Merge pull request #37 from octu0/v1.19.3

v1.19.3

Dockerfile.generator

@@ -5,10 +5,10 @@ WORKDIR /halide
 RUN set -eux && \
  apt update && \
  apt install -y wget clang g++ binutils libpng-dev libjpeg-dev && \
- wget https://github.com/halide/Halide/releases/download/v13.0.1/Halide-13.0.1-x86-64-linux-fb39d7e3149c0ee1e848bb9957be2ea18765c35d.tar.gz && \
- tar xzf Halide-13.0.1-x86-64-linux-fb39d7e3149c0ee1e848bb9957be2ea18765c35d.tar.gz && \
- rm Halide-13.0.1-x86-64-linux-fb39d7e3149c0ee1e848bb9957be2ea18765c35d.tar.gz && \
- mv Halide-13.0.1-x86-64-linux Halide-Runtime
+ wget https://github.com/halide/Halide/releases/download/v14.0.0/Halide-14.0.0-x86-64-linux-6b9ed2afd1d6d0badf04986602c943e287d44e46.tar.gz && \
+ tar xzf Halide-14.0.0-x86-64-linux-6b9ed2afd1d6d0badf04986602c943e287d44e46.tar.gz && \
+ rm Halide-14.0.0-x86-64-linux-6b9ed2afd1d6d0badf04986602c943e287d44e46.tar.gz && \
+ mv Halide-14.0.0-x86-64-linux Halide-Runtime
 
 COPY docker-entrypoint.generator.sh /usr/local/bin/docker-entrypoint.generator.sh
 ENTRYPOINT [ "docker-entrypoint.generator.sh" ]

Makefile

@@ -4,7 +4,7 @@ _NAME = $(shell grep -o 'AppName string = "[^"]*"' $(VERSION_GO) | cut -d
 _VERSION = $(shell grep -oE 'Version string = "[0-9]+\.[0-9]+\.[0-9]+"' $(VERSION_GO) | cut -d '"' -f2)
 
 _HALIDE = "generator"
-_HALIDE_VER = "13.0.1"
+_HALIDE_VER = "14.0.0"
 
 .PHONY: vet
 vet:
@@ -27,21 +27,21 @@ setup-halide-runtime_linux:
 ifeq ($(shell [ -d detector/Halide-Runtime ] && echo "1"),1)
  @echo "detector/Halide-Runtime exists"
 else
- curl -O -sSL https://github.com/halide/Halide/releases/download/v13.0.1/Halide-13.0.1-x86-64-linux-fb39d7e3149c0ee1e848bb9957be2ea18765c35d.tar.gz
- tar xzf Halide-13.0.1-x86-64-linux-fb39d7e3149c0ee1e848bb9957be2ea18765c35d.tar.gz
- mv Halide-13.0.1-x86-64-linux detector/Halide-Runtime
- rm Halide-13.0.1-x86-64-linux-fb39d7e3149c0ee1e848bb9957be2ea18765c35d.tar.gz
+ curl -O -sSL https://github.com/halide/Halide/releases/download/v14.0.0/Halide-14.0.0-x86-64-linux-6b9ed2afd1d6d0badf04986602c943e287d44e46.tar.gz
+ tar xzf Halide-14.0.0-x86-64-linux-6b9ed2afd1d6d0badf04986602c943e287d44e46.tar.gz
+ mv Halide-14.0.0-x86-64-linux detector/Halide-Runtime
+ rm Halide-14.0.0-x86-64-linux-6b9ed2afd1d6d0badf04986602c943e287d44e46.tar.gz
 endif
 
 .PHONY: setup-halide-runtime_darwin
 setup-halide-runtime_darwin:
 ifeq ($(shell [ -d detector/Halide-Runtime ] && echo "1"),1)
  @echo "detector/Halide-Runtime exists"
 else
- curl -O -sSL https://github.com/halide/Halide/releases/download/v13.0.1/Halide-13.0.1-x86-64-osx-fb39d7e3149c0ee1e848bb9957be2ea18765c35d.tar.gz
- tar xzf Halide-13.0.1-x86-64-osx-fb39d7e3149c0ee1e848bb9957be2ea18765c35d.tar.gz
- mv Halide-13.0.1-x86-64-osx ./Halide-Runtime
- rm Halide-13.0.1-x86-64-osx-fb39d7e3149c0ee1e848bb9957be2ea18765c35d.tar.gz
+ curl -O -sSL https://github.com/halide/Halide/releases/download/v14.0.0/Halide-14.0.0-x86-64-osx-6b9ed2afd1d6d0badf04986602c943e287d44e46.tar.gz
+ tar xzf Halide-14.0.0-x86-64-osx-6b9ed2afd1d6d0badf04986602c943e287d44e46.tar.gz
+ mv Halide-14.0.0-x86-64-osx ./Halide-Runtime
+ rm Halide-14.0.0-x86-64-osx-6b9ed2afd1d6d0badf04986602c943e287d44e46.tar.gz
 endif
 
 .PHONY: setup-halide-runtime

README.md

@@ -19,57 +19,57 @@ darwin/amd64 Intel(R) Core(TM) i7-8569U CPU @ 2.80GHz
 
 ```
 src 320x240
-BenchmarkJIT/cloneimg : 0.00752ms
-BenchmarkJIT/convert_from_argb : 0.02369ms
-BenchmarkJIT/convert_from_abgr : 0.03711ms
-BenchmarkJIT/convert_from_bgra : 0.02472ms
-BenchmarkJIT/convert_from_rabg : 0.03139ms
-BenchmarkJIT/convert_from_yuv_420 : 0.02957ms
-BenchmarkJIT/convert_from_yuv_444 : 0.02586ms
-BenchmarkJIT/convert_to_yuv_420 : 0.05634ms
-BenchmarkJIT/convert_to_yuv_444 : 0.06963ms
-BenchmarkJIT/rotate0 : 0.00739ms
-BenchmarkJIT/rotate90 : 0.02580ms
-BenchmarkJIT/rotate180 : 0.00746ms
-BenchmarkJIT/rotate270 : 0.02557ms
-BenchmarkJIT/crop : 0.06071ms
-BenchmarkJIT/scale : 0.14003ms
-BenchmarkJIT/scale_box : 0.19936ms
-BenchmarkJIT/scale_linear : 0.19754ms
-BenchmarkJIT/scale_gaussian : 0.22766ms
-BenchmarkJIT/blend_normal : 0.08383ms
-BenchmarkJIT/blend_sub : 0.08447ms
-BenchmarkJIT/blend_add : 0.08394ms
-BenchmarkJIT/blend_diff : 0.08423ms
-BenchmarkJIT/grayscale : 0.03839ms
-BenchmarkJIT/invert : 0.04330ms
-BenchmarkJIT/brightness : 0.04931ms
-BenchmarkJIT/gammacorrection : 0.08158ms
-BenchmarkJIT/contrast : 0.01506ms
-BenchmarkJIT/boxblur : 0.12091ms
-BenchmarkJIT/gaussianblur : 0.32293ms
-BenchmarkJIT/blockmozaic : 0.27398ms
-BenchmarkJIT/erosion : 0.12039ms
-BenchmarkJIT/dilation : 0.12439ms
-BenchmarkJIT/morphology_open : 0.10255ms
-BenchmarkJIT/morphology_close : 0.10472ms
-BenchmarkJIT/morphology_gradient : 0.08321ms
-BenchmarkJIT/emboss$1 : 0.05385ms
-BenchmarkJIT/laplacian : 0.03204ms
-BenchmarkJIT/highpass : 0.03783ms
-BenchmarkJIT/gradient : 0.03303ms
-BenchmarkJIT/edgedetect : 0.02638ms
-BenchmarkJIT/sobel : 0.06399ms
-BenchmarkJIT/canny : 0.29472ms
-BenchmarkJIT/canny_dilate : 0.36258ms
-BenchmarkJIT/canny_morphology_open : 0.39542ms
-BenchmarkJIT/canny_morphology_close : 0.40479ms
-BenchmarkJIT/match_template_sad : 6.64854ms
-BenchmarkJIT/match_template_ssd : 4.76639ms
-BenchmarkJIT/match_template_ncc : 9.37937ms
-BenchmarkJIT/prepared_match_template_ncc : 6.85107ms
-BenchmarkJIT/match_template_zncc : 13.29085ms
-BenchmarkJIT/prepared_match_template_zncc : 12.07535ms
+BenchmarkJIT/cloneimg : 0.00767ms
+BenchmarkJIT/convert_from_argb : 0.02328ms
+BenchmarkJIT/convert_from_abgr : 0.03573ms
+BenchmarkJIT/convert_from_bgra : 0.02430ms
+BenchmarkJIT/convert_from_rabg : 0.03159ms
+BenchmarkJIT/convert_from_yuv_420 : 0.02964ms
+BenchmarkJIT/convert_from_yuv_444 : 0.02652ms
+BenchmarkJIT/convert_to_yuv_420 : 0.05654ms
+BenchmarkJIT/convert_to_yuv_444 : 0.07356ms
+BenchmarkJIT/rotate0 : 0.00828ms
+BenchmarkJIT/rotate90 : 0.02600ms
+BenchmarkJIT/rotate180 : 0.00792ms
+BenchmarkJIT/rotate270 : 0.02560ms
+BenchmarkJIT/crop : 0.06128ms
+BenchmarkJIT/scale : 0.13941ms
+BenchmarkJIT/scale_box : 0.20701ms
+BenchmarkJIT/scale_linear : 0.20637ms
+BenchmarkJIT/scale_gaussian : 0.31937ms
+BenchmarkJIT/blend_normal : 0.09480ms
+BenchmarkJIT/blend_sub : 0.08381ms
+BenchmarkJIT/blend_add : 0.08439ms
+BenchmarkJIT/blend_diff : 0.08445ms
+BenchmarkJIT/grayscale : 0.03752ms
+BenchmarkJIT/invert : 0.03684ms
+BenchmarkJIT/brightness : 0.04142ms
+BenchmarkJIT/gammacorrection : 0.07600ms
+BenchmarkJIT/contrast : 0.01532ms
+BenchmarkJIT/boxblur : 0.10674ms
+BenchmarkJIT/gaussianblur : 0.31972ms
+BenchmarkJIT/blockmozaic : 0.27346ms
+BenchmarkJIT/erosion : 0.11407ms
+BenchmarkJIT/dilation : 0.11997ms
+BenchmarkJIT/morphology_open : 0.14157ms
+BenchmarkJIT/morphology_close : 0.10427ms
+BenchmarkJIT/morphology_gradient : 0.07612ms
+BenchmarkJIT/emboss  : 0.06083ms
+BenchmarkJIT/laplacian : 0.04251ms
+BenchmarkJIT/highpass : 0.03843ms
+BenchmarkJIT/gradient : 0.03320ms
+BenchmarkJIT/edgedetect : 0.02701ms
+BenchmarkJIT/sobel : 0.06392ms
+BenchmarkJIT/canny : 0.28839ms
+BenchmarkJIT/canny_dilate : 0.34880ms
+BenchmarkJIT/canny_morphology_open : 0.38943ms
+BenchmarkJIT/canny_morphology_close : 0.39011ms
+BenchmarkJIT/match_template_sad : 5.69188ms
+BenchmarkJIT/match_template_ssd : 4.75666ms
+BenchmarkJIT/match_template_ncc : 8.98426ms
+BenchmarkJIT/prepared_match_template_ncc : 6.23328ms
+BenchmarkJIT/match_template_zncc : 12.64066ms
+BenchmarkJIT/prepared_match_template_zncc : 11.67131ms
 ```
 
 ## AOT benchmarks

blurry.cpp

@@ -608,7 +608,8 @@ Func filter2d_gray(
  );
 
  conv.compute_at(gradient, yi)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
 
  gradient.compute_at(in, ti)
  .tile(x, y, xo, yo, xi, yi, 32, 32)
@@ -707,7 +708,8 @@ Func gaussian(Func in, Expr sigma, RDom rd, const char *name) {
  gaussian(x, y) += cast<uint8_t>(val / center_val);
 
  sum_kernel.compute_at(gaussian, y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  return gaussian;
 }
 
@@ -781,11 +783,14 @@ Func canny(Func in, Param<int32_t> threshold_max, Param<int32_t> threshold_min)
 
  gauss.compute_at(hy, ti)
  .vectorize(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  gy.compute_at(hy, ti)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  gx.compute_at(hy, ti)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
 
  nms.compute_at(hy, ti)
  .vectorize(x);
@@ -1971,10 +1976,12 @@ Func sobel_fn(Func input, Param<int32_t> width, Param<int32_t> height){
 
  gy.compute_at(sobel, yi)
  .parallel(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  gx.compute_at(sobel, yi)
  .parallel(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
 
  sobel.compute_at(in, ti)
  .tile(x, y, xo, yo, xi, yi, 32, 32)
@@ -2119,7 +2126,8 @@ Func emboss_fn(Func input, Param<int32_t> width, Param<int32_t> height){
  );
 
  conv.compute_at(emboss, yi)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
 
  emboss.compute_at(in, ti)
  .tile(x, y, xo, yo, xi, yi, 32, 32)
@@ -2552,7 +2560,7 @@ Func linearsum_xy(Func in, Expr size, Expr xfactor, Expr yfactor) {
  Var x("x"), y("y");
 
  RDom rd = RDom(0, size, "rd_linearsum");
- Func f = Func("linearsum");
+ Func f = Func("linearsum_xy");
  f(x, y) += in(x + (rd * xfactor), y + (rd * yfactor));
  return f;
 }
@@ -2869,28 +2877,36 @@ Func contour_line(Func binary_input, Expr width, Expr height, Expr size) {
 
  next_top.compute_at(f, ti)
  .vectorize(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  next_top_right.compute_at(f, ti)
  .vectorize(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  next_right.compute_at(f, ti)
  .vectorize(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  next_bottom_right.compute_at(f, ti)
  .vectorize(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  next_bottom.compute_at(f, ti)
  .vectorize(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  next_bottom_left.compute_at(f, ti)
  .vectorize(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  next_left.compute_at(f, ti)
  .vectorize(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
  next_top_left.compute_at(f, ti)
  .vectorize(y)
- .vectorize(x);
+ .vectorize(x)
+ .update(0).unscheduled();
 
  nb.compute_at(f, ti)
  .vectorize(y, 8)