Merge pull request #4 from zpye/halide

add Halide support

README.md

@@ -55,6 +55,19 @@ pip install build/python/
 python test/test_python/test_model.py
 ```
 
+## (Experimental) Halide Programming
+
+Note: Only upsample nearest layer has an implementation of Halide.
+
+1. Set environment `HALIDE_ROOT` for Halide installation path, using release packages from [https://github.com/halide/Halide/releases](https://github.com/halide/Halide/releases) is a good choice.
+
+2. Set `--halide=true` after `xmake config` and build: 
+
+```shell
+xmake config -a x64 -m release --halide=true
+xmake -w --all
+```
+
 ## Reference
 
 [KuiperInfer](https://github.com/zjhellofss/KuiperInfer) -> basic framework
@@ -75,4 +88,6 @@ python test/test_python/test_model.py
 
 [stb](https://github.com/nothings/stb) -> image loader and image writer
 
+[Halide](https://github.com/halide/Halide) -> Halide programming
+
 [tmp](https://github.com/zjhellofss/tmp) -> pnnx models

src/layer/halide/halide_upsample_nearest.cc

@@ -0,0 +1,25 @@
+#include "Halide.h"
+
+class HalideUpsampleNearest : public Halide::Generator<HalideUpsampleNearest> {
+public:
+    Input<float> scale_x{"scale_x"};
+    Input<float> scale_y{"scale_y"};
+
+    Input<Halide::Buffer<float, 4>> input{"input"};
+
+    Output<Halide::Buffer<float, 4>> output{"output"};
+
+    Halide::Var b, y, x, c;
+
+    void generate() {
+        Halide::Expr x_scale =
+            Halide::cast<int32_t>(Halide::cast<float>(x) * scale_x);
+        Halide::Expr y_scale =
+            Halide::cast<int32_t>(Halide::cast<float>(y) * scale_y);
+
+        // TODO: schedule
+        output(c, x, y, b) = input(c, x_scale, y_scale, b);
+    }
+};
+
+HALIDE_REGISTER_GENERATOR(HalideUpsampleNearest, halide_upsample_nearest)

src/layer/halide/xmake.lua

@@ -0,0 +1,16 @@
+target("halide_upsample_nearest")
+    set_kind("binary")
+    set_targetdir("$(buildir)/")
+    add_files("halide_upsample_nearest.cc", "$(env HALIDE_ROOT)/share/Halide/tools/GenGen.cpp")
+    add_options("halide")
+
+    after_build(function () 
+        os.run("xmake run halide_upsample_nearest -g halide_upsample_nearest -o . -f halide_upsample_nearest target=host")
+    end)
+
+target("halide_layers")
+    set_kind("headeronly")
+    add_defines("USE_HALIDE", { public = true })
+    add_includedirs("$(env HALIDE_ROOT)/include/", "$(buildir)/", { public = true })
+    add_linkdirs("$(buildir)/", { public = true })
+    add_links("halide_upsample_nearest", { public = true })

src/layer/upsample.cpp

@@ -2,6 +2,11 @@
 
 #include <cmath>
 
+#if defined(USE_HALIDE)
+#include "HalideBuffer.h"
+#include "halide_upsample_nearest.h"
+#endif
+
 namespace SimpleInfer {
 
 DEFINE_LAYER_REGISTRY(Upsample);
@@ -119,6 +124,25 @@ Status Upsample::Forward(const Tensor& input, Tensor& output) {
     const float scale_factor_w_inv = 1.0f / scale_factor_w_;
 
     // nearest
+#if defined(USE_HALIDE)
+    Halide::Runtime::Buffer<float> input_buffer(
+        input_eigen_tensor.data(),
+        {input_channel, input_width, input_height, input_batch});
+
+    Halide::Runtime::Buffer<float> output_buffer(
+        output_eigen_tensor.data(),
+        {output_channel, output_width, output_height, output_batch});
+
+    int ret = halide_upsample_nearest(scale_factor_w_inv,
+                                      scale_factor_h_inv,
+                                      input_buffer,
+                                      output_buffer);
+    if (0 != ret) {
+        LOG(ERROR) << "error in halide_upsample_nearest: " << ret;
+        return Status::kFail;
+    }
+
+#else
 #if 0
     for (int h = 0; h < output_height; ++h) {
         int h_in = (float)h * scale_factor_h_inv;
@@ -140,6 +164,7 @@ Status Upsample::Forward(const Tensor& input, Tensor& output) {
     output_eigen_tensor.device(*device) = output_eigen_tensor.generate(
         Nearest4D(input_eigen_tensor, scale_factor_h_inv, scale_factor_w_inv));
 #endif
+#endif  // USE_HALIDE
 
     return Status::kSuccess;
 }

test/test_layer/test_upsample.cpp

@@ -5,7 +5,7 @@
 #include <algorithm>
 #include <cmath>
 
-TEST_CASE("Test Upsample layer 0") {
+TEST_CASE("Test Upsample layer 0", "[Upsample]") {
     using namespace SimpleInfer;
 
     // set tensor
@@ -53,7 +53,7 @@ TEST_CASE("Test Upsample layer 0") {
     }
 }
 
-TEST_CASE("Test Upsample layer 1") {
+TEST_CASE("Test Upsample layer 1", "[Upsample]") {
     using namespace SimpleInfer;
 
     // set tensor

xmake.lua

@@ -1,12 +1,28 @@
 add_rules("mode.debug", "mode.release")
 
+set_languages("cxx17")
+
 option("build_python")
     set_default(false)
     set_showmenu(true)
 option_end()
 
+option("halide")
+    set_default(false)
+    set_showmenu(true)
+    add_includedirs("$(env HALIDE_ROOT)/include/", { public = true })
+    add_linkdirs("$(env HALIDE_ROOT)/bin/Release/",
+                 "$(env HALIDE_ROOT)/lib/Release/", { public = true })
+    add_links("Halide")
+    add_rpathdirs("$(env HALIDE_ROOT)/bin/Release/")
+option_end()
+
 includes("3rdparty")
 
+if has_config("halide") then
+    includes("src/layer/halide")
+end
+
 target("simple-infer")
     set_kind("static")
     add_includedirs("include/", { public = true })
@@ -17,28 +33,32 @@ target("simple-infer")
     add_vectorexts("sse", "sse2", "sse3", "ssse3")
     add_vectorexts("avx", "avx2")
 
+    if has_config("halide") then
+        add_deps("halide_layers")
+    end
+
 target("tools")
     set_kind("static")
     add_includedirs("tools/", { public = true })
     add_files("tools/**.cpp")
 
-if has_config("build_python") then
-    target("pybind11_export")
-        before_build(function () 
-            os.cp("python/simpleinfer/", "$(buildir)/python/")
-        end)
-
-        set_kind("shared")
-        set_basename("simpleinfer")
-        set_extension(".pyd")
-        set_targetdir("$(buildir)/python/simpleinfer")
-        add_files("python/pybind11_main.cpp")
-        add_deps("pybind11", "simple-infer")
-
-        after_build(function () 
-            os.cp("python/setup.py.in", "$(buildir)/python/setup.py")
-        end)
-end
+    if has_config("build_python") then
+        target("pybind11_export")
+            before_build(function () 
+                os.cp("python/simpleinfer/", "$(buildir)/python/")
+            end)
+
+            set_kind("shared")
+            set_basename("simpleinfer")
+            set_extension(".pyd")
+            set_targetdir("$(buildir)/python/simpleinfer")
+            add_files("python/pybind11_main.cpp")
+            add_deps("pybind11", "simple-infer")
+
+            after_build(function () 
+                os.cp("python/setup.py.in", "$(buildir)/python/setup.py")
+            end)
+    end
 
 -- tests
 target("test-eigen")