abadams

 (max(((17 - (v0 % 4))/4), ((max(((min((v1*2), ((v0 % 2) + 5)) + (v0/2))/2), ((v0/4) + v1)) - (((v0 + -6)/4) + v1)) + 2)) <= 1024)
 (max(((17 - (v0 % 4))/4), ((((min((v1*2), ((v0 % 2) + 20)) + (v0/2))/2) - (((v0 + -6)/4) + v1)) + 2)) <= 1024)
 (max(((8 - (((v0*161) + v1) % 2))/2), ((((min((v2*2), 159) + ((v0*161) + v1))/2) - (((((v0*161) + v1) + -3)/2) + v2)) + 2)) <= 1024)
 (max(max(min(v0, 12), (min(min(min(min(min(min(min(min(min(min(min(min(min(min(min(((v1 + -15)/16), (v2 + -1)), ((v1 + -14)/16)), ((v1 + -13)/16)), ((v1 + -2)/16)), ((v1 + -1)/16)), ((v1 + 1)/16)), ((v3 + -2)/16)), ((v3 + -1)/16)), ((v4 + -2)/16)), ((v4 + -1)/16)), ((v5 + -2)/16)), ((v5 + -1)/16)), ((v6 + -2)/16)), ((v6 + -1)/16)), 11) + 1)), 0) <= max(min(v0, 12), 0))
 (max(max(min(((v0 + 131)/144), 2), (min(((min((min(min(min(min(min(min((min((((v1 + v2) + 25)/2), v3) - ((v4*288) + v5)), v6), (min((((v1 + v2) + 29)/2), v3) - ((v4*288) + v5))), (v7 + 13)), v8), v9), (v7 + 15)) + 128), v8) + -140)/144), 1) + 1)), 0) <= max(min(((v0 + 131)

abadams

Keybase proof

I hereby claim:

• I am abadams on github.
• I am abadams (https://keybase.io/abadams) on keybase.
• I have a public key ASAm3P6tql93IPmgJWn3VxppYs7znurwEtj30edT-4Hvsgo

To claim this, I am signing this object:

abadams

ubuntu-on-windows-10 in branch
performance_inner_loop_parallel 	71.7ms
performance_parallel_performance	0.020
bilateral grid  			16.89 / 10.9
blur					7.00
camera pipe				7.162 / 32.739
conv layer				7.09 / 6.03 
interpolate				36.815
lens blur				8.10 / 36.44 
local laplacian				74.23 / 76.138

abadams

 Halide:

  1.42 │ e0:   vmovup -0x60(%rdi),%ymm0
  4.50 │       vmovup -0x40(%rdi),%ymm6
  4.97 │       vmovup -0x20(%rdi),%ymm7
  2.54 │       vmovup (%rdi),%ymm8
 30.71 │       vfmadd (%rsi),%ymm5,%ymm8
  7.82 │       vfmadd -0x20(%rsi),%ymm5,%ymm7
 22.69 │       vfmadd -0x40(%rsi),%ymm5,%ymm6

abadams

// This sucks. foo, bar, and baz are mentioned way too many times, mostly just to capture references to them.
bool my_function(Expr e) {

    Foo foo = ...;
    Bar bar = ...;
    Baz baz = ...;   
    
    class CheckForSpecialNode : public IRVisitor {
        using IRVisitor::visit;

abadams

abadams@abadams-mbp:/Applications/Adobe Photoshop CC 2018/Adobe Photoshop CC 2018.app/Contents/Required/Plug-ins/Extensions/HalideBottlenecks.plugin/Contents/MacOS
$ strings HalideBottlenecks | grep ^Halide

HalideBNBlend8ProcWrapper
HalideGenBNColorOnly_sse41
HalideGenBNColorOnlyConst_sse41
HalideGenBNColorOnlyConstDeep_sse41
HalideGenBNColorOnlyConstFloat_sse41
HalideGenBNColorOnlyDeep_sse41
HalideGenBNColorOnlyFloat_sse41

abadams

.version 4.0
.target sm_50
.address_size 64

	// .globl	kernel_out_s0_y_y___block_id_y // -- Begin function kernel_out_s0_y_y___block_id_y
                                        // @kernel_out_s0_y_y___block_id_y
.visible .entry kernel_out_s0_y_y___block_id_y(
	.param .u32 kernel_out_s0_y_y___block_id_y_param_0,
	.param .u32 kernel_out_s0_y_y___block_id_y_param_1,
	.param .u32 kernel_out_s0_y_y___block_id_y_param_2,

abadams

#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>

int silly_factorial(int i) {
    if (i == 0) {
        int x;
        printf("Hello from %p\n", &x);
        return 1;
    }

abadams

1>------ Rebuild All started: Project: ZERO_CHECK, Configuration: Release x64 ------
1>  Checking Build System
1>  CMake does not need to re-run because D:/b/slave/win-64-trunk/halide-build-Release/CMakeFiles/generate.stamp is up-to-date.
1>  CMake does not need to re-run because D:/b/slave/win-64-trunk/halide-build-Release/src/CMakeFiles/generate.stamp is up-to-date.
1>  CMake does not need to re-run because D:/b/slave/win-64-trunk/halide-build-Release/tools/CMakeFiles/generate.stamp is up-to-date.
1>  CMake does not need to re-run because D:/b/slave/win-64-trunk/halide-build-Release/test/CMakeFiles/generate.stamp is up-to-date.
1>  CMake does not need to re-run because D:/b/slave/win-64-trunk/halide-build-Release/tutorial/CMakeFiles/generate.stamp is up-to-date.
1>  CMake does not need to re-run because D:/b/slave/win-64-trunk/halide-build-Release/util/CMakeFiles/generate.stamp is up-to-date.
2>------ Rebuild All started: Project: bitcode2cpp, Configuration: Release x64 ------
3>------ Rebuild All started: 

abadams

#include "Halide.h"
#include "benchmark.h"

using namespace Halide;

int main(int argc, char **argv) {
    Var x, y, z, yi, xi;
    Func f1, f2, res;
    
    ImageParam input1(type_of<float>(), 3);