Use a C array for transparency LUT for debug perf

In a debug build, `std::array` accesses are function calls.

Timedemo on my machine:
* Before: 45.39 seconds	141.3 FPS
* After:  40.20 seconds	159.5 FPS

Source/engine.cpp

@@ -11,7 +11,6 @@
  * - Video playback
  */
 
-#include <array>
 #include <cassert>
 #include <cstdint>
 #include <cstring>
@@ -26,7 +25,7 @@ namespace {
 void DrawHalfTransparentUnalignedBlendedRectTo(const Surface &out, unsigned sx, unsigned sy, unsigned width, unsigned height)
 {
 	uint8_t *pix = out.at(static_cast<int>(sx), static_cast<int>(sy));
-	const std::array<uint8_t, 256> &lookupTable = paletteTransparencyLookup[0];
+	const uint8_t *const lookupTable = paletteTransparencyLookup[0];
 	const unsigned skipX = out.pitch() - width;
 	for (unsigned y = 0; y < height; ++y) {
 		for (unsigned x = 0; x < width; ++x, ++pix) {

Source/engine/palette.cpp

@@ -24,7 +24,11 @@ namespace devilution {
 std::array<SDL_Color, 256> logical_palette;
 std::array<SDL_Color, 256> system_palette;
 std::array<SDL_Color, 256> orig_palette;
-std::array<std::array<Uint8, 256>, 256> paletteTransparencyLookup;
+
+// This array is read from a lot on every frame.
+// We do not use `std::array` here to improve debug build performance.
+// In a debug build, `std::array` accesses are function calls.
+Uint8 paletteTransparencyLookup[256][256];
 
 #if DEVILUTIONX_PALETTE_TRANSPARENCY_BLACK_16_LUT
 uint16_t paletteTransparencyLookupBlack16[65536];
@@ -124,10 +128,10 @@ void CycleColors(int from, int to)
 	std::rotate(system_palette.begin() + from, system_palette.begin() + from + 1, system_palette.begin() + to + 1);
 
 	for (auto &palette : paletteTransparencyLookup) {
-		std::rotate(palette.begin() + from, palette.begin() + from + 1, palette.begin() + to + 1);
+		std::rotate(std::begin(palette) + from, std::begin(palette) + from + 1, std::begin(palette) + to + 1);
 	}
 
-	std::rotate(paletteTransparencyLookup.begin() + from, paletteTransparencyLookup.begin() + from + 1, paletteTransparencyLookup.begin() + to + 1);
+	std::rotate(&paletteTransparencyLookup[from][0], &paletteTransparencyLookup[from + 1][0], &paletteTransparencyLookup[to + 1][0]);
 }
 
 /**
@@ -140,10 +144,10 @@ void CycleColorsReverse(int from, int to)
 	std::rotate(system_palette.begin() + from, system_palette.begin() + to, system_palette.begin() + to + 1);
 
 	for (auto &palette : paletteTransparencyLookup) {
-		std::rotate(palette.begin() + from, palette.begin() + to, palette.begin() + to + 1);
+		std::rotate(std::begin(palette) + from, std::begin(palette) + to, std::begin(palette) + to + 1);
 	}
 
-	std::rotate(paletteTransparencyLookup.begin() + from, paletteTransparencyLookup.begin() + to, paletteTransparencyLookup.begin() + to + 1);
+	std::rotate(&paletteTransparencyLookup[from][0], &paletteTransparencyLookup[to][0], &paletteTransparencyLookup[to + 1][0]);
 }
 
 } // namespace

Source/engine/palette.h

@@ -36,7 +36,7 @@ extern std::array<SDL_Color, 256> logical_palette;
 extern std::array<SDL_Color, 256> system_palette;
 extern std::array<SDL_Color, 256> orig_palette;
 /** Lookup table for transparency */
-extern std::array<std::array<Uint8, 256>, 256> paletteTransparencyLookup;
+extern Uint8 paletteTransparencyLookup[256][256];
 
 #if DEVILUTIONX_PALETTE_TRANSPARENCY_BLACK_16_LUT
 /**

Source/engine/render/blit_impl.hpp

@@ -72,7 +72,7 @@ DVL_ALWAYS_INLINE DVL_ATTRIBUTE_HOT void BlitFillBlended(uint8_t *dst, unsigned
 {
 	assert(length != 0);
 	const uint8_t *end = dst + length;
-	const std::array<uint8_t, 256> &tbl = paletteTransparencyLookup[color];
+	const uint8_t *const tbl = paletteTransparencyLookup[color];
 	while (dst + 3 < end) {
 		*dst = tbl[*dst];
 		++dst;

Source/utils/sdl_bilinear_scale.cpp

@@ -148,7 +148,7 @@ void BilinearScale32(SDL_Surface *src, SDL_Surface *dst)
 	}
 }
 
-void BilinearDownscaleByHalf8(const SDL_Surface *src, const std::array<std::array<Uint8, 256>, 256> &paletteBlendingTable, SDL_Surface *dst, uint8_t transparentIndex)
+void BilinearDownscaleByHalf8(const SDL_Surface *src, const Uint8 paletteBlendingTable[256][256], SDL_Surface *dst, uint8_t transparentIndex)
 {
 	const auto *const srcPixelsBegin = static_cast<const uint8_t *>(src->pixels)
 	    + static_cast<size_t>(src->clip_rect.y * src->pitch + src->clip_rect.x);

Source/utils/sdl_bilinear_scale.hpp

@@ -1,6 +1,5 @@
 #pragma once
 
-#include <array>
 #include <cstdint>
 
 #include <SDL_version.h>
@@ -23,6 +22,6 @@ void BilinearScale32(SDL_Surface *src, SDL_Surface *dst);
  * @brief Streamlined bilinear downscaling using blended transparency table.
  * Requires `src` and `dst` to have the same pixel format (INDEX8).
  */
-void BilinearDownscaleByHalf8(const SDL_Surface *src, const std::array<std::array<Uint8, 256>, 256> &paletteBlendingTable, SDL_Surface *dst, uint8_t transparentIndex);
+void BilinearDownscaleByHalf8(const SDL_Surface *src, const Uint8 paletteBlendingTable[256][256], SDL_Surface *dst, uint8_t transparentIndex);
 
 } // namespace devilution