rygorous

void bc7_encode_single_color_block(U8 * output_bc7, const U8 rgba[4])
{
	U64 r = rgba[0];
	U64 g = rgba[1];
	U64 b = rgba[2];
	U64 a = rgba[3];

	const U64 bit6_mask = (0x40 << 8) | (0x40 << 22) | (0x40ull << 36);
	const U64 lo7_mask  = (0x7f << 8) | (0x7f << 22) | (0x7full << 36);
	U64 color_bits;

rygorous

reading: c:\devel\media\bc1speedlevel/M_MED_Kurohomura_Backpack_Textures_T_Kurohomura_BP_backpack_D.uasset_build.mip0.png
CompressBC7      : 65.720 millis, 3.51 kc/B, rate= 997.20 kB/s
per-pixel rmse : 0.6425
mode stats (8=invalid 9=solid):
[0]     567 (  0.87%)
[1]    5465 (  8.34%)
[2]     126 (  0.19%)
[3]   15855 ( 24.19%)
[4]     909 (  1.39%)
[5]   15072 ( 23.00%)

rygorous

Cixin Liu, "The Three-Body Problem", Chapter 4 "The Frontriers of Science"

Page numbers given from trade paperback, ISBN 978 0-7653-8203-0

Pg. 53: 

  Shi and the young cop said nothing. Both turned and went down the stairs. The two
  army officers watched them leave and seemed to sigh with relief.
  
  "What's wrong with that guy?" the major whispered to the other officer.

rygorous

#version 430 core

layout(local_size_x=1) in;

layout(binding=0) uniform isamplerBuffer coeff_data;

layout(binding=0, std430) restrict buffer debug_buf {
    uint debug[];
};

rygorous

#include <stdio.h>
#include <stdlib.h>
#include <GL/gl3w.h>
#include "util.h"
#include "radglx.h"

#define CHECK_RESULTS           // turn this off and we'll keep running until the driver crashes.
#define MIN_ALIGN       1       // setting this to 128 (or higher) fixes the bug (both incorrect results and crash)

typedef unsigned char U8;

rygorous

mov  rax, [codewords]
pext rax, [masks] ; coalesce code words
shlx rax, rax, rDestBitPos ; still need something like this
or   rOut, rax ; and this
add  rDestBitPos, [total] ; this and make SIMD code emit it, or load masks to a reg then do popcnt on it?

; so 5-6 insns per 4 codewords = 1.25-1.5 insns per codeword
; leaves 2.5-2.75 insns per codeword to assemble codewords SIMD
; our best bet here is, I guess, 8 16-bit codewords at once (AVX2)
;

rygorous

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

#ifdef __RADAVX__
#include <immintrin.h>
#endif

rygorous

// Returns 8 bit fields at the given positions (in bits) and of the
// given widths as 16-bit integers, with the values aligned with the
// MSB at the top and garbage in the lower-order bits.
//
// The individual lens must be <=8, the positions are bit offsets
// into the 128-bit "bytes".
template<
	int pos0, int len0,
	int pos1, int len1,
	int pos2, int len2,

rygorous

C:\devel\libs\icbc>\devel\projects\oodle2\cdepbuild\win64_release\textest bc1 c:\devel\media\blops3bc1\t7_decal_grunge_dust_wall_01_c_BC1_UNORM_sRGB.tga out_baseline.dds
test_framework 2.9.9 built Nov  8 2022 16:18:59 on Win-x64 msvc-1929
test_bc1
reading: c:\devel\media\blops3bc1\t7_decal_grunge_dust_wall_01_c_BC1_UNORM_sRGB.tga
auto alpha : RGBA with all Opaque : no alpha
CompressBCN      : 64.330 millis, 858.90 c/B, rate= 4.07 MB/s
writing: out_baseline.dds
t7_decal_grunge_dust_wall_01_c_BC1_UNORM_sRGB.tga BC1: rmse=3.4901 hash=0x7b37fc442d36f112
vendor rmse: 3.4953(AMD), 3.4948(NV), 3.4929(Intel), 3.5382(D3D)
rmse_total = 3.490, combined hash=0xe31c1ffb4f7950d3

rygorous

C:\devel\libs\icbc>cl /nologo icbc_test.cpp /O2 /arch:AVX2 && icbc_test -dec intel
icbc_test.cpp
Using 32 threads.
Encoding 'c:/devel/media/blops3bc1/t7_decal_grunge_dust_wall_01_c_BC1_UNORM_sRGB.tga':  RMSE = 3.510    PSNR = 37.224   TIME = 0.052029 (0.052029)
Encoding 'c:/devel/media/blops3bc1/c_hro_sarah_armor_c_BC1_UNORM_sRGB.tga':     RMSE = 4.406    PSNR = 35.251   TIME = 0.012941 (0.012941)
Average Results:
        RMSE = 3.707    PSNR = 36.751   TIME = 0.064970 (0.064970)

C:\devel\libs\icbc>\devel\projects\oodle2\cdepbuild\win64_release\textest bc1 c:\devel\media\blops3bc1\t7_decal_grunge_dust_wall_01_c_BC1_UNORM_sRGB.tga -e2
test_framework 2.9.9 built Nov  8 2022 16:18:59 on Win-x64 msvc-1929