Merge commit '04b33820291e8574ea2618aead9cd0be6a4df56b'

f4a71b1337 README.md: tweak credits
a8a25e17b5 update readme version numbers
0bc88af4de stb_image: optimizations
0ca75da4ec stb_image_resize2: remove whitespace
9d924f8a47 stb_image_resize: 2.04
4da08a1dbd stb_image: create_png_image_raw restructuring
c6d7c32e5d stb_image: Two warning fixes
07268cbf36 stb_image: New Paeth filter
ed64333410 tests: test_png_regress
45eb4ac158 global: very basic .gitignore for object files
e5f0e18d0f stb_image: Small PNG filter simplification
d373674115 stb_image: Fix zlib decoder end-of-stream handling
03f50e343d security notice
1d878bd2a3 security notice
beebb24b94 README.md: fix reference to stb_image_resize2.h
e81f294b16 stb_image_resize: switch GNU/clang from "asm" to "__asm__" for -c99 compatibility
c4bbb6e75f stb_image_resize2.h 2.00
REVERT: c39c7023eb stb_image: create_png_image_raw restructuring
REVERT: 8c3aa05487 stb_image: Two warning fixes
REVERT: 3aa1744a29 stb_image: New Paeth filter
REVERT: 2f3b7e47fa tests: test_png_regress
REVERT: d647af9ad2 global: very basic .gitignore for object files
REVERT: 5a44133dc5 stb_image: Small PNG filter simplification
REVERT: 9f1776a36d stb_image: Fix zlib decoder end-of-stream handling

git-subtree-dir: external/stb/stb
git-subtree-split: f4a71b13373436a2866c5d68f8f80ac6f0bc1ffe

external/stb/stb/README.md

@@ -5,16 +5,18 @@ stb
 
 single-file public domain (or MIT licensed) libraries for C/C++
 
+# This project discusses security-relevant bugs in public in Github Issues and Pull Requests, and it may take significant time for security fixes to be implemented or merged. If this poses an unreasonable risk to your project, do not use stb libraries.
+
 Noteworthy:
 
 * image loader: [stb_image.h](stb_image.h)
 * image writer: [stb_image_write.h](stb_image_write.h)
-* image resizer: [stb_image_resize.h](stb_image_resize.h)
+* image resizer: [stb_image_resize2.h](stb_image_resize2.h)
 * font text rasterizer: [stb_truetype.h](stb_truetype.h)
 * typesafe containers: [stb_ds.h](stb_ds.h)
 
-Most libraries by stb, except: stb_dxt by Fabian "ryg" Giesen, stb_image_resize
-by Jorge L. "VinoBS" Rodriguez, and stb_sprintf by Jeff Roberts.
+Most libraries by stb, except: stb_dxt by Fabian "ryg" Giesen, original stb_image_resize
+by Jorge L. "VinoBS" Rodriguez, and stb_image_resize2 and stb_sprintf by Jeff Roberts.
 
 <a name="stb_libs"></a>
 
@@ -22,10 +24,10 @@ library    | lastest version | category | LoC | description
 --------------------- | ---- | -------- | --- | --------------------------------
 **[stb_vorbis.c](stb_vorbis.c)** | 1.22 | audio | 5584 | decode ogg vorbis files from file/memory to float/16-bit signed output
 **[stb_hexwave.h](stb_hexwave.h)** | 0.5 | audio | 680 | audio waveform synthesizer
-**[stb_image.h](stb_image.h)** | 2.28 | graphics | 7987 | image loading/decoding from file/memory: JPG, PNG, TGA, BMP, PSD, GIF, HDR, PIC
+**[stb_image.h](stb_image.h)** | 2.29 | graphics | 7985 | image loading/decoding from file/memory: JPG, PNG, TGA, BMP, PSD, GIF, HDR, PIC
 **[stb_truetype.h](stb_truetype.h)** | 1.26 | graphics | 5077 | parse, decode, and rasterize characters from truetype fonts
 **[stb_image_write.h](stb_image_write.h)** | 1.16 | graphics | 1724 | image writing to disk: PNG, TGA, BMP
-**[stb_image_resize.h](stb_image_resize.h)** | 0.97 | graphics | 2634 | resize images larger/smaller with good quality
+**[stb_image_resize2.h](stb_image_resize2.h)** | 2.04 | graphics | 10325 | resize images larger/smaller with good quality
 **[stb_rect_pack.h](stb_rect_pack.h)** | 1.01 | graphics | 623 | simple 2D rectangle packer with decent quality
 **[stb_perlin.h](stb_perlin.h)** | 0.5 | graphics | 428 | perlin's revised simplex noise w/ different seeds
 **[stb_ds.h](stb_ds.h)** | 0.67 | utility | 1895 | typesafe dynamic array and hash tables for C, will compile in C++
@@ -43,7 +45,7 @@ library    | lastest version | category | LoC | description
 **[stb_include.h](stb_include.h)** | 0.02 | misc | 295 | implement recursive #include support, particularly for GLSL
 
 Total libraries: 21
-Total lines of C code: 43117
+Total lines of C code: 50806
 
 
 FAQ

external/stb/stb/stb_image.h

@@ -1,4 +1,4 @@
-/* stb_image - v2.28 - public domain image loader - http://nothings.org/stb
+/* stb_image - v2.29 - public domain image loader - http://nothings.org/stb
                                   no warranty implied; use at your own risk
 
    Do this:
@@ -48,6 +48,7 @@ LICENSE
 
 RECENT REVISION HISTORY:
 
+      2.29  (2023-05-xx) optimizations
       2.28  (2023-01-29) many error fixes, security errors, just tons of stuff
       2.27  (2021-07-11) document stbi_info better, 16-bit PNM support, bug fixes
       2.26  (2020-07-13) many minor fixes

external/stb/stb/stb_image_resize_test/dotimings.c

@@ -0,0 +1,224 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#ifdef _MSC_VER
+
+#define stop() __debugbreak()
+#include <windows.h>
+#define int64 __int64
+#pragma warning(disable:4127)
+
+#define get_milliseconds GetTickCount
+
+#else
+
+#define stop() __builtin_trap()
+#define int64 long long
+
+typedef unsigned int U32;
+typedef unsigned long long U64;
+
+#include <time.h>
+static int get_milliseconds()
+{
+  struct timespec ts;
+  clock_gettime( CLOCK_MONOTONIC, &ts );
+  return (U32) ( ( ((U64)(U32)ts.tv_sec) * 1000LL ) + (U64)(((U32)ts.tv_nsec+500000)/1000000) );
+}
+
+#endif
+
+#if defined(TIME_SIMD)
+  // default for most platforms
+#elif defined(TIME_SCALAR)
+  #define STBIR_NO_SIMD
+#else
+  #error You must define TIME_SIMD or TIME_SCALAR when compiling this file.
+#endif
+
+#define STBIR_PROFILE
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#define STBIR__V_FIRST_INFO_BUFFER v_info
+#include "stb_image_resize2.h"  // new one!
+
+#if defined(TIME_SIMD)  && !defined(STBIR_SIMD)
+#error Timing SIMD, but scalar was ON!
+#endif
+
+#if defined(TIME_SCALAR)  && defined(STBIR_SIMD)
+#error Timing scalar, but SIMD was ON!
+#endif
+
+#define HEADER 32
+
+
+static int file_write( const char *filename, void * buffer, size_t size ) 
+{
+  FILE * f = fopen( filename, "wb" );
+  if ( f == 0 ) return 0;
+  if ( fwrite( buffer, 1, size, f) != size ) return 0;
+  fclose(f);
+  return 1;
+}
+
+int64 nresize( void * o, int ox, int oy, int op, void * i, int ix, int iy, int ip, int buf, int type, int edg, int flt )
+{
+  STBIR_RESIZE resize;
+  int t;
+  int64 b;
+
+  stbir_resize_init( &resize, i, ix, iy, ip, o, ox, oy, op, buf, type );
+  stbir_set_edgemodes( &resize, edg, edg );
+  stbir_set_filters( &resize, flt, flt );
+  
+  stbir_build_samplers_with_splits( &resize, 1 );
+
+  b = 0x7fffffffffffffffULL;
+  for( t = 0 ; t < 16 ; t++ )
+  {
+    STBIR_PROFILE_INFO profile;
+    int64 v;
+    if(!stbir_resize_extended( &resize ) )
+      stop();
+    stbir_resize_extended_profile_info( &profile, &resize );
+    v = profile.clocks[1]+profile.clocks[2];
+    if ( v < b )
+    {
+      b = v;
+      t = 0;
+    }
+  }
+
+  stbir_free_samplers( &resize );
+
+  return b;
+}
+
+
+#define INSIZES 5
+#define TYPESCOUNT 5
+#define NUM 64
+
+static const int sizes[INSIZES]={63,126,252,520,772};
+static const int types[TYPESCOUNT]={STBIR_1CHANNEL,STBIR_2CHANNEL,STBIR_RGB,STBIR_4CHANNEL,STBIR_RGBA};
+static const int effective[TYPESCOUNT]={1,2,3,4,7};
+
+int main( int argc, char ** argv )
+{
+  unsigned char * input;
+  unsigned char * output;
+  int dimensionx, dimensiony;
+  int scalex, scaley;
+  int totalms;
+  int timing_count;
+  int ir;
+  int * file;
+  int * ts;
+  int64 totalcycles;
+
+  if ( argc != 6 )
+  {
+    printf("command: dotimings x_samps y_samps x_scale y_scale outfilename\n");
+    exit(1);
+  }
+
+  input = malloc( 4*1200*1200 );
+  memset( input, 0x80, 4*1200*1200 );
+  output = malloc( 4*10000*10000ULL );
+  
+  dimensionx = atoi( argv[1] );
+  dimensiony = atoi( argv[2] );
+  scalex = atoi( argv[3] );
+  scaley = atoi( argv[4] );
+
+  timing_count = dimensionx * dimensiony * INSIZES * TYPESCOUNT;
+
+  file = malloc( sizeof(int) * ( 2 * timing_count + HEADER ) );
+  ts = file + HEADER;
+
+  totalms = get_milliseconds();  
+  totalcycles = STBIR_PROFILE_FUNC();
+  for( ir = 0 ; ir < INSIZES ; ir++ )
+  {
+    int ix, iy, ty;
+    ix = iy = sizes[ir];
+
+    for( ty = 0 ; ty < TYPESCOUNT ; ty++ )
+    {
+      int h, hh;
+
+      h = 1;
+      for( hh = 0 ; hh < dimensiony; hh++ )
+      {
+        int ww, w = 1;
+        for( ww = 0 ; ww < dimensionx; ww++ )
+        {
+          int64 VF, HF;
+          int good;
+        
+          v_info.control_v_first = 2; // vertical first
+          VF = nresize( output, w, h, (w*4*1)&~3, input, ix, iy, ix*4*1, types[ty], STBIR_TYPE_UINT8, STBIR_EDGE_CLAMP, STBIR_FILTER_MITCHELL );
+          v_info.control_v_first = 1; // horizonal first
+          HF = nresize( output, w, h, (w*4*1)&~3, input, ix, iy, ix*4*1, types[ty], STBIR_TYPE_UINT8, STBIR_EDGE_CLAMP, STBIR_FILTER_MITCHELL );
+
+          good = ( ((HF<=VF) && (!v_info.v_first)) || ((VF<=HF) && (v_info.v_first)));
+
+//          printf("\r%d,%d, %d,%d, %d, %I64d,%I64d, // Good: %c(%c-%d)  CompEst: %.1f %.1f\n", ix, iy, w, h, ty, VF, HF, good?'y':'n', v_info.v_first?'v':'h', v_info.v_resize_classification, v_info.v_cost,v_info.h_cost );
+          ts[0] = (int)VF;
+          ts[1] = (int)HF;
+
+          ts += 2;
+
+          w += scalex;
+        }
+        printf(".");
+        h += scaley;  
+      }
+    }
+  }
+  totalms = get_milliseconds() - totalms;  
+  totalcycles = STBIR_PROFILE_FUNC() - totalcycles;
+
+  printf("\n");
+
+  file[0] = 'VFT1';
+
+  #if defined(_x86_64) || defined( __x86_64__ ) || defined( _M_X64 ) || defined(__x86_64) || defined(__SSE2__) || defined( _M_IX86_FP ) || defined(__i386) || defined( __i386__ ) || defined( _M_IX86 ) || defined( _X86_ )
+  file[1] = 1;  // x64
+  #elif defined( _M_AMD64 ) || defined( __aarch64__ ) || defined( __arm64__ ) || defined(__ARM_NEON__) || defined(__ARM_NEON) || defined(__arm__) || defined( _M_ARM )
+  file[1] = 2;  // arm
+  #else
+  file[1] = 99;  // who knows???
+  #endif
+  
+  #ifdef STBIR_SIMD8
+    file[2] = 2;  // simd-8
+  #elif defined( STBIR_SIMD )
+    file[2] = 1;  // simd-4
+  #else
+    file[2] = 0;  // nosimd
+  #endif
+  
+  file[3] = dimensionx; // dimx
+  file[4] = dimensiony; // dimy
+  file[5] = TYPESCOUNT;  // channel types
+  file[ 6] = types[0]; file[7] = types[1]; file[8] = types[2]; file[9] = types[3]; file[10] = types[4];  // buffer_type 
+  file[11] = effective[0]; file[12] = effective[1]; file[13] = effective[2]; file[14] = effective[3]; file[15] = effective[4];  // effective channels 
+  file[16] = INSIZES;  // resizes
+  file[17] = sizes[0]; file[18] = sizes[0]; // input sizes (w x h)
+  file[19] = sizes[1]; file[20] = sizes[1];
+  file[21] = sizes[2]; file[22] = sizes[2];
+  file[23] = sizes[3]; file[24] = sizes[3];
+  file[25] = sizes[4]; file[26] = sizes[4];
+  file[27] = scalex;   file[28] = scaley;  // scale the dimx and dimy amount ( for(i=0;i<dimx) outputx = 1 + i*scalex; )
+  file[29] = totalms;
+  ((int64*)(file+30))[0] = totalcycles;
+
+  if ( !file_write( argv[5], file, sizeof(int) * ( 2 * timing_count + HEADER ) ) )
+    printf( "Error writing file: %s\n", argv[5] );
+  else
+    printf( "Successfully wrote timing file: %s\n", argv[5] );
+
+  return 0;
+}

external/stb/stb/stb_image_resize_test/oldir.c

@@ -0,0 +1,56 @@
+#include <stdio.h>
+#include <stdlib.h>
+
+#ifdef _MSC_VER
+#define stop() __debugbreak()
+#else
+#define stop() __builtin_trap()
+#endif
+
+//#define HEAVYTM
+#include "tm.h"
+
+#define STBIR_SATURATE_INT
+#define STB_IMAGE_RESIZE_STATIC
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include "old_image_resize.h"  
+
+
+static int types[4] =    { STBIR_TYPE_UINT8, STBIR_TYPE_UINT8, STBIR_TYPE_UINT16, STBIR_TYPE_FLOAT };
+static int edges[4] =    { STBIR_EDGE_CLAMP, STBIR_EDGE_REFLECT, STBIR_EDGE_ZERO, STBIR_EDGE_WRAP };
+static int flts[5] =     { STBIR_FILTER_BOX, STBIR_FILTER_TRIANGLE, STBIR_FILTER_CUBICBSPLINE, STBIR_FILTER_CATMULLROM, STBIR_FILTER_MITCHELL };
+static int channels[20] = { 1, 2, 3, 4,      4,4,  2,2,  4,4, 2,2,  4,4, 2,2,  4,4, 2,2 }; 
+static int alphapos[20] = { -1, -1, -1, -1,  3,0,  1,0,   3,0,  1,0,   3,0,  1,0,   3,0,  1,0 }; 
+
+
+void oresize( void * o, int ox, int oy, int op, void * i, int ix, int iy, int ip, int buf, int type, int edg, int flt )
+{
+  int t = types[type];
+  int ic = channels[buf];
+  int alpha = alphapos[buf];
+  int e = edges[edg];
+  int f = flts[flt];
+  int space = ( type == 1 ) ? STBIR_COLORSPACE_SRGB : 0;
+  int flags = ( buf >= 16 ) ? STBIR_FLAG_ALPHA_PREMULTIPLIED : ( ( buf >= 12 ) ? STBIR_FLAG_ALPHA_OUT_PREMULTIPLIED : ( ( buf >= 8 ) ? (STBIR_FLAG_ALPHA_PREMULTIPLIED|STBIR_FLAG_ALPHA_OUT_PREMULTIPLIED) : 0 ) );
+  stbir_uint64 start;
+
+  ENTER( "Resize (old)" );
+  start = tmGetAccumulationStart( tm_mask );
+
+  if(!stbir_resize( i, ix, iy, ip, o, ox, oy, op, t, ic, alpha, flags, e, e, f, f, space, 0 ) )
+    stop();
+
+  #ifdef STBIR_PROFILE
+  tmEmitAccumulationZone( 0, 0, (tm_uint64 *)&start, 0, oldprofile.named.setup, "Setup (old)" );
+  tmEmitAccumulationZone( 0, 0, (tm_uint64 *)&start, 0, oldprofile.named.filters, "Filters (old)" );
+  tmEmitAccumulationZone( 0, 0, (tm_uint64 *)&start, 0, oldprofile.named.looping, "Looping (old)" );
+  tmEmitAccumulationZone( 0, 0, (tm_uint64 *)&start, 0, oldprofile.named.vertical, "Vertical (old)" );
+  tmEmitAccumulationZone( 0, 0, (tm_uint64 *)&start, 0, oldprofile.named.horizontal, "Horizontal (old)" );
+  tmEmitAccumulationZone( 0, 0, (tm_uint64 *)&start, 0, oldprofile.named.decode, "Scanline input (old)" );
+  tmEmitAccumulationZone( 0, 0, (tm_uint64 *)&start, 0, oldprofile.named.encode, "Scanline output (old)" );
+  tmEmitAccumulationZone( 0, 0, (tm_uint64 *)&start, 0, oldprofile.named.alpha, "Alpha weighting (old)" );
+  tmEmitAccumulationZone( 0, 0, (tm_uint64 *)&start, 0, oldprofile.named.unalpha, "Alpha unweighting (old)" );
+  #endif
+
+  LEAVE();
+}

external/stb/stb/stb_image_resize_test/stbirtest.c

@@ -0,0 +1,992 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+//#define HEAVYTM
+#include "tm.h"
+
+#ifdef RADUSETM3
+tm_api * g_tm_api;
+//#define PROFILE_MODE
+#endif
+
+#include <math.h>
+
+#ifdef _MSC_VER
+#define stop() __debugbreak()
+#include <windows.h>
+#define int64 __int64
+#define uint64 unsigned __int64
+#else
+#define stop() __builtin_trap()
+#define int64 long long
+#define uint64 unsigned long long
+#endif
+
+#ifdef _MSC_VER
+#pragma warning(disable:4127)
+#endif
+
+//#define NOCOMP
+
+
+//#define PROFILE_NEW_ONLY
+//#define PROFILE_MODE
+
+
+#if defined(_x86_64) || defined( __x86_64__ ) || defined( _M_X64 ) || defined(__x86_64) || defined(__SSE2__) || defined(STBIR_SSE) || defined( _M_IX86_FP ) || defined(__i386) || defined( __i386__ ) || defined( _M_IX86 ) || defined( _X86_ )
+
+#ifdef _MSC_VER
+
+  uint64 __rdtsc();
+  #define __cycles() __rdtsc()
+
+#else // non msvc
+
+  static inline uint64 __cycles() 
+  {
+    unsigned int lo, hi;
+    asm volatile ("rdtsc" : "=a" (lo), "=d" (hi) );
+    return ( ( (uint64) hi ) << 32 ) | ( (uint64) lo );
+  }
+
+#endif  // msvc
+
+#elif defined( _M_ARM64 ) || defined( __aarch64__ ) || defined( __arm64__ ) || defined(__ARM_NEON__) 
+
+#ifdef _MSC_VER
+
+  #define __cycles() _ReadStatusReg(ARM64_CNTVCT)
+
+#else
+
+  static inline uint64 __cycles()
+  {
+    uint64 tsc;
+    asm volatile("mrs %0, cntvct_el0" : "=r" (tsc));
+    return tsc;
+  }
+
+#endif
+
+#else // x64, arm
+
+#error Unknown platform for timing.
+
+#endif  //x64 and   
+
+
+#ifdef PROFILE_MODE
+
+#define STBIR_ASSERT(cond)
+
+#endif
+
+#ifdef _DEBUG
+#undef STBIR_ASSERT
+#define STBIR_ASSERT(cond) { if (!(cond)) stop(); }
+#endif
+
+
+#define SHRINKBYW 2
+#define ZOOMBYW 2
+#define SHRINKBYH 2
+#define ZOOMBYH 2
+
+
+int mem_count = 0;
+
+#ifdef TEST_WITH_VALLOC
+
+#define STBIR__SEPARATE_ALLOCATIONS
+
+#if TEST_WITH_LIMIT_AT_FRONT
+
+  void * wmalloc(SIZE_T size)
+  {
+     static unsigned int pagesize=0;
+     void* p;
+     SIZE_T s;
+
+     // get the page size, if we haven't yet
+     if (pagesize==0)
+     {
+       SYSTEM_INFO si;
+       GetSystemInfo(&si);
+       pagesize=si.dwPageSize;
+     }
+
+     // we need room for the size, 8 bytes to hide the original pointer and a
+     //   validation dword, and enough data to completely fill one page
+     s=(size+(pagesize-1))&~(pagesize-1);
+
+     // allocate the size plus a page (for the guard)
+     p=VirtualAlloc(0,(SIZE_T)s,MEM_RESERVE|MEM_COMMIT,PAGE_READWRITE);
+     
+     return p;
+   }
+
+  void wfree(void * ptr)
+  {
+    if (ptr)
+    {
+      if ( ((ptrdiff_t)ptr) & 4095 ) stop();
+      if ( VirtualFree(ptr,0,MEM_RELEASE) == 0 ) stop();
+    }
+  }
+
+#else
+
+  void * wmalloc(SIZE_T size)
+  {
+     static unsigned int pagesize=0;
+     void* p;
+     SIZE_T s;
+
+     // get the page size, if we haven't yet
+     if (pagesize==0)
+     {
+       SYSTEM_INFO si;
+       GetSystemInfo(&si);
+       pagesize=si.dwPageSize;
+     }
+
+     // we need room for the size, 8 bytes to hide the original pointer and a
+     //   validation dword, and enough data to completely fill one page
+     s=(size+16+(pagesize-1))&~(pagesize-1);
+
+     // allocate the size plus a page (for the guard)
+     p=VirtualAlloc(0,(SIZE_T)(s+pagesize+pagesize),MEM_RESERVE|MEM_COMMIT,PAGE_READWRITE);
+
+     if (p)
+     {
+       DWORD oldprot;
+       void* orig=p;
+
+       // protect the first page
+       VirtualProtect(((char*)p),pagesize,PAGE_NOACCESS,&oldprot);
+
+       // protect the final page
+       VirtualProtect(((char*)p)+s+pagesize,pagesize,PAGE_NOACCESS,&oldprot);
+
+       // now move the returned pointer so that it bumps right up against the
+       //   the next (protected) page (this may result in unaligned return
+       //   addresses - pre-align the sizes if you always want aligned ptrs)
+//#define ERROR_ON_FRONT
+#ifdef ERROR_ON_FRONT
+       p=((char*)p)+pagesize+16;
+#else
+       p=((char*)p)+(s-size)+pagesize;
+#endif
+
+       // hide the validation value and the original pointer (which we'll
+       //   need used for freeing) right behind the returned pointer
+       ((unsigned int*)p)[-1]=0x98765432;
+       ((void**)p)[-2]=orig;
+       ++mem_count;
+//printf("aloc: %p bytes: %d\n",p,(int)size);
+       return(p);
+     }
+
+     return 0;
+  }
+
+  void wfree(void * ptr)
+  {
+    if (ptr)
+    {
+      int err=0;
+
+      // is this one of our allocations?
+      if (((((unsigned int*)ptr)[-1])!=0x98765432) || ((((void**)ptr)[-2])==0))
+      {
+        err=1;
+      }
+
+      if (err)
+      {
+        __debugbreak();
+      }
+      else
+      {
+
+        // back up to find the original pointer
+        void* p=((void**)ptr)[-2];
+
+        // clear the validation value and the original pointer
+        ((unsigned int*)ptr)[-1]=0;
+        ((void**)ptr)[-2]=0;
+
+//printf("free: %p\n",ptr);
+
+        --mem_count;
+
+        // now free the pages
+        if (p)
+          VirtualFree(p,0,MEM_RELEASE);
+
+      }
+    }
+  }
+
+#endif
+
+#define STBIR_MALLOC(size,user_data) ((void)(user_data), wmalloc(size))
+#define STBIR_FREE(ptr,user_data)    ((void)(user_data), wfree(ptr))
+
+#endif
+
+#define STBIR_PROFILE
+//#define STBIR_NO_SIMD
+//#define STBIR_AVX
+//#define STBIR_AVX2
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include "stb_image_resize2.h"  // new one!
+
+#define STB_IMAGE_WRITE_IMPLEMENTATION
+#include "stb_image_write.h"
+
+int tsizes[5] =   { 1, 1, 2, 4, 2 };
+int ttypes[5] =   { STBIR_TYPE_UINT8, STBIR_TYPE_UINT8_SRGB, STBIR_TYPE_UINT16, STBIR_TYPE_FLOAT, STBIR_TYPE_HALF_FLOAT };
+
+int cedges[4] =   { STBIR_EDGE_CLAMP, STBIR_EDGE_REFLECT, STBIR_EDGE_ZERO, STBIR_EDGE_WRAP };
+int flts[5] =     { STBIR_FILTER_BOX, STBIR_FILTER_TRIANGLE, STBIR_FILTER_CUBICBSPLINE, STBIR_FILTER_CATMULLROM, STBIR_FILTER_MITCHELL };
+int buffers[20] = { STBIR_1CHANNEL, STBIR_2CHANNEL, STBIR_RGB, STBIR_4CHANNEL, 
+                    STBIR_BGRA, STBIR_ARGB, STBIR_RA, STBIR_AR,
+                    STBIR_RGBA_PM, STBIR_ARGB_PM, STBIR_RA_PM, STBIR_AR_PM,
+                    STBIR_RGBA, STBIR_ARGB, STBIR_RA, STBIR_AR,
+                    STBIR_RGBA_PM, STBIR_ARGB_PM, STBIR_RA_PM, STBIR_AR_PM,
+                  };
+int obuffers[20] = { STBIR_1CHANNEL, STBIR_2CHANNEL, STBIR_RGB, STBIR_4CHANNEL, 
+                     STBIR_BGRA, STBIR_ARGB, STBIR_RA, STBIR_AR,
+                     STBIR_RGBA_PM, STBIR_ARGB_PM, STBIR_RA_PM, STBIR_AR_PM,
+                     STBIR_RGBA_PM, STBIR_ARGB_PM, STBIR_RA_PM, STBIR_AR_PM,
+                     STBIR_RGBA, STBIR_ARGB, STBIR_RA, STBIR_AR,
+                   };
+
+int bchannels[20] = { 1, 2, 3, 4,  4,4, 2,2,  4,4, 2,2,  4,4, 2,2,  4,4, 2,2  }; 
+int alphapos[20] = { -1, -1, -1, -1,  3,0,  1,0,   3,0,  1,0,   3,0,  1,0,3,0,  1,0 }; 
+
+
+char const * buffstrs[20] = { "1ch", "2ch", "3ch", "4ch",  "RGBA", "ARGB", "RA", "AR",  "RGBA_both_pre", "ARGB_both_pre", "RA_both_pre", "AR_both_pre",  "RGBA_out_pre", "ARGB_out_pre", "RA_out_pre", "AR_out_pre",  "RGBA_in_pre", "ARGB_in_pre", "RA_in_pre", "AR_in_pre" };
+char const * typestrs[5] =  { "Bytes", "BytesSRGB", "Shorts", "Floats", "Half Floats"};
+char const * edgestrs[4] =  { "Clamp", "Reflect", "Zero", "Wrap" };
+char const * fltstrs[5] =   { "Box", "Triangle", "Cubic", "Catmullrom", "Mitchell" };
+
+#ifdef STBIR_PROFILE
+  static void do_acc_zones( STBIR_PROFILE_INFO * profile )
+  {
+    stbir_uint32 j;
+    stbir_uint64 start = tmGetAccumulationStart( tm_mask ); start=start;
+
+    for( j = 0 ; j < profile->count ; j++ )
+    {
+      if ( profile->clocks[j] )
+        tmEmitAccumulationZone( 0, 0, (tm_uint64*)&start, 0, profile->clocks[j], profile->descriptions[j] );
+    }
+  }
+#else
+  #define do_acc_zones(...)
+#endif
+
+int64 vert;
+
+//#define WINTHREADTEST
+#ifdef WINTHREADTEST
+
+static STBIR_RESIZE * thread_resize;
+static LONG which;
+static int threads_started = 0;
+static HANDLE threads[32];
+static HANDLE starts,stops;
+  
+static DWORD resize_shim( LPVOID p )
+{
+  for(;;)
+  {
+    LONG wh;
+
+    WaitForSingleObject( starts, INFINITE );
+
+    wh = InterlockedAdd( &which, 1 ) - 1;
+
+    ENTER( "Split %d", wh );
+    stbir_resize_split( thread_resize, wh, 1 );
+  #ifdef STBIR_PROFILE
+    { STBIR_PROFILE_INFO profile; stbir_resize_split_profile_info( &profile, thread_resize, wh, 1 ); do_acc_zones( &profile ); vert = profile.clocks[1]; }
+  #endif
+    LEAVE();
+
+    ReleaseSemaphore( stops, 1, 0 );
+  }
+}
+
+#endif
+
+void nresize( void * o, int ox, int oy, int op, void * i, int ix, int iy, int ip, int buf, int type, int edg, int flt )
+{
+  STBIR_RESIZE resize;
+ 
+  stbir_resize_init( &resize, i, ix, iy, ip, o, ox, oy, op, buffers[buf], ttypes[type] );
+  stbir_set_pixel_layouts( &resize, buffers[buf], obuffers[buf] );
+  stbir_set_edgemodes( &resize, cedges[edg], cedges[edg] );
+  stbir_set_filters( &resize, flts[flt], /*STBIR_FILTER_POINT_SAMPLE */ flts[flt] );
+  //stbir_set_input_subrect( &resize, 0.55f,0.333f,0.75f,0.50f);
+  //stbir_set_output_pixel_subrect( &resize, 00, 00, ox/2,oy/2);
+  //stbir_set_pixel_subrect(&resize, 1430,1361,30,30);
+  
+  ENTER( "Resize" );
+ 
+  #ifndef WINTHREADTEST
+
+  ENTER( "Filters" );
+  stbir_build_samplers_with_splits( &resize, 1 );
+  #ifdef STBIR_PROFILE
+    { STBIR_PROFILE_INFO profile; stbir_resize_build_profile_info( &profile, &resize ); do_acc_zones( &profile ); }
+  #endif
+  LEAVE();
+ 
+  ENTER( "Resize" );
+  if(!stbir_resize_extended( &resize ) )
+    stop();
+  #ifdef STBIR_PROFILE
+    { STBIR_PROFILE_INFO profile; stbir_resize_extended_profile_info( &profile, &resize ); do_acc_zones( &profile ); vert = profile.clocks[1]; }
+  #endif
+  LEAVE();
+
+  #else
+  {
+    int c, cnt;
+ 
+    ENTER( "Filters" );
+    cnt =  stbir_build_samplers_with_splits( &resize, 4 );
+    #ifdef STBIR_PROFILE
+      { STBIR_PROFILE_INFO profile; stbir_resize_build_profile_info( &profile, &resize ); do_acc_zones( &profile ); }
+    #endif
+    LEAVE();
+    
+    ENTER( "Thread start" );
+    if ( threads_started == 0 )
+    {
+      starts = CreateSemaphore( 0, 0, 32, 0 );
+      stops = CreateSemaphore( 0, 0, 32, 0 );
+    }
+    for( c = threads_started ; c < cnt ; c++ )
+      threads[ c ] = CreateThread( 0, 2048*1024, resize_shim, 0, 0, 0 );
+    
+    threads_started = cnt;
+    thread_resize = &resize;
+    which = 0;
+    LEAVE();
+
+    // starts the threads
+    ReleaseSemaphore( starts, cnt, 0 );
+
+    ENTER( "Wait" );
+    for( c = 0 ; c < cnt; c++ )
+      WaitForSingleObject( stops, INFINITE );
+    LEAVE();
+  }
+  #endif
+
+  ENTER( "Free" );
+  stbir_free_samplers( &resize );
+  LEAVE();
+  LEAVE();
+}
+
+
+#define STB_IMAGE_IMPLEMENTATION
+#include "stb_image.h"
+
+extern void oresize( void * o, int ox, int oy, int op, void * i, int ix, int iy, int ip, int buf, int type, int edg, int flt );
+
+
+
+#define TYPESTART 0
+#define TYPEEND   4
+
+#define LAYOUTSTART  0
+#define LAYOUTEND   19
+
+#define SIZEWSTART 0
+#define SIZEWEND   2
+
+#define SIZEHSTART 0
+#define SIZEHEND   2
+
+#define EDGESTART 0
+#define EDGEEND   3
+
+#define FILTERSTART 0
+#define FILTEREND   4
+
+#define HEIGHTSTART 0
+#define HEIGHTEND   2
+
+#define WIDTHSTART 0
+#define WIDTHEND   2
+
+
+
+
+static void * convert8to16( unsigned char * i, int w, int h, int c )
+{
+  unsigned short * ret;
+  int p;
+
+  ret = malloc( w*h*c*sizeof(short) );
+  for(p = 0 ; p < (w*h*c) ; p++ )
+  {
+    ret[p]=(short)((((int)i[p])<<8)+i[p]);
+  }
+
+  return ret;
+}
+
+static void * convert8tof( unsigned char * i, int w, int h, int c )
+{
+  float * ret;
+  int p;
+
+  ret = malloc( w*h*c*sizeof(float) );
+  for(p = 0 ; p < (w*h*c) ; p++ )
+  {
+    ret[p]=((float)i[p])*(1.0f/255.0f);
+  }
+
+  return ret;
+}
+
+static void * convert8tohf( unsigned char * i, int w, int h, int c )
+{
+  stbir__FP16 * ret;
+  int p;
+
+  ret = malloc( w*h*c*sizeof(stbir__FP16) );
+  for(p = 0 ; p < (w*h*c) ; p++ )
+  {
+    ret[p]=stbir__float_to_half(((float)i[p])*(1.0f/255.0f));
+  }
+
+  return ret;
+}
+
+static void * convert8tohff( unsigned char * i, int w, int h, int c )
+{
+  float * ret;
+  int p;
+
+  ret = malloc( w*h*c*sizeof(float) );
+  for(p = 0 ; p < (w*h*c) ; p++ )
+  {
+    ret[p]=stbir__half_to_float(stbir__float_to_half(((float)i[p])*(1.0f/255.0f)));
+  }
+
+  return ret;
+}
+
+static int isprime( int v )
+{
+  int i;
+
+  if ( v <= 3 )
+    return ( v > 1 );
+  if ( ( v & 1 ) == 0 )
+    return 0;
+  if ( ( v % 3 ) == 0 )
+    return 0;
+  i = 5;
+  while ( (i*i) <= v )
+  {
+    if ( ( v % i ) == 0 )
+      return 0;
+    if ( ( v % ( i + 2 ) ) == 0 )
+      return 0;
+    i += 6;
+  }
+
+  return 1;
+}
+
+static int getprime( int v )
+{
+  int i;
+  i = 0;
+  for(;;)
+  {
+    if ( i >= v )
+      return v;  // can't find any, just return orig
+    if (isprime(v - i))
+      return v - i;
+    if (isprime(v + i))
+      return v + i;
+    ++i;
+  }
+}
+
+
+int main( int argc, char ** argv )
+{
+  int ix, iy, ic;
+  unsigned char * input[6];
+  char * ir1;
+  char * ir2;
+  int szhs[3];
+  int szws[3];
+  int aw, ah, ac;
+  unsigned char * correctalpha;
+  int layouts, types, heights, widths, edges, filters;
+
+  if ( argc != 2 )
+  {
+    printf("command: stbirtest [imagefile]\n");
+    exit(1);
+  }
+
+  SetupTM( "127.0.0.1" );
+
+  correctalpha = stbi_load( "correctalpha.png", &aw, &ah, &ac, 0 );
+
+  input[0] = stbi_load( argv[1], &ix, &iy, &ic, 0 );
+  input[1] = input[0];
+  input[2] = convert8to16( input[0], ix, iy, ic );
+  input[3] = convert8tof( input[0], ix, iy, ic );
+  input[4] = convert8tohf( input[0], ix, iy, ic );
+  input[5] = convert8tohff( input[0], ix, iy, ic );
+
+  printf("Input %dx%d (%d channels)\n",ix,iy,ic);
+
+  ir1 = malloc( 4 * 4 * 3000 * 3000ULL );
+  ir2 = malloc( 4 * 4 * 3000 * 3000ULL );
+
+  szhs[0] = getprime( iy/SHRINKBYH );
+  szhs[1] = iy;
+  szhs[2] = getprime( iy*ZOOMBYH );
+
+  szws[0] = getprime( ix/SHRINKBYW );
+  szws[1] = ix;
+  szws[2] = getprime( ix*ZOOMBYW );
+
+  #if 1
+  for( types = TYPESTART ; types <= TYPEEND ; types++ )
+  #else
+  for( types = 1 ; types <= 1 ; types++ )
+  #endif
+  {
+    ENTER( "Test type: %s",typestrs[types]);
+    #if 1
+    for( layouts = LAYOUTSTART ; layouts <= LAYOUTEND ; layouts++ )
+    #else
+    for( layouts = 16; layouts <= 16 ; layouts++ )
+    #endif
+    {
+      ENTER( "Test layout: %s",buffstrs[layouts]);
+      
+      #if 0
+      for( heights = HEIGHTSTART ; heights <= HEIGHTEND ; heights++ )
+      {
+        int w, h = szhs[heights];
+      #else
+      for( heights = 0 ; heights <= 11 ; heights++ )
+      {
+        static int szhsz[12]={32, 200, 350, 400, 450, 509, 532, 624, 700, 824, 1023, 2053 };
+        int w, h = szhsz[heights];
+      #endif
+
+        ENTER( "Test height: %d %s %d",iy,(h<iy)?"Down":((h>iy)?"Up":"Same"),h);
+  
+        #if 0
+        for( widths = WIDTHSTART ; widths <= WIDTHEND ; widths++ )
+        {
+          w = szws[widths];
+        #else
+        for( widths = 0 ; widths <= 12 ; widths++ )
+        {
+          static int szwsz[13]={2, 32, 200, 350, 400, 450, 509, 532, 624, 700, 824, 1023, 2053 };
+          w = szwsz[widths];
+        #endif
+
+          ENTER( "Test width: %d %s %d",ix, (w<ix)?"Down":((w>ix)?"Up":"Same"), w);
+
+          #if 0
+          for( edges = EDGESTART ; edges <= EDGEEND ; edges++ )
+          #else
+          for( edges = 0 ; edges <= 0 ; edges++ )
+          #endif
+          {
+            ENTER( "Test edge: %s",edgestrs[edges]);
+            #if 0
+            for( filters = FILTERSTART ; filters <= FILTEREND ; filters++ )
+            #else
+            for( filters = 3 ; filters <= 3 ; filters++ )
+            #endif
+            {
+              int op, opw, np,npw, c, a;
+              #ifdef COMPARE_SAME                        
+              int oldtypes = types;
+              #else
+              int oldtypes = (types==4)?3:types;
+              #endif
+            
+              ENTER( "Test filter: %s",fltstrs[filters]);
+              {
+                c = bchannels[layouts];
+                a = alphapos[layouts];
+
+                op = w*tsizes[oldtypes]*c + 60;
+                opw = w*tsizes[oldtypes]*c;
+
+                np = w*tsizes[types]*c + 60;
+                npw = w*tsizes[types]*c;
+
+                printf( "%s:layout: %s  w: %d h: %d edge: %s filt: %s\n", typestrs[types],buffstrs[layouts], w, h, edgestrs[edges], fltstrs[filters] );
+
+
+                // clear pixel area to different, right edge to zero
+                #ifndef NOCLEAR
+                ENTER( "Test clear padding" );
+                {
+                  int d;
+                  for( d = 0 ; d < h ; d++ )
+                  {
+                    int oofs = d * op;
+                    int nofs = d * np;
+                    memset( ir1 + oofs, 192, opw );
+                    memset( ir1 + oofs+opw, 79, op-opw );
+                    memset( ir2 + nofs, 255, npw );
+                    memset( ir2 + nofs+npw, 79, np-npw );
+                  }
+                }
+                LEAVE();
+
+                #endif
+
+                #ifdef COMPARE_SAME                        
+                #define TIMINGS 1
+                #else 
+                #define TIMINGS 1
+                #endif
+                ENTER( "Test both" );
+                {
+                  #ifndef PROFILE_NEW_ONLY
+                    {
+                      int ttt, max = 0x7fffffff;
+                      ENTER( "Test old" );
+                      for( ttt = 0 ; ttt < TIMINGS ; ttt++ )
+                      {
+                        int64 m = __cycles();
+
+                        oresize( ir1, w, h, op, 
+                        #ifdef COMPARE_SAME                        
+                        input[types], 
+                        #else
+                        input[(types==4)?5:types], 
+                        #endif
+                        ix, iy, ix*ic*tsizes[oldtypes], layouts, oldtypes, edges, filters );
+
+                        m = __cycles() - m;
+                        if ( ( (int)m ) < max )
+                        max = (int) m;
+                      }
+                      LEAVE();
+                      printf("old: %d\n", max );
+                    }
+                  #endif
+     
+                  {
+                    int ttt, max = 0x7fffffff, maxv = 0x7fffffff;
+                    ENTER( "Test new" );
+                    for( ttt = 0 ; ttt < TIMINGS ; ttt++ )
+                    {
+                      int64 m = __cycles();
+
+                      nresize( ir2, w, h, np, input[types], ix, iy, ix*ic*tsizes[types], layouts, types, edges, filters );
+        
+                      m = __cycles() - m;
+                      if ( ( (int)m ) < max )
+                        max = (int) m;
+                      if ( ( (int)vert ) < maxv )
+                        maxv = (int) vert;
+                    }
+                    LEAVE(); // test new
+                    printf("new: %d (v: %d)\n", max, maxv );
+                  }
+                }
+                LEAVE();  // test both
+
+                if ( mem_count!= 0 )
+                  stop();
+
+              #ifndef NOCOMP
+                ENTER( "Test compare" );
+                {
+                  int x,y,ch;
+                  int nums = 0;
+                  for( y = 0 ; y < h ; y++ )
+                  {
+                    for( x = 0 ; x < w ; x++ )
+                    {
+                      switch(types)
+                      {
+                        case 0:
+                        case 1: //SRGB 
+                        {
+                          unsigned char * p1 = (unsigned char *)&ir1[y*op+x*c];
+                          unsigned char * p2 = (unsigned char *)&ir2[y*np+x*c];
+                          for( ch = 0 ; ch < c ; ch++ )
+                          {
+                            float pp1,pp2,d;
+                            float av = (a==-1)?1.0f:((float)p1[a]/255.0f);
+
+                            pp1 = p1[ch];
+                            pp2 = p2[ch];
+
+                            // compare in premult space
+                            #ifndef COMPARE_SAME                        
+                            if ( ( ( layouts >=4 ) && ( layouts <= 7 ) ) || ( ( layouts >=16 ) && ( layouts <= 19 ) ) )
+                            {
+                              pp1 *= av;
+                              pp2 *= av;
+                            }
+                            #endif
+
+                            d = pp1 - pp2;
+                            if ( d < 0 ) d = -d;
+
+                            #ifdef COMPARE_SAME                        
+                            if ( d > 0 ) 
+                            #else 
+                            if ( d > 1 )
+                            #endif
+                            {
+                              printf("Error at %d x %d (chan %d) (d: %g a: %g) [%d %d %d %d] [%d %d %d %d]\n",x,y,ch, d,av, p1[0],p1[1],p1[2],p1[3], p2[0],p2[1],p2[2],p2[3]);
+                              ++nums;
+                              if ( nums > 16 ) goto ex;
+                              //if (d) exit(1);
+                              //goto ex;
+                            }
+                          }  
+                        }
+                        break;
+
+                        case 2:
+                        {
+                          unsigned short * p1 = (unsigned short *)&ir1[y*op+x*c*sizeof(short)];
+                          unsigned short * p2 = (unsigned short *)&ir2[y*np+x*c*sizeof(short)];
+                          for( ch = 0 ; ch < c ; ch++ )
+                          {
+                            float thres,pp1,pp2,d;
+                            float av = (a==-1)?1.0f:((float)p1[a]/65535.0f);
+
+                            pp1 = p1[ch];
+                            pp2 = p2[ch];
+
+                            // compare in premult space
+                            #ifndef COMPARE_SAME                        
+                            if ( ( ( layouts >=4 ) && ( layouts <= 7 ) ) || ( ( layouts >= 16 ) && ( layouts <= 19 ) ) )
+                            {
+                              pp1 *= av;
+                              pp2 *= av;
+                            }
+                            #endif
+
+                            d = pp1 - pp2;
+                            if ( d < 0 ) d = -d;
+
+                            thres=((float)p1[ch]*0.007f)+2.0f;
+                            if (thres<4) thres = 4;
+  
+                            #ifdef COMPARE_SAME                        
+                            if ( d > 0 ) 
+                            #else 
+                            if ( d > thres)
+                            #endif
+                            {
+                              printf("Error at %d x %d (chan %d) %d %d [df: %g th: %g al: %g] (%d %d %d %d) (%d %d %d %d)\n",x,y,ch, p1[ch],p2[ch],d,thres,av,p1[0],p1[1],p1[2],p1[3],p2[0],p2[1],p2[2],p2[3]);
+                              ++nums;
+                              if ( nums > 16 ) goto ex;
+                              //if (d) exit(1);
+                              //goto ex;
+                            }
+                          }
+                        }
+                        break;
+
+                        case 3:
+                        {
+                          float * p1 = (float *)&ir1[y*op+x*c*sizeof(float)];
+                          float * p2 = (float *)&ir2[y*np+x*c*sizeof(float)];
+                          for( ch = 0 ; ch < c ; ch++ )
+                          {
+                            float pp1 = p1[ch], pp2 = p2[ch];
+                            float av = (a==-1)?1.0f:p1[a];
+                            float thres, d;
+
+                            // clamp
+                            if (pp1<=0.0f) pp1 = 0;
+                            if (pp2<=0.0f) pp2 = 0;
+                            if (av<=0.0f) av = 0;
+                            if (pp1>1.0f) pp1 = 1.0f;
+                            if (pp2>1.0f) pp2 = 1.0f;
+                            if (av>1.0f) av = 1.0f;
+
+                            // compare in premult space
+                            #ifndef COMPARE_SAME                        
+                            if ( ( ( layouts >=4 ) && ( layouts <= 7 ) ) || ( ( layouts >= 16 ) && ( layouts <= 19 ) ) )
+                            {
+                              pp1 *= av;
+                              pp2 *= av;
+                            }
+                            #endif
+
+                            d = pp1 - pp2;
+                            if ( d < 0 ) d = -d;
+
+                            thres=(p1[ch]*0.002f)+0.0002f;
+                            if ( thres < 0 ) thres = -thres;
+
+                            #ifdef COMPARE_SAME                        
+                            if ( d != 0.0f ) 
+                            #else 
+                            if ( d > thres )
+                            #endif
+                            {
+                              printf("Error at %d x %d (chan %d) %g %g [df: %g th: %g al: %g] (%g %g %g %g) (%g %g %g %g)\n",x,y,ch, p1[ch],p2[ch],d,thres,av,p1[0],p1[1],p1[2],p1[3],p2[0],p2[1],p2[2],p2[3]);
+                              ++nums;
+                              if ( nums > 16 ) goto ex;
+                              //if (d) exit(1);
+                              //goto ex;
+                            }
+                          }
+                        }
+                        break;
+
+                        case 4:
+                        {
+                          #ifdef COMPARE_SAME                        
+                          stbir__FP16 * p1 = (stbir__FP16 *)&ir1[y*op+x*c*sizeof(stbir__FP16)];
+                          #else
+                          float * p1 = (float *)&ir1[y*op+x*c*sizeof(float)];
+                          #endif
+                          stbir__FP16 * p2 = (stbir__FP16 *)&ir2[y*np+x*c*sizeof(stbir__FP16)];
+                          for( ch = 0 ; ch < c ; ch++ )
+                          {
+                            #ifdef COMPARE_SAME                        
+                            float pp1  = stbir__half_to_float(p1[ch]);
+                            float av = (a==-1)?1.0f:stbir__half_to_float(p1[a]);
+                            #else
+                            float pp1 = stbir__half_to_float(stbir__float_to_half(p1[ch]));
+                            float av = (a==-1)?1.0f:stbir__half_to_float(stbir__float_to_half(p1[a]));
+                            #endif
+                            float pp2 = stbir__half_to_float(p2[ch]);
+                            float d, thres;
+
+                            // clamp
+                            if (pp1<=0.0f) pp1 = 0;
+                            if (pp2<=0.0f) pp2 = 0;
+                            if (av<=0.0f) av = 0;
+                            if (pp1>1.0f) pp1 = 1.0f;
+                            if (pp2>1.0f) pp2 = 1.0f;
+                            if (av>1.0f) av = 1.0f;
+
+                            thres=(pp1*0.002f)+0.0002f;
+
+                            // compare in premult space
+                            #ifndef COMPARE_SAME                        
+                            if ( ( ( layouts >=4 ) && ( layouts <= 7 ) ) || ( ( layouts >= 16 ) && ( layouts <= 19 ) ) )
+                            {
+                              pp1 *= av;
+                              pp2 *= av;
+                            }
+                            #endif
+
+                            d = pp1 - pp2;
+                            if ( d < 0 ) d = -d;
+
+
+                            #ifdef COMPARE_SAME                        
+                            if ( d != 0.0f ) 
+                            #else 
+                            if ( d > thres )
+                            #endif
+                            {
+                              printf("Error at %d x %d (chan %d) %g %g [df: %g th: %g al: %g] (%g %g %g %g) (%g %g %g %g)\n",x,y,ch, 
+                              #ifdef COMPARE_SAME                        
+                              stbir__half_to_float(p1[ch]), 
+                              #else
+                              p1[ch],
+                              #endif
+                              stbir__half_to_float(p2[ch]), 
+                              d,thres,av,
+                              #ifdef COMPARE_SAME                        
+                              stbir__half_to_float(p1[0]),stbir__half_to_float(p1[1]),stbir__half_to_float(p1[2]),stbir__half_to_float(p1[3]),
+                              #else
+                              p1[0],p1[1],p1[2],p1[3],
+                              #endif
+                              stbir__half_to_float(p2[0]),stbir__half_to_float(p2[1]),stbir__half_to_float(p2[2]),stbir__half_to_float(p2[3]) );
+                              ++nums;
+                              if ( nums > 16 ) goto ex;
+                              //if (d) exit(1);
+                              //goto ex;
+                            }
+                          }
+                        }
+                        break;
+                      }
+                    }
+
+                    for( x = (w*c)*tsizes[oldtypes]; x < op; x++ )
+                    {
+                      if ( ir1[y*op+x] != 79 )
+                      {
+                        printf("Margin error at %d x %d %d (should be 79) OLD!\n",x,y,(unsigned char)ir1[y*op+x]);
+                        goto ex;
+                      }
+                    }
+
+                    for( x = (w*c)*tsizes[types]; x < np; x++ )
+                    {
+                      if ( ir2[y*np+x] != 79 )
+                      {
+                        printf("Margin error at %d x %d %d (should be 79) NEW\n",x,y,(unsigned char)ir2[y*np+x]);
+                        goto ex;
+                      }
+                    }
+                  }
+                
+                  ex:
+                  ENTER( "OUTPUT IMAGES" );
+                  printf("  tot pix: %d, errs: %d\n", w*h*c,nums );
+
+                  if (nums)
+                  {
+                    stbi_write_png("old.png", w, h, c, ir1, op);
+                    stbi_write_png("new.png", w, h, c, ir2, np);
+                    exit(1);
+                  }
+
+                  LEAVE(); // output images
+                }
+                LEAVE(); //test compare
+              #endif
+
+
+
+              }
+              LEAVE(); // test filter
+            }
+            LEAVE(); // test edge
+          }
+          LEAVE(); // test width
+        }
+        LEAVE(); // test height
+      }
+      LEAVE(); // test type
+    }
+    LEAVE();  // test layout
+  }
+
+  CloseTM();
+  return 0;
+}

external/stb/stb/stb_image_resize_test/vf_train.c

@@ -0,0 +1,999 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#define stop() __debugbreak()
+#include <windows.h>
+#define int64 __int64
+
+#pragma warning(disable:4127)
+
+#define STBIR__WEIGHT_TABLES  
+#define STBIR_PROFILE
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include "stb_image_resize2.h"  
+
+static int * file_read( char const * filename )
+{
+  size_t s;
+  int * m;
+  FILE * f = fopen( filename, "rb" );
+  if ( f == 0 ) return 0;
+  
+  fseek( f, 0, SEEK_END);
+  s = ftell( f );
+  fseek( f, 0, SEEK_SET);
+  m = malloc( s + 4 );
+  m[0] = (int)s;
+  fread( m+1, 1, s, f);
+  fclose(f);
+
+  return( m );
+}
+
+typedef struct fileinfo
+{
+  int * timings;
+  int timing_count;
+  int dimensionx, dimensiony;
+  int numtypes;
+  int * types;
+  int * effective;
+  int cpu;
+  int simd;
+  int numinputrects;
+  int * inputrects;
+  int outputscalex, outputscaley;
+  int milliseconds;
+  int64 cycles;
+  double scale_time;
+  int bitmapx, bitmapy;
+  char const * filename;
+} fileinfo;
+
+int numfileinfo;
+fileinfo fi[256];
+unsigned char * bitmap;
+int bitmapw, bitmaph, bitmapp;
+
+static int use_timing_file( char const * filename, int index )
+{
+  int * base = file_read( filename );
+  int * file = base;
+
+  if ( base == 0 ) return 0;
+
+  ++file; // skip file image size;
+  if ( *file++ != 'VFT1' ) return 0;
+  fi[index].cpu = *file++;
+  fi[index].simd = *file++;
+  fi[index].dimensionx = *file++;
+  fi[index].dimensiony = *file++;
+  fi[index].numtypes = *file++;
+  fi[index].types = file; file += fi[index].numtypes;
+  fi[index].effective = file; file += fi[index].numtypes;
+  fi[index].numinputrects = *file++;
+  fi[index].inputrects = file; file += fi[index].numinputrects * 2;
+  fi[index].outputscalex = *file++;
+  fi[index].outputscaley = *file++;
+  fi[index].milliseconds = *file++;
+  fi[index].cycles = ((int64*)file)[0]; file += 2;
+  fi[index].filename = filename;
+
+  fi[index].timings = file;
+  fi[index].timing_count = (int) ( ( base[0] - ( ((char*)file - (char*)base - sizeof(int) ) ) ) / (sizeof(int)*2) );
+
+  fi[index].scale_time = (double)fi[index].milliseconds / (double)fi[index].cycles;
+
+  return 1;
+}
+
+static int vert_first( float weights_table[STBIR_RESIZE_CLASSIFICATIONS][4], int ox, int oy, int ix, int iy, int filter, STBIR__V_FIRST_INFO * v_info )
+{
+  float h_scale=(float)ox/(float)(ix);
+  float v_scale=(float)oy/(float)(iy);
+  stbir__support_callback * support = stbir__builtin_supports[filter];
+  int vertical_filter_width = stbir__get_filter_pixel_width(support,v_scale,0);
+  int vertical_gather = ( v_scale >= ( 1.0f - stbir__small_float ) ) || ( vertical_filter_width <= STBIR_FORCE_GATHER_FILTER_SCANLINES_AMOUNT );
+
+  return stbir__should_do_vertical_first( weights_table, stbir__get_filter_pixel_width(support,h_scale,0), h_scale, ox, vertical_filter_width, v_scale, oy, vertical_gather, v_info );
+} 
+
+#define STB_IMAGE_WRITE_IMPLEMENTATION
+#include "stb_image_write.h"
+
+static void alloc_bitmap()
+{
+  int findex;
+  int x = 0, y = 0;
+  int w = 0, h = 0;
+
+  for( findex = 0 ; findex < numfileinfo ; findex++ )
+  {
+    int nx, ny;
+    int thisw, thish;
+
+    thisw = ( fi[findex].dimensionx * fi[findex].numtypes ) + ( fi[findex].numtypes - 1 );
+    thish = ( fi[findex].dimensiony * fi[findex].numinputrects ) + ( fi[findex].numinputrects - 1 );
+
+    for(;;)
+    {
+      nx = x + ((x)?4:0) + thisw;
+      ny = y + ((y)?4:0) + thish;
+      if ( ( nx <= 3600 ) || ( x == 0 ) )
+      { 
+        fi[findex].bitmapx = x + ((x)?4:0);
+        fi[findex].bitmapy = y + ((y)?4:0);
+        x = nx;
+        if ( x > w ) w = x;
+        if ( ny > h ) h = ny;
+        break;
+      }
+      else
+      {
+        x = 0;
+        y = h;
+      }
+    }
+  }
+
+  w = (w+3) & ~3;
+  bitmapw = w;
+  bitmaph = h;
+  bitmapp = w * 3; // RGB
+  bitmap = malloc( bitmapp * bitmaph );
+
+  memset( bitmap, 0, bitmapp * bitmaph );
+}
+
+static void build_bitmap( float weights[STBIR_RESIZE_CLASSIFICATIONS][4], int do_channel_count_index, int findex )
+{
+  static int colors[STBIR_RESIZE_CLASSIFICATIONS];
+  STBIR__V_FIRST_INFO v_info = {0};
+
+  int * ts;
+  int ir;
+  unsigned char * bitm = bitmap + ( fi[findex].bitmapx*3 ) + ( fi[findex].bitmapy*bitmapp) ;
+
+  for( ir = 0; ir < STBIR_RESIZE_CLASSIFICATIONS ; ir++ ) colors[ ir ] = 127*ir/STBIR_RESIZE_CLASSIFICATIONS+128;
+
+  ts = fi[findex].timings;
+
+  for( ir = 0 ; ir < fi[findex].numinputrects ; ir++ )
+  {
+    int ix, iy, chanind;
+    ix = fi[findex].inputrects[ir*2];
+    iy = fi[findex].inputrects[ir*2+1];
+
+    for( chanind = 0 ; chanind < fi[findex].numtypes ; chanind++ )
+    {
+      int ofs, h, hh;
+
+      // just do the type that we're on
+      if ( chanind != do_channel_count_index )
+      {
+        ts += 2 * fi[findex].dimensionx * fi[findex].dimensiony;
+        continue;
+      }
+
+      // bitmap offset
+      ofs=chanind*(fi[findex].dimensionx+1)*3+ir*(fi[findex].dimensiony+1)*bitmapp;
+
+      h = 1;
+      for( hh = 0 ; hh < fi[findex].dimensiony; hh++ )
+      {
+        int ww, w = 1;
+        for( ww = 0 ; ww < fi[findex].dimensionx; ww++ )
+        {
+          int good, v_first, VF, HF;
+
+          VF = ts[0];
+          HF = ts[1];
+
+          v_first = vert_first( weights, w, h, ix, iy, STBIR_FILTER_MITCHELL, &v_info );
+
+          good = ( ((HF<=VF) && (!v_first)) || ((VF<=HF) && (v_first)));
+
+          if ( good )
+          {
+            bitm[ofs+2] = 0;
+            bitm[ofs+1] = (unsigned char)colors[v_info.v_resize_classification];
+          }
+          else
+          {
+            double r;
+
+            if ( HF < VF )
+              r = (double)(VF-HF)/(double)HF;
+            else
+              r = (double)(HF-VF)/(double)VF;
+            
+            if ( r > 0.4f) r = 0.4;
+            r *= 1.0f/0.4f;   
+
+            bitm[ofs+2] = (char)(255.0f*r);
+            bitm[ofs+1] = (char)(((float)colors[v_info.v_resize_classification])*(1.0f-r));
+          }
+          bitm[ofs] = 0;
+
+          ofs += 3;
+          ts += 2;
+          w += fi[findex].outputscalex;
+        }
+        ofs += bitmapp - fi[findex].dimensionx*3;
+        h += fi[findex].outputscaley;
+      }
+    }
+  }
+}
+
+static void build_comp_bitmap( float weights[STBIR_RESIZE_CLASSIFICATIONS][4], int do_channel_count_index )
+{
+  int * ts0;
+  int * ts1;
+  int ir;
+  unsigned char * bitm = bitmap + ( fi[0].bitmapx*3 ) + ( fi[0].bitmapy*bitmapp) ;
+
+  ts0 = fi[0].timings;
+  ts1 = fi[1].timings;
+
+  for( ir = 0 ; ir < fi[0].numinputrects ; ir++ )
+  {
+    int ix, iy, chanind;
+    ix = fi[0].inputrects[ir*2];
+    iy = fi[0].inputrects[ir*2+1];
+
+    for( chanind = 0 ; chanind < fi[0].numtypes ; chanind++ )
+    {
+      int ofs, h, hh;
+
+      // just do the type that we're on
+      if ( chanind != do_channel_count_index )
+      {
+        ts0 += 2 * fi[0].dimensionx * fi[0].dimensiony;
+        ts1 += 2 * fi[0].dimensionx * fi[0].dimensiony;
+        continue;
+      }
+
+      // bitmap offset
+      ofs=chanind*(fi[0].dimensionx+1)*3+ir*(fi[0].dimensiony+1)*bitmapp;
+
+      h = 1;
+      for( hh = 0 ; hh < fi[0].dimensiony; hh++ )
+      {
+        int ww, w = 1;
+        for( ww = 0 ; ww < fi[0].dimensionx; ww++ )
+        {
+          int v_first, time0, time1;
+
+          v_first = vert_first( weights, w, h, ix, iy, STBIR_FILTER_MITCHELL, 0 );
+
+          time0 = ( v_first ) ? ts0[0] : ts0[1];
+          time1 = ( v_first ) ? ts1[0] : ts1[1];
+
+          if ( time0 < time1 )
+          {
+            double r = (double)(time1-time0)/(double)time0;
+            if ( r > 0.4f) r = 0.4;
+            r *= 1.0f/0.4f;   
+            bitm[ofs+2] = 0;
+            bitm[ofs+1] = (char)(255.0f*r);
+            bitm[ofs] = (char)(64.0f*(1.0f-r));
+          }
+          else
+          {
+            double r = (double)(time0-time1)/(double)time1;
+            if ( r > 0.4f) r = 0.4;
+            r *= 1.0f/0.4f;   
+            bitm[ofs+2] = (char)(255.0f*r);
+            bitm[ofs+1] = 0;
+            bitm[ofs] = (char)(64.0f*(1.0f-r));
+          }
+          ofs += 3;
+          ts0 += 2;
+          ts1 += 2;
+          w += fi[0].outputscalex;
+        }
+        ofs += bitmapp - fi[0].dimensionx*3;
+        h += fi[0].outputscaley;
+      }
+    }
+  }
+}
+
+static void write_bitmap()
+{
+  stbi_write_png( "results.png", bitmapp / 3, bitmaph, 3|STB_IMAGE_BGR, bitmap, bitmapp );
+}
+
+
+static void calc_errors( float weights_table[STBIR_RESIZE_CLASSIFICATIONS][4], int * curtot, double * curerr, int do_channel_count_index )
+{
+  int th, findex;
+  STBIR__V_FIRST_INFO v_info = {0};
+
+  for(th=0;th<STBIR_RESIZE_CLASSIFICATIONS;th++)
+  {
+    curerr[th]=0;
+    curtot[th]=0;
+  }
+
+  for( findex = 0 ; findex < numfileinfo ; findex++ )
+  {
+    int * ts;
+    int ir;
+    ts = fi[findex].timings;
+
+    for( ir = 0 ; ir < fi[findex].numinputrects ; ir++ )
+    {
+      int ix, iy, chanind;
+      ix = fi[findex].inputrects[ir*2];
+      iy = fi[findex].inputrects[ir*2+1];
+
+      for( chanind = 0 ; chanind < fi[findex].numtypes ; chanind++ )
+      {
+        int h, hh;
+
+        // just do the type that we're on
+        if ( chanind != do_channel_count_index )
+        {
+          ts += 2 * fi[findex].dimensionx * fi[findex].dimensiony;
+          continue;
+        }
+
+        h = 1;
+        for( hh = 0 ; hh < fi[findex].dimensiony; hh++ )
+        {
+          int ww, w = 1;
+          for( ww = 0 ; ww < fi[findex].dimensionx; ww++ )
+          {
+            int good, v_first, VF, HF;
+
+            VF = ts[0];
+            HF = ts[1];
+
+            v_first = vert_first( weights_table, w, h, ix, iy, STBIR_FILTER_MITCHELL, &v_info );
+
+            good = ( ((HF<=VF) && (!v_first)) || ((VF<=HF) && (v_first)));
+
+            if ( !good )
+            {
+              double diff;
+              if ( VF < HF )
+                diff = ((double)HF-(double)VF) * fi[findex].scale_time;
+              else
+                diff = ((double)VF-(double)HF) * fi[findex].scale_time;
+
+              curtot[v_info.v_resize_classification] += 1;
+              curerr[v_info.v_resize_classification] += diff;
+            }
+
+            ts += 2;
+            w += fi[findex].outputscalex;
+          }
+          h += fi[findex].outputscaley;
+        }
+      }
+    }
+  }
+}
+
+#define TRIESPERWEIGHT 32
+#define MAXRANGE ((TRIESPERWEIGHT+1) * (TRIESPERWEIGHT+1) * (TRIESPERWEIGHT+1) * (TRIESPERWEIGHT+1) - 1)
+
+static void expand_to_floats( float * weights, int range )
+{
+  weights[0] = (float)( range % (TRIESPERWEIGHT+1) ) / (float)TRIESPERWEIGHT;
+  weights[1] = (float)( range/(TRIESPERWEIGHT+1) % (TRIESPERWEIGHT+1) ) / (float)TRIESPERWEIGHT;
+  weights[2] = (float)( range/(TRIESPERWEIGHT+1)/(TRIESPERWEIGHT+1) % (TRIESPERWEIGHT+1) ) / (float)TRIESPERWEIGHT;
+  weights[3] = (float)( range/(TRIESPERWEIGHT+1)/(TRIESPERWEIGHT+1)/(TRIESPERWEIGHT+1) % (TRIESPERWEIGHT+1) ) / (float)TRIESPERWEIGHT;
+}
+
+static char const * expand_to_string( int range )
+{
+  static char str[128];
+  int w0,w1,w2,w3;
+  w0 = range % (TRIESPERWEIGHT+1);
+  w1 = range/(TRIESPERWEIGHT+1) % (TRIESPERWEIGHT+1);
+  w2 = range/(TRIESPERWEIGHT+1)/(TRIESPERWEIGHT+1) % (TRIESPERWEIGHT+1);
+  w3 = range/(TRIESPERWEIGHT+1)/(TRIESPERWEIGHT+1)/(TRIESPERWEIGHT+1) % (TRIESPERWEIGHT+1);
+  sprintf( str, "[ %2d/%d %2d/%d %2d/%d %2d/%d ]",w0,TRIESPERWEIGHT,w1,TRIESPERWEIGHT,w2,TRIESPERWEIGHT,w3,TRIESPERWEIGHT );
+  return str;
+}
+
+static void print_weights( float weights[STBIR_RESIZE_CLASSIFICATIONS][4], int channel_count_index, int * tots, double * errs )
+{
+  int th;
+  printf("ChInd: %d  Weights:\n",channel_count_index);
+  for(th=0;th<STBIR_RESIZE_CLASSIFICATIONS;th++)
+  {
+    float * w = weights[th];
+    printf("  %d: [%1.5f %1.5f %1.5f %1.5f] (%d %.4f)\n",th, w[0], w[1], w[2], w[3], tots[th], errs[th] );
+  }
+  printf("\n");
+}
+
+static int windowranges[ 16 ];
+static int windowstatus = 0;
+static DWORD trainstart = 0;
+
+static void opt_channel( float best_output_weights[STBIR_RESIZE_CLASSIFICATIONS][4], int channel_count_index )
+{
+  int newbest = 0;
+  float weights[STBIR_RESIZE_CLASSIFICATIONS][4] = {0};
+  double besterr[STBIR_RESIZE_CLASSIFICATIONS];
+  int besttot[STBIR_RESIZE_CLASSIFICATIONS];
+  int best[STBIR_RESIZE_CLASSIFICATIONS]={0};
+
+  double curerr[STBIR_RESIZE_CLASSIFICATIONS];
+  int curtot[STBIR_RESIZE_CLASSIFICATIONS];
+  int th, range;
+  DWORD lasttick = 0;
+
+  for(th=0;th<STBIR_RESIZE_CLASSIFICATIONS;th++) 
+  {
+    besterr[th]=1000000000000.0;
+    besttot[th]=0x7fffffff;
+  }
+
+  newbest = 0;
+
+  // try the whole range  
+  range = MAXRANGE;
+  do
+  {
+    for(th=0;th<STBIR_RESIZE_CLASSIFICATIONS;th++)
+      expand_to_floats( weights[th], range );
+
+    calc_errors( weights, curtot, curerr, channel_count_index );
+
+    for(th=0;th<STBIR_RESIZE_CLASSIFICATIONS;th++)
+    {
+      if ( curerr[th] < besterr[th] )
+      {
+        besterr[th] = curerr[th];
+        besttot[th] = curtot[th];
+        best[th] = range;
+        expand_to_floats( best_output_weights[th], best[th] );
+        newbest = 1;
+      }
+    }
+
+    {
+      DWORD t = GetTickCount();
+      if ( range == 0 )
+        goto do_bitmap;
+
+      if ( newbest )
+      {
+        if ( ( GetTickCount() - lasttick ) > 200 )
+        {
+          int findex;
+        
+         do_bitmap:
+          lasttick = t;
+          newbest = 0;
+
+          for( findex = 0 ; findex < numfileinfo ; findex++ )
+            build_bitmap( best_output_weights, channel_count_index, findex );
+
+          lasttick = GetTickCount();
+        }
+      }
+    }
+   
+    windowranges[ channel_count_index ] = range;
+
+    // advance all the weights and loop
+    --range;
+  } while( ( range >= 0 ) && ( !windowstatus ) );
+
+  // if we hit here, then we tried all weights for this opt, so save them 
+}
+
+static void print_struct( float weight[5][STBIR_RESIZE_CLASSIFICATIONS][4], char const * name )
+{
+  printf("\n\nstatic float %s[5][STBIR_RESIZE_CLASSIFICATIONS][4]=\n{", name );
+  {
+    int i;
+    for(i=0;i<5;i++) 
+    { 
+      int th;
+      for(th=0;th<STBIR_RESIZE_CLASSIFICATIONS;th++)
+      {
+        int j;
+        printf("\n  "); 
+        for(j=0;j<4;j++) 
+          printf("%1.5ff, ", weight[i][th][j] ); 
+      }
+      printf("\n");
+    }
+    printf("\n};\n");
+  }
+}
+
+static float retrain_weights[5][STBIR_RESIZE_CLASSIFICATIONS][4];
+
+static DWORD __stdcall retrain_shim( LPVOID p )
+{
+  int chanind = (int) (size_t)p;
+  opt_channel( retrain_weights[chanind], chanind );
+  return 0;
+}
+
+static char const * gettime( int ms )
+{
+  static char time[32];
+  if (ms > 60000)
+    sprintf( time, "%dm %ds",ms/60000, (ms/1000)%60 );
+  else  
+    sprintf( time, "%ds",ms/1000 );
+  return time;
+}
+
+static BITMAPINFOHEADER bmiHeader;
+static DWORD extrawindoww, extrawindowh;
+static HINSTANCE instance;
+static int curzoom = 1;
+
+static LRESULT WINAPI WindowProc( HWND   window,
+                                  UINT   message,
+                                  WPARAM wparam,
+                                  LPARAM lparam )
+{
+  switch( message )
+  {
+    case WM_CHAR:
+      if ( wparam != 27 )
+        break;
+      // falls through
+
+    case WM_CLOSE:
+    {
+      int i;
+      int max = 0;
+      
+      for( i = 0 ; i < fi[0].numtypes ; i++ )
+        if( windowranges[i] > max ) max = windowranges[i];
+   
+      if ( ( max == 0 ) || ( MessageBox( window, "Cancel before training is finished?", "Vertical First Training", MB_OKCANCEL|MB_ICONSTOP ) == IDOK ) )
+      {
+        for( i = 0 ; i < fi[0].numtypes ; i++ )
+          if( windowranges[i] > max ) max = windowranges[i];
+        if ( max )
+          windowstatus = 1;
+        DestroyWindow( window );
+      }
+    }
+    return 0;
+       
+    case WM_PAINT:
+    {
+      PAINTSTRUCT ps;
+      HDC dc;
+
+      dc = BeginPaint( window, &ps );
+      StretchDIBits( dc, 
+        0, 0, bitmapw*curzoom, bitmaph*curzoom,
+        0, 0, bitmapw, bitmaph,
+        bitmap, (BITMAPINFO*)&bmiHeader, DIB_RGB_COLORS, SRCCOPY );
+
+      PatBlt( dc, bitmapw*curzoom, 0, 4096, 4096, WHITENESS );
+      PatBlt( dc, 0, bitmaph*curzoom, 4096, 4096, WHITENESS );
+
+      SetTextColor( dc, RGB(0,0,0)  );
+      SetBkColor( dc, RGB(255,255,255) );
+      SetBkMode( dc, OPAQUE );
+
+      {
+        int i, l = 0, max = 0;
+        char buf[1024];
+        RECT rc;
+        POINT p;
+
+        for( i = 0 ; i < fi[0].numtypes ; i++ )
+        {
+          l += sprintf( buf + l, "channels: %d %s\n", fi[0].effective[i], windowranges[i] ? expand_to_string( windowranges[i] ) : "Done." );
+          if ( windowranges[i] > max ) max = windowranges[i];
+        }
+
+        rc.left = 32; rc.top = bitmaph*curzoom+10;
+        rc.right = 512; rc.bottom = rc.top + 512;
+        DrawText( dc, buf, -1, &rc, DT_TOP );
+
+        l = 0;
+        if ( max == 0 )
+        {
+          static DWORD traindone = 0;
+          if ( traindone == 0 ) traindone = GetTickCount();
+          l = sprintf( buf, "Finished in %s.", gettime( traindone - trainstart ) );
+        }
+        else if ( max != MAXRANGE )
+          l = sprintf( buf, "Done in %s...", gettime( (int) ( ( ( (int64)max * ( (int64)GetTickCount() - (int64)trainstart ) ) ) / (int64) ( MAXRANGE - max ) ) ) );
+
+        GetCursorPos( &p );
+        ScreenToClient( window, &p );
+
+        if ( ( p.x >= 0 ) && ( p.y >= 0 ) && ( p.x < (bitmapw*curzoom) ) && ( p.y < (bitmaph*curzoom) ) )
+        {
+          int findex;
+          int x, y, w, h, sx, sy, ix, iy, ox, oy;
+          int ir, chanind;
+          int * ts;
+          char badstr[64];
+          STBIR__V_FIRST_INFO v_info={0};
+
+          p.x /= curzoom;
+          p.y /= curzoom;
+
+          for( findex = 0 ; findex < numfileinfo ; findex++ )
+          {
+            x = fi[findex].bitmapx;
+            y = fi[findex].bitmapy;
+            w = x + ( fi[findex].dimensionx + 1 ) * fi[findex].numtypes;
+            h = y + ( fi[findex].dimensiony + 1 ) * fi[findex].numinputrects;
+
+            if ( ( p.x >= x ) && ( p.y >= y ) && ( p.x < w ) && ( p.y < h ) )
+              goto found;
+          }
+          goto nope;
+         
+         found:
+            
+          ir = ( p.y - y ) / ( fi[findex].dimensiony + 1 );
+          sy = ( p.y - y ) % ( fi[findex].dimensiony + 1 );
+          if ( sy >= fi[findex].dimensiony ) goto nope;
+
+          chanind = ( p.x - x ) / ( fi[findex].dimensionx + 1 );
+          sx = ( p.x - x ) % ( fi[findex].dimensionx + 1 );
+          if ( sx >= fi[findex].dimensionx ) goto nope;
+
+          ix = fi[findex].inputrects[ir*2];
+          iy = fi[findex].inputrects[ir*2+1];
+
+          ts = fi[findex].timings + ( ( fi[findex].dimensionx * fi[findex].dimensiony * fi[findex].numtypes * ir ) + ( fi[findex].dimensionx * fi[findex].dimensiony * chanind ) + ( fi[findex].dimensionx * sy ) + sx ) * 2;
+
+          ox = 1+fi[findex].outputscalex*sx;
+          oy = 1+fi[findex].outputscaley*sy;
+
+          if ( windowstatus != 2 )
+          {
+            int VF, HF, v_first, good;
+            VF = ts[0];
+            HF = ts[1];
+
+            v_first = vert_first( retrain_weights[chanind], ox, oy, ix, iy, STBIR_FILTER_MITCHELL, &v_info );
+
+            good = ( ((HF<=VF) && (!v_first)) || ((VF<=HF) && (v_first)));
+
+            if ( good )
+              badstr[0] = 0;
+            else
+            {
+              double r;
+
+              if ( HF < VF )
+                r = (double)(VF-HF)/(double)HF;
+              else
+                r = (double)(HF-VF)/(double)VF;
+              sprintf( badstr, " %.1f%% off", r*100 );
+            }
+            sprintf( buf + l, "\n\n%s\nCh: %d Resize: %dx%d to %dx%d\nV: %d H: %d  Order: %c (%s%s)\nClass: %d Scale: %.2f %s", fi[findex].filename,fi[findex].effective[chanind], ix,iy,ox,oy, VF, HF, v_first?'V':'H', good?"Good":"Wrong", badstr, v_info.v_resize_classification, (double)oy/(double)iy, v_info.is_gather ? "Gather" : "Scatter" );
+          }
+          else
+          {
+            int v_first, time0, time1;
+            float (* weights)[4] = stbir__compute_weights[chanind];
+            int * ts1;
+            char b0[32], b1[32];
+
+            ts1 = fi[1].timings + ( ts - fi[0].timings );
+
+            v_first = vert_first( weights, ox, oy, ix, iy, STBIR_FILTER_MITCHELL, &v_info );
+
+            time0 = ( v_first ) ? ts[0] : ts[1];
+            time1 = ( v_first ) ? ts1[0] : ts1[1];
+            
+            b0[0] = b1[0] = 0;
+            if ( time0 < time1 )
+              sprintf( b0," (%.f%% better)", ((double)time1-(double)time0)*100.0f/(double)time0);
+            else
+              sprintf( b1," (%.f%% better)", ((double)time0-(double)time1)*100.0f/(double)time1);
+
+            sprintf( buf + l, "\n\n0: %s\n1: %s\nCh: %d Resize: %dx%d to %dx%d\nClass: %d Scale: %.2f %s\nTime0: %d%s\nTime1: %d%s", fi[0].filename, fi[1].filename, fi[0].effective[chanind], ix,iy,ox,oy, v_info.v_resize_classification, (double)oy/(double)iy, v_info.is_gather ? "Gather" : "Scatter", time0, b0, time1, b1 );
+          }
+        }
+       nope:
+
+        rc.left = 32+320; rc.right = 512+320; 
+        SetTextColor( dc, RGB(0,0,128) );
+        DrawText( dc, buf, -1, &rc, DT_TOP );
+
+      }
+      EndPaint( window, &ps );
+      return 0;
+    }
+
+    case WM_TIMER:
+      InvalidateRect( window, 0, 0 );
+      return 0;
+
+    case WM_DESTROY:
+      PostQuitMessage( 0 );
+      return 0;
+  }
+   
+
+  return DefWindowProc( window, message, wparam, lparam );
+}
+
+static void SetHighDPI(void)
+{
+  typedef HRESULT WINAPI setdpitype(int v);
+  HMODULE h=LoadLibrary("Shcore.dll");
+  if (h)
+  {
+    setdpitype * sd = (setdpitype*)GetProcAddress(h,"SetProcessDpiAwareness");
+    if (sd )
+      sd(1);
+  }
+} 
+
+static void draw_window()
+{
+  WNDCLASS wc;
+  HWND w;
+  MSG msg;
+  
+  instance = GetModuleHandle(NULL);
+
+  wc.style = 0;
+  wc.lpfnWndProc = WindowProc;
+  wc.cbClsExtra = 0;
+  wc.cbWndExtra = 0;
+  wc.hInstance = instance;
+  wc.hIcon = 0;
+  wc.hCursor = LoadCursor(NULL, IDC_ARROW);
+  wc.hbrBackground = 0;
+  wc.lpszMenuName = 0;
+  wc.lpszClassName = "WHTrain";
+
+  if ( !RegisterClass( &wc ) )
+    exit(1);
+
+  SetHighDPI();
+
+  bmiHeader.biSize          =  sizeof(BITMAPINFOHEADER);
+  bmiHeader.biWidth         =  bitmapp/3;
+  bmiHeader.biHeight        =  -bitmaph;
+  bmiHeader.biPlanes        =  1;
+  bmiHeader.biBitCount      =  24;
+  bmiHeader.biCompression   =  BI_RGB;
+
+  w = CreateWindow( "WHTrain",
+                    "Vertical First Training",
+                    WS_CAPTION | WS_POPUP| WS_CLIPCHILDREN |
+                    WS_SYSMENU | WS_MINIMIZEBOX | WS_SIZEBOX,
+                    CW_USEDEFAULT,CW_USEDEFAULT,
+                    CW_USEDEFAULT,CW_USEDEFAULT,
+                    0, 0, instance, 0 );
+
+  {
+    RECT r, c;
+    GetWindowRect( w, &r );
+    GetClientRect( w, &c );
+    extrawindoww = ( r.right - r.left ) - ( c.right - c.left );
+    extrawindowh = ( r.bottom - r.top ) - ( c.bottom - c.top );
+    SetWindowPos( w, 0, 0, 0, bitmapw * curzoom + extrawindoww, bitmaph * curzoom + extrawindowh + 164, SWP_NOMOVE );
+  }
+
+  ShowWindow( w, SW_SHOWNORMAL );
+  SetTimer( w, 1, 250, 0 );
+
+  {
+    BOOL ret;
+    while( ( ret = GetMessage( &msg, w, 0, 0 ) ) != 0 )
+    { 
+      if ( ret == -1 )
+        break;
+      TranslateMessage( &msg ); 
+      DispatchMessage( &msg ); 
+    }
+  }
+}
+
+static void retrain()
+{
+  HANDLE threads[ 16 ];
+  int chanind;
+
+  trainstart = GetTickCount();
+  for( chanind = 0 ; chanind < fi[0].numtypes ; chanind++ )
+    threads[ chanind ] = CreateThread( 0, 2048*1024, retrain_shim, (LPVOID)(size_t)chanind, 0, 0 );
+
+  draw_window();
+
+  for( chanind = 0 ; chanind < fi[0].numtypes ; chanind++ )
+  {
+    WaitForSingleObject( threads[ chanind ], INFINITE );
+    CloseHandle( threads[ chanind ] );
+  }
+
+  write_bitmap();
+
+  print_struct( retrain_weights, "retained_weights" );
+  if ( windowstatus ) printf( "CANCELLED!\n" );
+}
+
+static void info()
+{
+  int findex;
+
+  // display info about each input file
+  for( findex = 0 ; findex < numfileinfo ; findex++ )
+  {
+    int i, h,m,s;
+    if ( findex ) printf( "\n" );
+    printf( "Timing file: %s\n", fi[findex].filename );
+    printf( "CPU type: %d  %s\n", fi[findex].cpu, fi[findex].simd?(fi[findex].simd==2?"SIMD8":"SIMD4"):"Scalar" );
+    h = fi[findex].milliseconds/3600000;
+    m = (fi[findex].milliseconds-h*3600000)/60000;
+    s = (fi[findex].milliseconds-h*3600000-m*60000)/1000;
+    printf( "Total time in test: %dh %dm %ds  Cycles/sec: %.f\n", h,m,s, 1000.0/fi[findex].scale_time );
+    printf( "Each tile of samples is %dx%d, and is scaled by %dx%d.\n", fi[findex].dimensionx,fi[findex].dimensiony, fi[findex].outputscalex,fi[findex].outputscaley );
+    printf( "So the x coords are: " );
+    for( i=0; i < fi[findex].dimensionx ; i++ ) printf( "%d ",1+i*fi[findex].outputscalex );
+    printf( "\n" );
+    printf( "And the y coords are: " );
+    for( i=0; i < fi[findex].dimensiony ; i++ ) printf( "%d ",1+i*fi[findex].outputscaley );
+    printf( "\n" );
+    printf( "There are %d channel counts and they are: ", fi[findex].numtypes );
+    for( i=0; i < fi[findex].numtypes ; i++ ) printf( "%d ",fi[findex].effective[i] );
+    printf( "\n" );
+    printf( "There are %d input rect sizes and they are: ", fi[findex].numinputrects );
+    for( i=0; i < fi[findex].numtypes ; i++ ) printf( "%dx%d ",fi[findex].inputrects[i*2],fi[findex].inputrects[i*2+1] );
+    printf( "\n" );
+  }
+}
+
+static void current( int do_win, int do_bitmap )
+{
+  int i, findex;
+
+  trainstart = GetTickCount();
+
+  // clear progress
+  memset( windowranges, 0, sizeof( windowranges ) );
+  // copy in appropriate weights
+  memcpy( retrain_weights, stbir__compute_weights, sizeof( retrain_weights ) );
+
+  // build and print current errors and build current bitmap
+  for( i = 0 ; i < fi[0].numtypes ; i++ )
+  {
+    double curerr[STBIR_RESIZE_CLASSIFICATIONS];
+    int curtot[STBIR_RESIZE_CLASSIFICATIONS];
+    float (* weights)[4] = retrain_weights[i];
+ 
+    calc_errors( weights, curtot, curerr, i );
+    if ( !do_bitmap )
+      print_weights( weights, i, curtot, curerr );
+
+    for( findex = 0 ; findex < numfileinfo ; findex++ )
+      build_bitmap( weights, i, findex );
+  }
+
+  if ( do_win )
+    draw_window();
+
+  if ( do_bitmap )
+    write_bitmap();
+}
+
+static void compare()
+{
+  int i;
+
+  trainstart = GetTickCount();
+  windowstatus = 2; // comp mode
+
+  // clear progress
+  memset( windowranges, 0, sizeof( windowranges ) );
+
+  if ( ( fi[0].numtypes != fi[1].numtypes ) || ( fi[0].numinputrects != fi[1].numinputrects ) ||
+       ( fi[0].dimensionx != fi[1].dimensionx ) || ( fi[0].dimensiony != fi[1].dimensiony ) || 
+       ( fi[0].outputscalex != fi[1].outputscalex ) || ( fi[0].outputscaley != fi[1].outputscaley ) )
+  {
+   err:
+    printf( "Timing files don't match.\n" );
+    exit(5);
+  }
+
+  for( i=0; i < fi[0].numtypes ; i++ )
+  {
+    if ( fi[0].effective[i]      != fi[1].effective[i] ) goto err;
+    if ( fi[0].inputrects[i*2]   != fi[1].inputrects[i*2] ) goto err;
+    if ( fi[0].inputrects[i*2+1] != fi[1].inputrects[i*2+1] ) goto err;
+  }
+    
+  alloc_bitmap( 1 );
+  
+  for( i = 0 ; i < fi[0].numtypes ; i++ )
+  {
+    float (* weights)[4] = stbir__compute_weights[i];
+    build_comp_bitmap( weights, i );
+  }
+
+  draw_window();
+}
+
+static void load_files( char ** args, int count )
+{
+  int i;
+
+  if ( count == 0 )
+  {
+    printf( "No timing files listed!" );
+    exit(3);
+  }
+
+  for ( i = 0 ; i < count ; i++ )
+  {
+    if ( !use_timing_file( args[i], i ) )
+    {
+      printf( "Bad timing file %s\n", args[i] );
+      exit(2);
+    }
+  }
+  numfileinfo = count;
+}  
+
+int main( int argc, char ** argv )
+{
+  int check;
+  if ( argc < 3 )
+  {
+   err:
+    printf( "vf_train retrain [timing_filenames....] - recalcs weights for all the files on the command line.\n");
+    printf( "vf_train info [timing_filenames....] - shows info about each timing file.\n");
+    printf( "vf_train check [timing_filenames...] - show results for the current weights for all files listed.\n");
+    printf( "vf_train compare <timing file1> <timing file2> - compare two timing files (must only be two files and same resolution).\n");
+    printf( "vf_train bitmap [timing_filenames...] - write out results.png, comparing against the current weights for all files listed.\n");
+    exit(1);
+  }
+  
+  check = ( strcmp( argv[1], "check" ) == 0 );
+  if ( ( check ) || ( strcmp( argv[1], "bitmap" ) == 0 ) )
+  {
+    load_files( argv + 2, argc - 2 );
+    alloc_bitmap( numfileinfo );
+    current( check, !check );
+  }
+  else if ( strcmp( argv[1], "info" ) == 0 ) 
+  {
+    load_files( argv + 2, argc - 2 );
+    info();
+  }
+  else if ( strcmp( argv[1], "compare" ) == 0 ) 
+  {
+    if ( argc != 4 )
+    {
+      printf( "You must specify two files to compare.\n" );
+      exit(4);
+    }
+
+    load_files( argv + 2, argc - 2 );
+    compare();
+  }
+  else if ( strcmp( argv[1], "retrain" ) == 0 ) 
+  {
+    load_files( argv + 2, argc - 2 );
+    alloc_bitmap( numfileinfo );
+    retrain();  
+  }
+  else
+  {
+    goto err;
+  }
+
+  return 0;
+}

external/stb/stb/tests/resample_test.cpp

@@ -64,7 +64,7 @@ void stbir_progress(float p)
 #define STBIR_PROGRESS_REPORT stbir_progress
 #define STB_IMAGE_RESIZE_IMPLEMENTATION
 #define STB_IMAGE_RESIZE_STATIC
-#include "stb_image_resize.h"
+#include "stb_image_resize2.h"
 
 #define STB_IMAGE_WRITE_IMPLEMENTATION
 #include "stb_image_write.h"
@@ -143,7 +143,7 @@ void resizer(int argc, char **argv)
 	out_h = h*3;
 	output_pixels = (unsigned char*) malloc(out_w*out_h*n);
 	//stbir_resize_uint8_srgb(input_pixels, w, h, 0, output_pixels, out_w, out_h, 0, n, -1,0);
-	stbir_resize_uint8(input_pixels, w, h, 0, output_pixels, out_w, out_h, 0, n);
+	stbir_resize_uint8_linear(input_pixels, w, h, 0, output_pixels, out_w, out_h, 0, (stbir_pixel_layout) n);
 	stbi_write_png("output.png", out_w, out_h, n, output_pixels, 0);
 	exit(0);
 }
@@ -171,9 +171,9 @@ void performance(int argc, char **argv)
 	output_pixels = (unsigned char*) malloc(out_w*out_h*n);
     for (i=0; i < count; ++i)
         if (srgb)
-	        stbir_resize_uint8_srgb(input_pixels, w, h, 0, output_pixels, out_w, out_h, 0, n,-1,0);
+	        stbir_resize_uint8_srgb(input_pixels, w, h, 0, output_pixels, out_w, out_h, 0, (stbir_pixel_layout) n);
         else
-	        stbir_resize(input_pixels, w, h, 0, output_pixels, out_w, out_h, 0, STBIR_TYPE_UINT8, n,-1, 0, STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT, STBIR_COLORSPACE_LINEAR, NULL);
+	        stbir_resize_uint8_linear(input_pixels, w, h, 0, output_pixels, out_w, out_h, 0, (stbir_pixel_layout) n);
 	exit(0);
 }
 
@@ -188,6 +188,7 @@ int main(int argc, char** argv)
 	return 0;
 }
 
+#if 0
 void resize_image(const char* filename, float width_percent, float height_percent, stbir_filter filter, stbir_edge edge, stbir_colorspace colorspace, const char* output_filename)
 {
 	int w, h, n;
@@ -1120,3 +1121,7 @@ void test_suite(int argc, char **argv)
 	resize_image("gamma_2.2.jpg", .5f, .5f, STBIR_FILTER_CATMULLROM, STBIR_EDGE_REFLECT, STBIR_COLORSPACE_SRGB, "test-output/gamma_2.2.jpg");
 	resize_image("gamma_dalai_lama_gray.jpg", .5f, .5f, STBIR_FILTER_CATMULLROM, STBIR_EDGE_REFLECT, STBIR_COLORSPACE_SRGB, "test-output/gamma_dalai_lama_gray.jpg");
 }
+#endif
+void test_suite(int argc, char **argv)
+{
+}

external/stb/stb/tests/resize.dsp

@@ -88,7 +88,7 @@ SOURCE=.\resample_test.cpp
 # End Source File
 # Begin Source File
 
-SOURCE=..\stb_image_resize.h
+SOURCE=..\stb_image_resize2.h
 # End Source File
 # End Target
 # End Project

external/stb/stb/tests/stb.dsp

@@ -130,10 +130,6 @@ SOURCE=..\stb_image.h
 # End Source File
 # Begin Source File
 
-SOURCE=..\stb_image_resize.h
-# End Source File
-# Begin Source File
-
 SOURCE=..\stb_image_write.h
 # End Source File
 # Begin Source File

external/stb/stb/tests/test_c_compilation.c

@@ -1,3 +1,6 @@
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include "stb_image_resize2.h"
+
 #define STB_SPRINTF_IMPLEMENTATION
 #include "stb_sprintf.h"
 
@@ -7,7 +10,6 @@
 #define STB_DIVIDE_IMPLEMENTATION
 #define STB_IMAGE_IMPLEMENTATION
 #define STB_HERRINGBONE_WANG_TILE_IMEPLEMENTATIOn
-#define STB_IMAGE_RESIZE_IMPLEMENTATION
 #define STB_RECT_PACK_IMPLEMENTATION
 #define STB_VOXEL_RENDER_IMPLEMENTATION
 #define STB_EASY_FONT_IMPLEMENTATION
@@ -20,7 +22,6 @@
 #include "stb_perlin.h"
 #include "stb_c_lexer.h"
 #include "stb_divide.h"
-#include "stb_image_resize.h"
 #include "stb_rect_pack.h"
 #include "stb_dxt.h"
 #include "stb_include.h"

external/stb/stb/tests/test_cpp_compilation.cpp

@@ -70,7 +70,7 @@ void my_free(void *) { }
 #include "stb_leakcheck.h"
 
 #define STB_IMAGE_RESIZE_IMPLEMENTATION
-#include "stb_image_resize.h"
+#include "stb_image_resize2.h"
 
 //#include "stretchy_buffer.h"  // deprecating
 

external/stb/stb/tools/README.header.md

@@ -3,16 +3,18 @@ stb
 
 single-file public domain (or MIT licensed) libraries for C/C++
 
+# This project discusses security-relevant bugs in public in Github Issues and Pull Requests, and it may take significant time for security fixes to be implemented or merged. If this poses an unreasonable risk to your project, do not use stb libraries.
+
 Noteworthy:
 
 * image loader: [stb_image.h](stb_image.h)
 * image writer: [stb_image_write.h](stb_image_write.h)
-* image resizer: [stb_image_resize.h](stb_image_resize.h)
+* image resizer: [stb_image_resize2.h](stb_image_resize2.h)
 * font text rasterizer: [stb_truetype.h](stb_truetype.h)
 * typesafe containers: [stb_ds.h](stb_ds.h)
 
-Most libraries by stb, except: stb_dxt by Fabian "ryg" Giesen, stb_image_resize
-by Jorge L. "VinoBS" Rodriguez, and stb_sprintf by Jeff Roberts.
+Most libraries by stb, except: stb_dxt by Fabian "ryg" Giesen, original stb_image_resize
+by Jorge L. "VinoBS" Rodriguez, and stb_image_resize2 and stb_sprintf by Jeff Roberts.
 
 <a name="stb_libs"></a>
 

external/stb/stb/tools/README.list

@@ -3,7 +3,7 @@ stb_hexwave.h               | audio            | audio waveform synthesizer
 stb_image.h                 | graphics         | image loading/decoding from file/memory: JPG, PNG, TGA, BMP, PSD, GIF, HDR, PIC
 stb_truetype.h              | graphics         | parse, decode, and rasterize characters from truetype fonts
 stb_image_write.h           | graphics         | image writing to disk: PNG, TGA, BMP
-stb_image_resize.h          | graphics         | resize images larger/smaller with good quality
+stb_image_resize2.h         | graphics         | resize images larger/smaller with good quality
 stb_rect_pack.h             | graphics         | simple 2D rectangle packer with decent quality
 stb_perlin.h                | graphics         | perlin's revised simplex noise w/ different seeds
 stb_ds.h                    | utility          | typesafe dynamic array and hash tables for C, will compile in C++