Show bad blocks

andrewevstyukhin · Jan 21, 2020 · e9be191 · e9be191
1 parent 5a4898a
commit e9be191
Show file tree

Hide file tree

Showing 5 changed files with 141 additions and 15 deletions.
diff --git a/src/Bc7Compress.cpp b/src/Bc7Compress.cpp
@@ -93,7 +93,7 @@ static void PackTexture(uint8_t* dst_bc7, uint8_t* src_bgra, uint8_t* mask_agrb,
 
 	auto finish = std::chrono::high_resolution_clock::now();
 
-	if (blockKernel != DecompressKernel)
+	if (blockKernel == CompressKernel)
 	{
 		int pixels = src_h * src_w;
 
@@ -104,10 +104,7 @@ static void PackTexture(uint8_t* dst_bc7, uint8_t* src_bgra, uint8_t* mask_agrb,
 		PRINTF("    Compressed %d blocks, elapsed %i ms, throughput %d.%03d Mpx/s", pixels >> 4, span, kpx_s / 1000, kpx_s % 1000);
 
 #if defined(OPTION_COUNTERS)
-		if (blockKernel == CompressKernel)
-		{
-			CompressStatistics();
-		}
+		CompressStatistics();
 #endif
 	}
 }
@@ -249,6 +246,7 @@ int Bc7MainWithArgs(const std::vector<std::string>& args)
 	const char* dst_name = nullptr;
 	const char* result_name = nullptr;
 	const char* partitions_name = nullptr;
+	const char* bad_name = nullptr;
 
 	for (int i = 0, n = (int)args.size(); i < n; i++)
 	{
@@ -327,6 +325,14 @@ int Bc7MainWithArgs(const std::vector<std::string>& args)
 				}
 				continue;
 			}
+			else if (strcmp(arg, "/bad") == 0)
+			{
+				if (++i < n)
+				{
+					bad_name = args[i].c_str();
+				}
+				continue;
+			}
 #if defined(WIN32)
 			else
 			{
@@ -504,6 +510,13 @@ int Bc7MainWithArgs(const std::vector<std::string>& args)
 
 		PackTexture(dst_bc7, dst_texture_bgra, mask_agrb, src_texture_stride, src_texture_w, src_texture_h, &DecompressKernel, 16, mse_alpha, mse_color, ssim);
 
+		if ((bad_name != nullptr) && bad_name[0])
+		{
+			ShowBadBlocks(src_texture_bgra, dst_texture_bgra, mask_agrb, src_texture_stride, src_texture_w, src_texture_h);
+
+			WriteImage(bad_name, mask_agrb, src_texture_w, src_texture_h, flip);
+		}
+
 		if ((partitions_name != nullptr) && partitions_name[0])
 		{
 			VisualizePartitionsGRB(dst_bc7, Size);
@@ -535,7 +548,7 @@ int __cdecl main(int argc, char* argv[])
 	if (argc < 2)
 	{
 		PRINTF("Usage: Bc7Compress [/fast | /normal | /slow | /draft] [/retina] [/nomask] [/noflip] src");
-		PRINTF("                   [dst.ktx] [/debug result.png] [/map partitions.png]");
+		PRINTF("                   [dst.ktx] [/debug result.png] [/map partitions.png] [/bad bad.png]");
 		return 1;
 	}
 

diff --git a/src/Metrics.cpp b/src/Metrics.cpp
@@ -129,8 +129,6 @@ BlockSSIM CompareBlocksSSIM(const Cell& cell1, const Cell& cell2) noexcept
 	return BlockSSIM(_mm_cvtsd_f64(mssim_ga), ssim * (1.0 / kColor));
 }
 
-#if defined(OPTION_COUNTERS)
-
 bool DetectGlitches(const Cell& input, const Cell& output) noexcept
 {
 	const __m128i msign = _mm_set1_epi8(-0x80);
@@ -169,5 +167,3 @@ bool DetectGlitches(const Cell& input, const Cell& output) noexcept
 
 	return _mm_movemask_epi8(me) != 0;
 }
-
-#endif
diff --git a/src/Metrics.h b/src/Metrics.h
@@ -49,8 +49,4 @@ BlockError CompareBlocks(const Cell& cell1, const Cell& cell2) noexcept;
 
 BlockSSIM CompareBlocksSSIM(const Cell& cell1, const Cell& cell2) noexcept;
 
-#if defined(OPTION_COUNTERS)
-
 bool DetectGlitches(const Cell& input, const Cell& output) noexcept;
-
-#endif
diff --git a/src/Worker.cpp b/src/Worker.cpp
@@ -1,6 +1,7 @@
 
 #include "pch.h"
 #include "Worker.h"
+#include "Metrics.h"
 
 #if defined(WIN32)
 #include <windows.h>
@@ -242,7 +243,7 @@ class Worker
 
 			if (_Running <= 0)
 				break;
-	}
+		}
 #endif
 
 		pErrorAlpha = _errorAlpha;
@@ -389,3 +390,121 @@ void ProcessTexture(uint8_t* dst, uint8_t* src_bgra, uint8_t* mask_agrb, int str
 
 	worker.Run(blockKernel, stride, pErrorAlpha, pErrorColor, pssim);
 }
+
+void ShowBadBlocks(const uint8_t* src_bgra, const uint8_t* dst_bgra, uint8_t* mask_agrb, int stride, int src_w, int src_h) noexcept
+{
+	Cell input;
+	Cell output;
+
+	for (int y = 0; y < src_h; y += 4)
+	{
+		const uint8_t* src = src_bgra + y * stride;
+		const uint8_t* dst = dst_bgra + y * stride;
+		uint8_t* mask = mask_agrb + y * stride;
+
+		for (int x = 0; x < src_w; x += 4)
+		{
+			{
+				const uint8_t* p = src;
+
+				input.ImageRows_U8[0] = ConvertBgraToAgrb(_mm_loadu_si128((const __m128i*)p));
+
+				p += stride;
+
+				input.ImageRows_U8[1] = ConvertBgraToAgrb(_mm_loadu_si128((const __m128i*)p));
+
+				p += stride;
+
+				input.ImageRows_U8[2] = ConvertBgraToAgrb(_mm_loadu_si128((const __m128i*)p));
+
+				p += stride;
+
+				input.ImageRows_U8[3] = ConvertBgraToAgrb(_mm_loadu_si128((const __m128i*)p));
+			}
+
+			{
+				const uint8_t* p = mask;
+
+				input.MaskRows_S8[0] = _mm_loadu_si128((const __m128i*)p);
+
+				p += stride;
+
+				input.MaskRows_S8[1] = _mm_loadu_si128((const __m128i*)p);
+
+				p += stride;
+
+				input.MaskRows_S8[2] = _mm_loadu_si128((const __m128i*)p);
+
+				p += stride;
+
+				input.MaskRows_S8[3] = _mm_loadu_si128((const __m128i*)p);
+			}
+
+			{
+				const uint8_t* p = dst;
+
+				output.ImageRows_U8[0] = ConvertBgraToAgrb(_mm_loadu_si128((const __m128i*)p));
+
+				p += stride;
+
+				output.ImageRows_U8[1] = ConvertBgraToAgrb(_mm_loadu_si128((const __m128i*)p));
+
+				p += stride;
+
+				output.ImageRows_U8[2] = ConvertBgraToAgrb(_mm_loadu_si128((const __m128i*)p));
+
+				p += stride;
+
+				output.ImageRows_U8[3] = ConvertBgraToAgrb(_mm_loadu_si128((const __m128i*)p));
+			}
+
+			bool bad = DetectGlitches(input, output);
+			if (bad)
+			{
+				const uint8_t* r = src;
+				uint8_t* p = mask;
+
+				_mm_storeu_si128((__m128i*)p, _mm_loadu_si128((const __m128i*)r));
+
+				r += stride;
+				p += stride;
+
+				_mm_storeu_si128((__m128i*)p, _mm_loadu_si128((const __m128i*)r));
+
+				r += stride;
+				p += stride;
+
+				_mm_storeu_si128((__m128i*)p, _mm_loadu_si128((const __m128i*)r));
+
+				r += stride;
+				p += stride;
+
+				_mm_storeu_si128((__m128i*)p, _mm_loadu_si128((const __m128i*)r));
+			}
+			else
+			{
+				__m128i mc = _mm_setzero_si128();
+
+				uint8_t* p = mask;
+
+				_mm_storeu_si128((__m128i*)p, mc);
+
+				p += stride;
+
+				_mm_storeu_si128((__m128i*)p, mc);
+
+				p += stride;
+
+				_mm_storeu_si128((__m128i*)p, mc);
+
+				p += stride;
+
+				_mm_storeu_si128((__m128i*)p, mc);
+			}
+
+			src += 16;
+			dst += 16;
+			mask += 16;
+		}
+	}
+}
diff --git a/src/Worker.h b/src/Worker.h
@@ -27,3 +27,5 @@ void DecompressKernel(const WorkerItem* begin, const WorkerItem* end, int stride
 void CompressKernel(const WorkerItem* begin, const WorkerItem* end, int stride, int64_t& pErrorAlpha, int64_t& pErrorColor, BlockSSIM& pssim) noexcept;
 
 void ProcessTexture(uint8_t* dst, uint8_t* src_bgra, uint8_t* mask_agrb, int stride, int src_w, int src_h, PBlockKernel blockKernel, size_t block_size, int64_t& pErrorAlpha, int64_t& pErrorColor, BlockSSIM& pssim);
+
+void ShowBadBlocks(const uint8_t* src_bgra, const uint8_t* dst_bgra, uint8_t* mask_agrb, int stride, int src_w, int src_h) noexcept;
Original file line number	Diff line number	Diff line change
Expand Up		@@ -27,3 +27,5 @@ void DecompressKernel(const WorkerItem* begin, const WorkerItem* end, int stride
		void CompressKernel(const WorkerItem* begin, const WorkerItem* end, int stride, int64_t& pErrorAlpha, int64_t& pErrorColor, BlockSSIM& pssim) noexcept;

		void ProcessTexture(uint8_t* dst, uint8_t* src_bgra, uint8_t* mask_agrb, int stride, int src_w, int src_h, PBlockKernel blockKernel, size_t block_size, int64_t& pErrorAlpha, int64_t& pErrorColor, BlockSSIM& pssim);

		void ShowBadBlocks(const uint8_t* src_bgra, const uint8_t* dst_bgra, uint8_t* mask_agrb, int stride, int src_w, int src_h) noexcept;