examples: graph: add int8 sdpa example

Author: TaoLv

examples/CMakeLists.txt

@@ -109,6 +109,7 @@ if(NOT ONEDNN_BUILD_GRAPH)
         ${CMAKE_CURRENT_SOURCE_DIR}/graph/gated_mlp_int4.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/graph/sdpa_bottom_right_causal_mask.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/graph/gqa_training.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/graph/int8_sdpa.cpp
         )
 endif()
 

examples/graph/int8_sdpa.cpp

@@ -0,0 +1,368 @@
+/*******************************************************************************
+* Copyright 2025 Intel Corporation
+*
+* Licensed under the Apache License, Version 2.0 (the "License");
+* you may not use this file except in compliance with the License.
+* You may obtain a copy of the License at
+*
+*     http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing, software
+* distributed under the License is distributed on an "AS IS" BASIS,
+* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+* See the License for the specific language governing permissions and
+* limitations under the License.
+*******************************************************************************/
+
+#include <cassert>
+#include <chrono>
+#include <iomanip>
+#include <iostream>
+#include <memory>
+#include <random>
+#include <string>
+#include <vector>
+
+#include "oneapi/dnnl/dnnl.hpp"
+#include "oneapi/dnnl/dnnl_graph.hpp"
+
+#include "graph_example_utils.hpp"
+
+using namespace dnnl;
+
+using namespace dnnl::graph;
+using layout_type = logical_tensor::layout_type;
+using data_type = logical_tensor::data_type;
+using dim = logical_tensor::dim;
+using dims = logical_tensor::dims;
+
+struct sdpa_dims_t {
+    dim mb;
+    dim seq_len;
+    dim head_num;
+    dim head_size;
+};
+
+static const int min_runs = 4;
+
+// this is changed from the fill_random() function in matmul_perf.cpp.
+void fill_random(std::vector<float> &out) {
+    static std::vector<float> random_data_f;
+    constexpr size_t nrand = 1037;
+
+    if (random_data_f.empty()) {
+        std::mt19937 generator;
+        std::uniform_real_distribution<float> dist_f(-1.0f, 1.0f);
+
+        random_data_f.resize(nrand);
+        for (auto &d : random_data_f)
+            d = dist_f(generator);
+    }
+
+    for (size_t i = 0; i < out.size(); i += nrand) {
+        size_t chunk = std::min(nrand, out.size() - i);
+        std::memcpy(&out[i], random_data_f.data(), chunk * sizeof(float));
+    }
+}
+
+// initialize the mask with first 3/4 elements with 0s and the last 1/4 elements
+// with -inf.
+void fill_mask(std::vector<float> &mask, size_t seq_len) {
+    const size_t pos = seq_len * 3 / 4;
+    for (size_t i = 0; i < mask.size(); ++i) {
+        if (i % seq_len < pos)
+            mask[i] = 0.f;
+        else
+            mask[i] = -1 * std::numeric_limits<float>::infinity();
+    }
+}
+
+const char *get_type_string(logical_tensor::data_type dt) {
+    const char *type_string = "unknown";
+
+#define TYPE_CASE(T) \
+    if (dt == logical_tensor::data_type::T) type_string = #T;
+    TYPE_CASE(f16);
+    TYPE_CASE(f32);
+    TYPE_CASE(bf16);
+    TYPE_CASE(u8);
+    TYPE_CASE(s8);
+#undef TYPE_CASE
+
+    return type_string;
+}
+
+void print_test_case(logical_tensor::data_type dt, const sdpa_dims_t &p) {
+    std::cout << '[' << std::setw(4) << get_type_string(dt);
+    std::cout << " mb = " << p.mb << ", seq_len = " << p.seq_len
+              << ", head_num = " << p.head_num
+              << ", head_size = " << p.head_size;
+    std::cout << "] " << std::flush;
+}
+
+void bench_int8_sdpa(
+        engine::kind ekind, const sdpa_dims_t &p, double time_limit = 0.) {
+    const bool quick_test = (time_limit == 0.);
+    print_test_case(data_type::u8, p);
+
+    allocator alloc = create_allocator(ekind);
+
+    // Create execution dnnl::engine.
+    dnnl::engine eng = make_engine_with_allocator(ekind, 0, alloc);
+    // Create dnnl::stream.
+    dnnl::stream strm(eng);
+
+    // Prepare input and output shapes to construct the sdpa graph.
+    const dims qkv_sz = {p.mb, p.head_num, p.seq_len, p.head_size};
+    const dims score_sz = {p.mb, p.head_num, p.seq_len, p.seq_len};
+    const dims scale_sz = {1};
+    const dims mask_sz = {p.mb, 1, 1, p.seq_len};
+
+    // Incremental IDs used to create logical tensors and operations.
+    size_t id = 0;
+
+    // insert the dequant for u8 query to f32 query
+    auto q_u8
+            = logical_tensor(id++, data_type::u8, qkv_sz, layout_type::strided);
+    auto q_f32 = logical_tensor(
+            id++, data_type::f32, qkv_sz, layout_type::strided);
+    auto q_deq = op(id++, op::kind::Dequantize, "q_deq");
+    q_deq.set_attr<std::string>(op::attr::qtype, "per_tensor");
+    q_deq.set_attr<float>(op::attr::scales, 0.25f);
+    q_deq.set_attr<int64_t>(op::attr::zps, 128);
+    q_deq.add_input(q_u8);
+    q_deq.add_output(q_f32);
+
+    // insert the dequant for u8 key to f32 key
+    auto k_u8
+            = logical_tensor(id++, data_type::u8, qkv_sz, layout_type::strided);
+    auto k_f32 = logical_tensor(
+            id++, data_type::f32, qkv_sz, layout_type::strided);
+    auto k_deq = op(id++, op::kind::Dequantize, "k_deq");
+    k_deq.set_attr<std::string>(op::attr::qtype, "per_tensor");
+    k_deq.set_attr<float>(op::attr::scales, 0.25f);
+    k_deq.set_attr<int64_t>(op::attr::zps, 128);
+    k_deq.add_input(k_u8);
+    k_deq.add_output(k_f32);
+
+    // score = query x key.T.
+    auto score = logical_tensor(
+            id++, data_type::f32, score_sz, layout_type::strided);
+    auto bmm1 = op(id++, op::kind::MatMul, "bmm1");
+    bmm1.set_attr<bool>(op::attr::transpose_b, true);
+    bmm1.add_inputs({q_f32, k_f32});
+    bmm1.add_output(score);
+
+    // scaled_score = score / scale
+    auto scale = logical_tensor(
+            id++, data_type::f32, scale_sz, layout_type::strided);
+    auto scaled_score = logical_tensor(
+            id++, data_type::f32, score_sz, layout_type::strided);
+    auto scale_div = op(id++, op::kind::Divide, "scale_div");
+    scale_div.add_inputs({score, scale});
+    scale_div.add_outputs({scaled_score});
+
+    // masked_score = scaled_score + mask
+    auto mask = logical_tensor(
+            id++, data_type::f32, mask_sz, layout_type::strided);
+    auto masked_score = logical_tensor(
+            id++, data_type::f32, score_sz, layout_type::strided);
+    auto mask_add = op(id++, op::kind::Add, "mask_add");
+    mask_add.add_inputs({scaled_score, mask});
+    mask_add.add_outputs({masked_score});
+
+    // attention_probs = softmax(masked_score)
+    auto probs = logical_tensor(
+            id++, data_type::f32, score_sz, layout_type::strided);
+    auto softmax = op(id++, op::kind::SoftMax, "softmax");
+    softmax.set_attr<int64_t>(op::attr::axis, -1);
+    softmax.add_inputs({masked_score});
+    softmax.add_outputs({probs});
+
+    // quantize the probs from f32 to u8
+    auto probs_u8 = logical_tensor(
+            id++, data_type::u8, score_sz, layout_type::strided);
+    auto p_quant = op(id++, op::kind::Quantize, "p_quant");
+    p_quant.set_attr<std::string>(op::attr::qtype, "per_tensor");
+    p_quant.set_attr<float>(op::attr::scales, 0.25f);
+    p_quant.set_attr<int64_t>(op::attr::zps, 128);
+    p_quant.add_input(probs);
+    p_quant.add_output(probs_u8);
+
+    // dequant the probs from u8 to f32
+    auto probs_f32 = logical_tensor(
+            id++, data_type::f32, score_sz, layout_type::strided);
+    auto p_deq = op(id++, op::kind::Dequantize, "p_deq");
+    p_deq.set_attr<std::string>(op::attr::qtype, "per_tensor");
+    p_deq.set_attr<float>(op::attr::scales, 0.25f);
+    p_deq.set_attr<int64_t>(op::attr::zps, 128);
+    p_deq.add_input(probs_u8);
+    p_deq.add_output(probs_f32);
+
+    // dequant the value from u8 to f32
+    auto v_u8
+            = logical_tensor(id++, data_type::u8, qkv_sz, layout_type::strided);
+    auto v_f32 = logical_tensor(
+            id++, data_type::f32, qkv_sz, layout_type::strided);
+    auto v_deq = op(id++, op::kind::Dequantize, "v_deq");
+    v_deq.set_attr<std::string>(op::attr::qtype, "per_tensor");
+    v_deq.set_attr<float>(op::attr::scales, 0.25f);
+    v_deq.set_attr<int64_t>(op::attr::zps, 128);
+    v_deq.add_input(v_u8);
+    v_deq.add_output(v_f32);
+
+    // attention_output = attention_probs x value.
+    auto output = logical_tensor(
+            id++, data_type::f32, qkv_sz, layout_type::strided);
+    auto bmm2 = op(id++, op::kind::MatMul, "bmm2");
+    bmm2.add_inputs({probs_f32, v_f32});
+    bmm2.add_outputs({output});
+
+    // quantize the output from f32 to u8
+    auto output_u8
+            = logical_tensor(id++, data_type::u8, qkv_sz, layout_type::strided);
+    auto o_quant = op(id++, op::kind::Quantize, "o_quant");
+    o_quant.set_attr<std::string>(op::attr::qtype, "per_tensor");
+    o_quant.set_attr<float>(op::attr::scales, 0.25f);
+    o_quant.set_attr<int64_t>(op::attr::zps, 128);
+    o_quant.add_input(output);
+    o_quant.add_output(output_u8);
+
+    // Construct a sdpa graph with engine kind and operations.
+    dnnl::graph::graph sdpa(ekind);
+    sdpa.add_op(q_deq);
+    sdpa.add_op(k_deq);
+    sdpa.add_op(bmm1);
+    sdpa.add_op(scale_div);
+    sdpa.add_op(mask_add);
+    sdpa.add_op(softmax);
+    sdpa.add_op(p_quant);
+    sdpa.add_op(p_deq);
+    sdpa.add_op(v_deq);
+    sdpa.add_op(bmm2);
+    sdpa.add_op(o_quant);
+    sdpa.finalize();
+
+    // Get partitions from the sdpa graph.
+    std::vector<partition> partitions = sdpa.get_partitions();
+    // This is just for oneDNN testing purpose.
+    if (partitions.size() != 1) {
+        std::cout << "unsupported sdpa" << std::endl;
+        return;
+    }
+
+    // Compile the partition with inputs, outputs, and an engine.
+    compiled_partition cp = partitions[0].compile(
+            {q_u8, k_u8, scale, mask, v_u8}, {output_u8}, eng);
+
+    // Create tensor objects
+    auto ts_query = tensor(q_u8, eng);
+    auto ts_key = tensor(k_u8, eng);
+    auto ts_scale = tensor(scale, eng);
+    auto ts_mask = tensor(mask, eng);
+    auto ts_value = tensor(v_u8, eng);
+    auto ts_output = tensor(output_u8, eng);
+
+    // Allocate user data.
+    std::vector<float> query_data(product(qkv_sz));
+    std::vector<float> key_data(product(qkv_sz));
+    std::vector<float> scale_data(
+            product(scale_sz), (float)std::sqrt(p.head_size));
+    std::vector<float> mask_data(product(mask_sz));
+    std::vector<float> value_data(product(qkv_sz));
+    std::vector<float> output_data(product(qkv_sz));
+
+    fill_random(query_data);
+    fill_random(key_data);
+    fill_random(value_data);
+    fill_mask(mask_data, static_cast<size_t>(p.seq_len));
+
+    // Write data to tensor object's handle.
+    write_to_dnnl_tensor(query_data.data(), ts_query);
+    write_to_dnnl_tensor(key_data.data(), ts_key);
+    write_to_dnnl_tensor(scale_data.data(), ts_scale);
+    write_to_dnnl_tensor(mask_data.data(), ts_mask);
+    write_to_dnnl_tensor(value_data.data(), ts_value);
+
+    // Warmup run.
+    // Execute the compiled partition of sdpa.
+    cp.execute(
+            strm, {ts_query, ts_key, ts_scale, ts_mask, ts_value}, {ts_output});
+
+    // Wait for the computation to finish.
+    strm.wait();
+
+    // First run.
+    auto start_first = std::chrono::steady_clock::now();
+    cp.execute(
+            strm, {ts_query, ts_key, ts_scale, ts_mask, ts_value}, {ts_output});
+    strm.wait();
+    auto end_first = std::chrono::steady_clock::now();
+    std::chrono::duration<double, std::milli> dur_first
+            = end_first - start_first;
+
+    if (quick_test) return;
+
+    // Timing runs.
+    const int runs = std::max(min_runs, int(time_limit / dur_first.count()));
+    auto start = std::chrono::steady_clock::now();
+    for (int i = 0; i <= runs; i++)
+        cp.execute(strm, {ts_query, ts_key, ts_scale, ts_mask, ts_value},
+                {ts_output});
+    strm.wait();
+    auto end = std::chrono::steady_clock::now();
+    std::chrono::duration<double, std::milli> duration = end - start;
+
+    // Display the results.
+    double avg_time = (duration.count() - dur_first.count()) / runs;
+    std::cout << "graph runs: " << runs + 1 << "; ";
+    std::cout << "avg_time: " << avg_time << " ms" << std::endl;
+}
+
+void bad_args() {
+    std::cerr << "Usage: graph-int8-sdpa-cpp [cpu|gpu]\n"
+                 "       graph-int8-sdpa-cpp [cpu|gpu] <mb> <seq_len> "
+                 "<head_num> <head_size>\n\n";
+    throw std::invalid_argument("Incorrect input arguments.");
+}
+
+void bench(engine::kind ekind, const sdpa_dims_t &p, double time_limit = 0.) {
+    try {
+        bench_int8_sdpa(ekind, p, time_limit);
+        get_mem_pool().clear();
+    } catch (dnnl::error &e) {
+        // Catch and report unimplemented cases.
+        if (e.status == dnnl_unimplemented) {
+            std::cout << "unsupported sdpa: " << std::endl;
+        } else
+            throw;
+    }
+}
+
+void sdpa_perf(engine::kind ekind, int argc, char **argv) {
+    // default testing parameters
+    sdpa_dims_t params = {32, 384, 16, 64};
+
+    if (argc > 2) {
+        if (argc == 6) {
+            params.mb = std::atoi(argv[2]);
+            params.seq_len = std::atoi(argv[3]);
+            params.head_num = std::atoi(argv[4]);
+            params.head_size = std::atoi(argv[5]);
+        } else {
+            bad_args();
+        }
+
+        if (params.mb <= 0 || params.seq_len <= 0 || params.head_num <= 0
+                || params.head_size <= 0) {
+            bad_args();
+        }
+    }
+
+    bench(ekind, params, 2000.0 /*ms*/);
+}
+
+int main(int argc, char **argv) {
+    return handle_example_errors(
+            sdpa_perf, parse_engine_kind(argc, argv, 4), argc, argv);
+}