Skip to content
This repository has been archived by the owner on Dec 22, 2021. It is now read-only.

LLVM inlines constant generation for swizzle #371

Open
omnisip opened this issue Oct 5, 2020 · 1 comment
Open

LLVM inlines constant generation for swizzle #371

omnisip opened this issue Oct 5, 2020 · 1 comment
Labels
toolchain

Comments

@omnisip
Copy link

omnisip commented Oct 5, 2020 •
edited 

#include <wasm_simd128.h>
#define SIMDPP_ARCH_X86_SSE4_1 1
#include "simdpp/simd.h"
#include <xmmintrin.h>
#include <tmmintrin.h>
#include <iomanip>
#include <array>
#include <cstdio>
#include <thread>
#include <future>
#include <iostream>

   // SWIZZLE constants
   
  const static __u32x4 sw1 __attribute__((require_constant_initialization)) = {0xffffff00, 0xffffff01, 0xffffff02, 0xffffff03};
   const static __u32x4 sw2  __attribute__((require_constant_initialization))  = {0xffffff04, 0xffffff05, 0xffffff06, 0xffffff07};
   const static __u32x4 sw3  __attribute__((require_constant_initialization)) = {0xffffff08, 0xffffff09, 0xffffff0a, 0xffffff0b};
   const static __u32x4 sw4   __attribute__((require_constant_initialization)) = {0xffffff0c, 0xffffff0d, 0xffffff0e, 0xffffff0f};
   

inline void sumMatrixPass1(uint8_t const* pImgData1, uint8_t const* pImgData2, uint32_t* pSumArray,
       
                           unsigned width, unsigned height, unsigned hBegin, unsigned hEnd) {
   using namespace simdpp;
   uint32x4 zero = make_zero();


      
  for (unsigned h = hBegin; h < hEnd; ++h)
  {
    uint32x4 lastValSplatX = zero;
    uint32x4 lastValSplatY = zero;
    uint32x4 lastValSplatXX = zero;
    uint32x4 lastValSplatYY = zero;
    uint32x4 lastValSplatXY = zero;
    unsigned w = 0;
    for (; w+15 < width; w += 16)
    {
      uint8x16 imgDataXOrig  = uint8x16(load(pImgData1 + h * width + w));
      uint8x16 imgDataYOrig = uint8x16(load(pImgData2 + h * width + w));
        {
        // a_0 a_1 a_2 a_3
            
        uint32v4::base_vector_type   x = wasm_v8x16_swizzle(imgDataXOrig.native(), sw1);
        uint32v4::base_vector_type  y = wasm_v8x16_swizzle(imgDataYOrig.native(), sw1);
        uint32v4::base_vector_type xx = mul_lo(x,x);
        uint32v4::base_vector_type yy = mul_lo(y, y);
        uint32v4::base_vector_type xy = mul_lo(x, y);

        // a_0 a_0+a_1 a_1+a_2 a_2+a_3
        x = add(x, move4_r<1>(x));
        x = add(x, move4_r<2>(x));
        x = add(x, lastValSplatX);
        lastValSplatX = permute4<3,3,3,3>(x);
        store(pSumArray+h*width+w+0*4, x);
        y = add(y, move4_r<1>(y));
        y = add(y, move4_r<2>(y));
        y = add(y, lastValSplatY);
        lastValSplatY = permute4<3,3,3,3>(y);
        store(width*height+pSumArray+h*width+w+0*4, y);
        xx = add(xx, move4_r<1>(xx));
        xx = add(xx, move4_r<2>(xx));
        xx = add(xx, lastValSplatXX);
        lastValSplatXX = permute4<3,3,3,3>(xx);
        store(2*width*height+pSumArray+h*width+w+0*4, xx);
        yy = add(yy, move4_r<1>(yy));
        yy = add(yy, move4_r<2>(yy));
        yy = add(yy, lastValSplatYY);
        lastValSplatYY = permute4<3,3,3,3>(yy);
        store(3*width*height+pSumArray+h*width+w+0*4, yy);
        xy = add(xy, move4_r<1>(xy));
        xy = add(xy, move4_r<2>(xy));
        xy = add(xy, lastValSplatXY);
        lastValSplatXY = permute4<3,3,3,3>(xy);
        store(4*width*height+pSumArray+h*width+w+0*4, xy);
      }
      {
        uint32v4::base_vector_type   x = wasm_v8x16_swizzle(imgDataXOrig.native(), sw2);
        uint32v4::base_vector_type  y = wasm_v8x16_swizzle(imgDataYOrig.native(), sw2);
        uint32v4::base_vector_type xx = mul_lo(x,x);
        uint32v4::base_vector_type yy = mul_lo(y, y);
        uint32v4::base_vector_type xy = mul_lo(x, y);
        // a_0 a_0+a_1 a_1+a_2 a_2+a_3
        x = add(x, move4_r<1>(x));
        x = add(x, move4_r<2>(x));
        x = add(x, lastValSplatX);
        lastValSplatX = permute4<3,3,3,3>(x);
        store(pSumArray+h*width+w+1*4, x);
        y = add(y, move4_r<1>(y));
        y = add(y, move4_r<2>(y));
        y = add(y, lastValSplatY);
        lastValSplatY = permute4<3,3,3,3>(y);
        store(width*height+pSumArray+h*width+w+1*4, y);
        xx = add(xx, move4_r<1>(xx));
        xx = add(xx, move4_r<2>(xx));
        xx = add(xx, lastValSplatXX);
        lastValSplatXX = permute4<3,3,3,3>(xx);
        store(2*width*height+pSumArray+h*width+w+1*4, xx);
        yy = add(yy, move4_r<1>(yy));
        yy = add(yy, move4_r<2>(yy));
        yy = add(yy, lastValSplatYY);
        lastValSplatYY = permute4<3,3,3,3>(yy);
        store(3*width*height+pSumArray+h*width+w+1*4, yy);
        xy = add(xy, move4_r<1>(xy));
        xy = add(xy, move4_r<2>(xy));
        xy = add(xy, lastValSplatXY);
        lastValSplatXY = permute4<3,3,3,3>(xy);
        store(4*width*height+pSumArray+h*width+w+1*4, xy);
      }
      
    {
        uint32v4::base_vector_type   x = wasm_v8x16_swizzle(imgDataXOrig.native(), sw3);
        uint32v4::base_vector_type  y = wasm_v8x16_swizzle(imgDataYOrig.native(), sw3);
        uint32v4::base_vector_type xx = mul_lo(x,x);
        uint32v4::base_vector_type yy = mul_lo(y, y);
        uint32v4::base_vector_type xy = mul_lo(x, y);
        // a_0 a_0+a_1 a_1+a_2 a_2+a_3
        x = add(x, move4_r<1>(x));
        x = add(x, move4_r<2>(x));
        x = add(x, lastValSplatX);
        lastValSplatX = permute4<3,3,3,3>(x);
        store(pSumArray+h*width+w+2*4, x);
        y = add(y, move4_r<1>(y));
        y = add(y, move4_r<2>(y));
        y = add(y, lastValSplatY);
        lastValSplatY = permute4<3,3,3,3>(y);
        store(width*height+pSumArray+h*width+w+2*4, y);
        xx = add(xx, move4_r<1>(xx));
        xx = add(xx, move4_r<2>(xx));
        xx = add(xx, lastValSplatXX);
        lastValSplatXX = permute4<3,3,3,3>(xx);
        store(2*width*height+pSumArray+h*width+w+2*4, xx);
        yy = add(yy, move4_r<1>(yy));
        yy = add(yy, move4_r<2>(yy));
        yy = add(yy, lastValSplatYY);
        lastValSplatYY = permute4<3,3,3,3>(yy);
        store(3*width*height+pSumArray+h*width+w+2*4, yy);
        xy = add(xy, move4_r<1>(xy));
        xy = add(xy, move4_r<2>(xy));
        xy = add(xy, lastValSplatXY);
        lastValSplatXY = permute4<3,3,3,3>(xy);
        store(4*width*height+pSumArray+h*width+w+2*4, xy);
      }
      
          {
        uint32v4::base_vector_type   x = wasm_v8x16_swizzle(imgDataXOrig.native(), sw4);
        uint32v4::base_vector_type  y = wasm_v8x16_swizzle(imgDataYOrig.native(), sw4);
        uint32v4::base_vector_type xx = mul_lo(x,x);
        uint32v4::base_vector_type yy = mul_lo(y, y);
        uint32v4::base_vector_type xy = mul_lo(x, y);
        // a_0 a_0+a_1 a_1+a_2 a_2+a_3
        x = add(x, move4_r<1>(x));
        x = add(x, move4_r<2>(x));
        x = add(x, lastValSplatX);
        lastValSplatX = permute4<3,3,3,3>(x);
        store(pSumArray+h*width+w+2*4, x);
        y = add(y, move4_r<1>(y));
        y = add(y, move4_r<2>(y));
        y = add(y, lastValSplatY);
        lastValSplatY = permute4<3,3,3,3>(y);
        store(width*height+pSumArray+h*width+w+3*4, y);
        xx = add(xx, move4_r<1>(xx));
        xx = add(xx, move4_r<2>(xx));
        xx = add(xx, lastValSplatXX);
        lastValSplatXX = permute4<3,3,3,3>(xx);
        store(2*width*height+pSumArray+h*width+w+3*4, xx);
        yy = add(yy, move4_r<1>(yy));
        yy = add(yy, move4_r<2>(yy));
        yy = add(yy, lastValSplatYY);
        lastValSplatYY = permute4<3,3,3,3>(yy);
        store(3*width*height+pSumArray+h*width+w+3*4, yy);
        xy = add(xy, move4_r<1>(xy));
        xy = add(xy, move4_r<2>(xy));
        xy = add(xy, lastValSplatXY);
        lastValSplatXY = permute4<3,3,3,3>(xy);
        store(4*width*height+pSumArray+h*width+w+3*4, xy);
      }
      
    }
  }
}


.Ltmp149:
        .loc    2 44 68                         # rgb2y-sample.cpp:44:68
        local.get       15
        local.get       4
        i32.add 
.Ltmp150:
        .loc    52 34 9                         # ./simdpp/detail/insn/load.h:34:9
        v128.load       0
        local.tee       26
        i64.const       -1090921693440
.Ltmp151:
        .loc    2 60 42                         # rgb2y-sample.cpp:60:42
        i64x2.splat
        i64.const       -1082331758846
        i64x2.replace_lane      1
        local.tee       27
        v8x16.swizzle

In this particular code sample, only 2 of the swizzles regenerate the shuffle mask in the loop. The other 2 treat them as constants...
@omnisip
Copy link
Author

omnisip commented Oct 5, 2020

@tlively This is what you asked for, I think.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Assignees
No one assigned
Labels
toolchain
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@omnisip @tlively