f32", "concat.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("concat_f16", "concat.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}}); + string_to_spv("concat_i32", "concat.comp", {{"A_TYPE", "int"}, {"B_TYPE", "int"}, {"D_TYPE", "int"}}); + + string_to_spv("upscale_f32", "upscale.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}); + + string_to_spv("gelu_f32", "gelu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("gelu_quick_f32", "gelu_quick.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("silu_f32", "silu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("relu_f32", "relu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("leaky_relu_f32", "leaky_relu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("tanh_f32", "tanh.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + + string_to_spv("diag_mask_inf_f32", "diag_mask_inf.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + + string_to_spv("soft_max_f32", "soft_max.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}})); + string_to_spv("soft_max_f32_f16", "soft_max.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}})); + + string_to_spv("rope_norm_f32", "rope_norm.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("rope_norm_f16", "rope_norm.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}}); + + string_to_spv("rope_neox_f32", "rope_neox.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}}); + string_to_spv("rope_neox_f16", "rope_neox.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}}); + + string_to_spv("argsort_f32", "argsort.comp", {{"A_TYPE", "float"}}); + + string_to_spv("sum_rows_f32", "sum_rows.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); + + string_to_spv("im2col_f32", "im2col.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); + string_to_spv("im2col_f32_f16", "im2col.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}})); + + string_to_spv("timestep_embedding_f32", "timestep_embedding.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); + + string_to_spv("pool2d_f32", "pool2d.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}})); + + for (auto &c : compiles) { + c.wait(); + } } void write_output_files() { @@ -591,12 +508,7 @@ int main(int argc, char** argv) { } } - std::vector> tasks; - process_shaders(tasks); - - for (auto& task : tasks) { - task.get(); - } + process_shaders(); write_output_files();