From fe2864d3962c38af11f513d0287792bbf03619a2 Mon Sep 17 00:00:00 2001 From: Magnus Norddahl Date: Sat, 22 Apr 2017 22:15:09 +0200 Subject: [PATCH] - add the AVX-2 softpoly drawer to the repository for completeness sake (not currently used) --- src/polyrenderer/drawers/poly_drawer32_avx2.h | 739 ++++++++++++++++++ 1 file changed, 739 insertions(+) create mode 100644 src/polyrenderer/drawers/poly_drawer32_avx2.h diff --git a/src/polyrenderer/drawers/poly_drawer32_avx2.h b/src/polyrenderer/drawers/poly_drawer32_avx2.h new file mode 100644 index 0000000000..9091ae21ae --- /dev/null +++ b/src/polyrenderer/drawers/poly_drawer32_avx2.h @@ -0,0 +1,739 @@ +/* +** Projected triangle drawer +** Copyright (c) 2016 Magnus Norddahl +** +** This software is provided 'as-is', without any express or implied +** warranty. In no event will the authors be held liable for any damages +** arising from the use of this software. +** +** Permission is granted to anyone to use this software for any purpose, +** including commercial applications, and to alter it and redistribute it +** freely, subject to the following restrictions: +** +** 1. The origin of this software must not be misrepresented; you must not +** claim that you wrote the original software. If you use this software +** in a product, an acknowledgment in the product documentation would be +** appreciated but is not required. +** 2. Altered source versions must be plainly marked as such, and must not be +** misrepresented as being the original software. +** 3. This notice may not be removed or altered from any source distribution. +** +*/ + +#pragma once + +#include "screen_triangle.h" + +#ifdef _MSC_VER +#pragma warning(disable: 4752) // warning C4752 : found Intel(R) Advanced Vector Extensions; consider using /arch:AVX +#endif + +namespace TriScreenDrawerModes +{ + template + FORCEINLINE unsigned int VECTORCALL Sample32_AVX2(int32_t u, int32_t v, const uint32_t *texPixels, int texWidth, int texHeight, uint32_t oneU, uint32_t oneV, uint32_t color, const uint32_t *translation) + { + uint32_t texel; + if (SamplerT::Mode == (int)Samplers::Shaded || SamplerT::Mode == (int)Samplers::Stencil || SamplerT::Mode == (int)Samplers::Fill || SamplerT::Mode == (int)Samplers::Fuzz) + { + return color; + } + else if (SamplerT::Mode == (int)Samplers::Translated) + { + const uint8_t *texpal = (const uint8_t *)texPixels; + uint32_t texelX = ((((uint32_t)u << 8) >> 16) * texWidth) >> 16; + uint32_t texelY = ((((uint32_t)v << 8) >> 16) * texHeight) >> 16; + return translation[texpal[texelX * texHeight + texelY]]; + } + else if (FilterModeT::Mode == (int)FilterModes::Nearest) + { + uint32_t texelX = ((((uint32_t)u << 8) >> 16) * texWidth) >> 16; + uint32_t texelY = ((((uint32_t)v << 8) >> 16) * texHeight) >> 16; + texel = texPixels[texelX * texHeight + texelY]; + } + else + { + u -= oneU >> 1; + v -= oneV >> 1; + + unsigned int frac_x0 = (((uint32_t)u << 8) >> FRACBITS) * texWidth; + unsigned int frac_x1 = ((((uint32_t)u << 8) + oneU) >> FRACBITS) * texWidth; + unsigned int frac_y0 = (((uint32_t)v << 8) >> FRACBITS) * texHeight; + unsigned int frac_y1 = ((((uint32_t)v << 8) + oneV) >> FRACBITS) * texHeight; + unsigned int x0 = frac_x0 >> FRACBITS; + unsigned int x1 = frac_x1 >> FRACBITS; + unsigned int y0 = frac_y0 >> FRACBITS; + unsigned int y1 = frac_y1 >> FRACBITS; + + unsigned int p00 = texPixels[x0 * texHeight + y0]; + unsigned int p01 = texPixels[x0 * texHeight + y1]; + unsigned int p10 = texPixels[x1 * texHeight + y0]; + unsigned int p11 = texPixels[x1 * texHeight + y1]; + + unsigned int inv_a = (frac_x1 >> (FRACBITS - 4)) & 15; + unsigned int inv_b = (frac_y1 >> (FRACBITS - 4)) & 15; + unsigned int a = 16 - inv_a; + unsigned int b = 16 - inv_b; + + unsigned int sred = (RPART(p00) * (a * b) + RPART(p01) * (inv_a * b) + RPART(p10) * (a * inv_b) + RPART(p11) * (inv_a * inv_b) + 127) >> 8; + unsigned int sgreen = (GPART(p00) * (a * b) + GPART(p01) * (inv_a * b) + GPART(p10) * (a * inv_b) + GPART(p11) * (inv_a * inv_b) + 127) >> 8; + unsigned int sblue = (BPART(p00) * (a * b) + BPART(p01) * (inv_a * b) + BPART(p10) * (a * inv_b) + BPART(p11) * (inv_a * inv_b) + 127) >> 8; + unsigned int salpha = (APART(p00) * (a * b) + APART(p01) * (inv_a * b) + APART(p10) * (a * inv_b) + APART(p11) * (inv_a * inv_b) + 127) >> 8; + + texel = (salpha << 24) | (sred << 16) | (sgreen << 8) | sblue; + } + + if (SamplerT::Mode == (int)Samplers::Skycap) + { + int start_fade = 2; // How fast it should fade out + + int alpha_top = clamp(v >> (16 - start_fade), 0, 256); + int alpha_bottom = clamp(((2 << 24) - v) >> (16 - start_fade), 0, 256); + int a = MIN(alpha_top, alpha_bottom); + int inv_a = 256 - a; + + uint32_t r = RPART(texel); + uint32_t g = GPART(texel); + uint32_t b = BPART(texel); + uint32_t fg_a = APART(texel); + uint32_t bg_red = RPART(color); + uint32_t bg_green = GPART(color); + uint32_t bg_blue = BPART(color); + r = (r * a + bg_red * inv_a + 127) >> 8; + g = (g * a + bg_green * inv_a + 127) >> 8; + b = (b * a + bg_blue * inv_a + 127) >> 8; + return MAKEARGB(fg_a, r, g, b); + } + else + { + return texel; + } + } + + template + FORCEINLINE unsigned int VECTORCALL SampleShade32_AVX2(int32_t u, int32_t v, const uint32_t *texPixels, int texWidth, int texHeight, int &fuzzpos) + { + if (SamplerT::Mode == (int)Samplers::Shaded) + { + const uint8_t *texpal = (const uint8_t *)texPixels; + uint32_t texelX = ((((uint32_t)u << 8) >> 16) * texWidth) >> 16; + uint32_t texelY = ((((uint32_t)v << 8) >> 16) * texHeight) >> 16; + unsigned int sampleshadeout = texpal[texelX * texHeight + texelY]; + sampleshadeout += sampleshadeout >> 7; // 255 -> 256 + return sampleshadeout; + } + else if (SamplerT::Mode == (int)Samplers::Stencil) + { + uint32_t texelX = ((((uint32_t)u << 8) >> 16) * texWidth) >> 16; + uint32_t texelY = ((((uint32_t)v << 8) >> 16) * texHeight) >> 16; + unsigned int sampleshadeout = APART(texPixels[texelX * texHeight + texelY]); + sampleshadeout += sampleshadeout >> 7; // 255 -> 256 + return sampleshadeout; + } + else if (SamplerT::Mode == (int)Samplers::Fuzz) + { + uint32_t texelX = ((((uint32_t)u << 8) >> 16) * texWidth) >> 16; + uint32_t texelY = ((((uint32_t)v << 8) >> 16) * texHeight) >> 16; + unsigned int sampleshadeout = APART(texPixels[texelX * texHeight + texelY]); + sampleshadeout += sampleshadeout >> 7; // 255 -> 256 + sampleshadeout = (sampleshadeout * fuzzcolormap[fuzzpos++]) >> 5; + if (fuzzpos >= FUZZTABLE) fuzzpos = 0; + return sampleshadeout; + } + else + { + return 0; + } + } + + template + FORCEINLINE __m256i VECTORCALL Shade32_AVX2(__m256i fgcolor, __m256i mlight, __m256i desaturate, __m256i inv_desaturate, __m256i shade_fade, __m256i shade_light) + { + if (ShadeModeT::Mode == (int)ShadeMode::Simple) + { + fgcolor = _mm256_srli_epi16(_mm256_mullo_epi16(fgcolor, mlight), 8); + } + else if (ShadeModeT::Mode == (int)ShadeMode::Advanced) + { + __m256i intensity = _mm256_mullo_epi16(fgcolor, _mm256_set_epi16(0, 77, 143, 37, 0, 77, 143, 37, 0, 77, 143, 37, 0, 77, 143, 37)); + intensity = _mm256_add_epi16(intensity, _mm256_srli_epi64(intensity, 32)); + intensity = _mm256_add_epi16(intensity, _mm256_srli_epi64(intensity, 16)); + intensity = _mm256_srli_epi16(intensity, 8); + intensity = _mm256_mullo_epi16(intensity, desaturate); + intensity = _mm256_shufflehi_epi16(_mm256_shufflelo_epi16(intensity, _MM_SHUFFLE(3, 0, 0, 0)), _MM_SHUFFLE(3, 0, 0, 0)); + + fgcolor = _mm256_srli_epi16(_mm256_add_epi16(_mm256_mullo_epi16(fgcolor, inv_desaturate), intensity), 8); + fgcolor = _mm256_mullo_epi16(fgcolor, mlight); + fgcolor = _mm256_srli_epi16(_mm256_add_epi16(shade_fade, fgcolor), 8); + fgcolor = _mm256_srli_epi16(_mm256_mullo_epi16(fgcolor, shade_light), 8); + } + return fgcolor; + } + + template + FORCEINLINE __m256i VECTORCALL Blend32_AVX2(__m256i fgcolor, __m256i bgcolor, __m256i ifgcolor, __m256i ifgshade, __m256i srcalpha, __m256i destalpha) + { + if (BlendT::Mode == (int)BlendModes::Opaque) + { + __m256i outcolor = fgcolor; + outcolor = _mm256_packus_epi16(outcolor, _mm256_setzero_si256()); + return outcolor; + } + else if (BlendT::Mode == (int)BlendModes::Masked) + { + __m256i mask = _mm256_cmpeq_epi32(_mm256_packus_epi16(fgcolor, _mm256_setzero_si256()), _mm256_setzero_si256()); + mask = _mm256_unpacklo_epi8(mask, _mm256_setzero_si256()); + __m256i outcolor = _mm256_or_si256(_mm256_and_si256(mask, bgcolor), _mm256_andnot_si256(mask, fgcolor)); + outcolor = _mm256_packus_epi16(outcolor, _mm256_setzero_si256()); + outcolor = _mm256_or_si256(outcolor, _mm256_set1_epi32(0xff000000)); + return outcolor; + } + else if (BlendT::Mode == (int)BlendModes::AddSrcColorOneMinusSrcColor) + { + __m256i inv_srccolor = _mm256_sub_epi16(_mm256_set1_epi16(256), _mm256_add_epi16(fgcolor, _mm256_srli_epi16(fgcolor, 7))); + __m256i outcolor = _mm256_add_epi16(fgcolor, _mm256_srli_epi16(_mm256_mullo_epi16(bgcolor, inv_srccolor), 8)); + outcolor = _mm256_packus_epi16(outcolor, _mm256_setzero_si256()); + return outcolor; + } + else if (BlendT::Mode == (int)BlendModes::Shaded) + { + ifgshade = _mm256_srli_epi32(_mm256_add_epi32(_mm256_mul_epu32(ifgshade, srcalpha), _mm256_set1_epi32(128)), 8); + __m256i alpha = _mm256_shufflehi_epi16(_mm256_shufflelo_epi16(ifgshade, _MM_SHUFFLE(0, 0, 0, 0)), _MM_SHUFFLE(0, 0, 0, 0)); + __m256i inv_alpha = _mm256_sub_epi16(_mm256_set1_epi16(256), alpha); + + fgcolor = _mm256_mullo_epi16(fgcolor, alpha); + bgcolor = _mm256_mullo_epi16(bgcolor, inv_alpha); + __m256i outcolor = _mm256_srli_epi16(_mm256_add_epi16(fgcolor, bgcolor), 8); + outcolor = _mm256_packus_epi16(outcolor, _mm256_setzero_si256()); + outcolor = _mm256_or_si256(outcolor, _mm256_set1_epi32(0xff000000)); + return outcolor; + } + else if (BlendT::Mode == (int)BlendModes::AddClampShaded) + { + ifgshade = _mm256_srli_epi32(_mm256_add_epi32(_mm256_mul_epu32(ifgshade, srcalpha), _mm256_set1_epi32(128)), 8); + __m256i alpha = _mm256_shufflehi_epi16(_mm256_shufflelo_epi16(ifgshade, _MM_SHUFFLE(0, 0, 0, 0)), _MM_SHUFFLE(0, 0, 0, 0)); + __m256i inv_alpha = _mm256_sub_epi16(_mm256_set1_epi16(256), alpha); + + fgcolor = _mm256_srli_epi16(_mm256_mullo_epi16(fgcolor, alpha), 8); + __m256i outcolor = _mm256_add_epi16(fgcolor, bgcolor); + outcolor = _mm256_packus_epi16(outcolor, _mm256_setzero_si256()); + outcolor = _mm256_or_si256(outcolor, _mm256_set1_epi32(0xff000000)); + return outcolor; + } + else + { + __m256i alpha = _mm256_shufflehi_epi16(_mm256_shufflelo_epi16(ifgcolor, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)); + alpha = _mm256_srli_epi16(_mm256_add_epi16(alpha, _mm256_srli_epi16(alpha, 7)), 1); // 255->128 + __m256i inv_alpha = _mm256_sub_epi16(_mm256_set1_epi16(128), alpha); + + __m256i bgalpha = _mm256_srli_epi16(_mm256_add_epi16(_mm256_add_epi16(_mm256_mullo_epi16(destalpha, alpha), _mm256_slli_epi16(inv_alpha, 8)), _mm256_set1_epi32(64)), 7); + __m256i fgalpha = _mm256_srli_epi16(_mm256_add_epi16(_mm256_mullo_epi16(srcalpha, alpha), _mm256_set1_epi32(64)), 7); + + fgcolor = _mm256_mullo_epi16(fgcolor, fgalpha); + bgcolor = _mm256_mullo_epi16(bgcolor, bgalpha); + + __m256i fg_lo = _mm256_unpacklo_epi16(fgcolor, _mm256_setzero_si256()); + __m256i bg_lo = _mm256_unpacklo_epi16(bgcolor, _mm256_setzero_si256()); + __m256i fg_hi = _mm256_unpackhi_epi16(fgcolor, _mm256_setzero_si256()); + __m256i bg_hi = _mm256_unpackhi_epi16(bgcolor, _mm256_setzero_si256()); + + __m256i out_lo, out_hi; + if (BlendT::Mode == (int)BlendModes::AddClamp) + { + out_lo = _mm256_add_epi32(fg_lo, bg_lo); + out_hi = _mm256_add_epi32(fg_hi, bg_hi); + } + else if (BlendT::Mode == (int)BlendModes::SubClamp) + { + out_lo = _mm256_sub_epi32(fg_lo, bg_lo); + out_hi = _mm256_sub_epi32(fg_hi, bg_hi); + } + else if (BlendT::Mode == (int)BlendModes::RevSubClamp) + { + out_lo = _mm256_sub_epi32(bg_lo, fg_lo); + out_hi = _mm256_sub_epi32(bg_hi, fg_hi); + } + + out_lo = _mm256_srai_epi32(out_lo, 8); + out_hi = _mm256_srai_epi32(out_hi, 8); + __m256i outcolor = _mm256_packs_epi32(out_lo, out_hi); + outcolor = _mm256_packus_epi16(outcolor, _mm256_setzero_si256()); + outcolor = _mm256_or_si256(outcolor, _mm256_set1_epi32(0xff000000)); + return outcolor; + } + } +} + +template +class TriScreenDrawer32_AVX2 +{ +public: + static void Execute(int x, int y, uint32_t mask0, uint32_t mask1, const TriDrawTriangleArgs *args) + { + using namespace TriScreenDrawerModes; + + bool is_simple_shade = args->uniforms->SimpleShade(); + + if (SamplerT::Mode == (int)Samplers::Texture) + { + bool is_nearest_filter = args->uniforms->NearestFilter(); + + if (is_simple_shade) + { + if (is_nearest_filter) + DrawBlock(x, y, mask0, mask1, args); + else + DrawBlock(x, y, mask0, mask1, args); + } + else + { + if (is_nearest_filter) + DrawBlock(x, y, mask0, mask1, args); + else + DrawBlock(x, y, mask0, mask1, args); + } + } + else if (SamplerT::Mode == (int)Samplers::Fuzz) + { + DrawBlock(x, y, mask0, mask1, args); + } + else // no linear filtering for translated, shaded, stencil, fill or skycap + { + if (is_simple_shade) + { + DrawBlock(x, y, mask0, mask1, args); + } + else + { + DrawBlock(x, y, mask0, mask1, args); + } + } + } + +private: + template + FORCEINLINE static void VECTORCALL DrawBlock(int destX, int destY, uint32_t mask0, uint32_t mask1, const TriDrawTriangleArgs *args) + { + using namespace TriScreenDrawerModes; + + bool is_fixed_light = args->uniforms->FixedLight(); + __m128i lightmask = _mm_set1_epi32(is_fixed_light ? 0 : 0xffffffff); + __m256i srcalpha = _mm256_set1_epi16(args->uniforms->SrcAlpha()); + __m256i destalpha = _mm256_set1_epi16(args->uniforms->DestAlpha()); + + int fuzzpos = (ScreenTriangle::FuzzStart + destX * 123 + destY) % FUZZTABLE; + + // Light + uint32_t light = args->uniforms->Light(); + float shade = MIN(2.0f - (light + 12.0f) / 128.0f, 31.0f / 32.0f); + float globVis = args->uniforms->GlobVis() * (1.0f / 32.0f); + light += (light >> 7); // 255 -> 256 + light <<= 8; + __m128i fixedlight = _mm_set1_epi32(light); + + // Calculate gradients + const TriVertex &v1 = *args->v1; + __m128 gradientX = _mm_setr_ps(args->gradientX.W, args->gradientX.U, args->gradientX.V, 0.0f); + __m128 gradientY = _mm_setr_ps(args->gradientY.W, args->gradientY.U, args->gradientY.V, 0.0f); + __m128 blockPosY = _mm_add_ps(_mm_add_ps( + _mm_setr_ps(v1.w, v1.u * v1.w, v1.v * v1.w, globVis), + _mm_mul_ps(gradientX, _mm_set1_ps(destX - v1.x))), + _mm_mul_ps(gradientY, _mm_set1_ps(destY - v1.y))); + gradientX = _mm_mul_ps(gradientX, _mm_set1_ps(8.0f)); + + // Output + uint32_t * RESTRICT destOrg = (uint32_t*)args->dest; + int pitch = args->pitch; + uint32_t *dest = destOrg + destX + destY * pitch; + int offset_next_line = pitch - 8; + + // Sampling stuff + uint32_t color = args->uniforms->Color(); + const uint32_t * RESTRICT translation = (const uint32_t *)args->uniforms->Translation(); + const uint32_t * RESTRICT texPixels = (const uint32_t *)args->uniforms->TexturePixels(); + uint32_t texWidth = args->uniforms->TextureWidth(); + uint32_t texHeight = args->uniforms->TextureHeight(); + uint32_t oneU, oneV; + if (SamplerT::Mode != (int)Samplers::Fill) + { + oneU = ((0x800000 + texWidth - 1) / texWidth) * 2 + 1; + oneV = ((0x800000 + texHeight - 1) / texHeight) * 2 + 1; + } + else + { + oneU = 0; + oneV = 0; + } + + // Shade constants + __m256i inv_desaturate, shade_fade, shade_light; + __m256i desaturate; + if (ShadeModeT::Mode == (int)ShadeMode::Advanced) + { + inv_desaturate = _mm256_setr_epi16( + 256, 256 - args->uniforms->ShadeDesaturate(), 256 - args->uniforms->ShadeDesaturate(), 256 - args->uniforms->ShadeDesaturate(), + 256, 256 - args->uniforms->ShadeDesaturate(), 256 - args->uniforms->ShadeDesaturate(), 256 - args->uniforms->ShadeDesaturate(), + 256, 256 - args->uniforms->ShadeDesaturate(), 256 - args->uniforms->ShadeDesaturate(), 256 - args->uniforms->ShadeDesaturate(), + 256, 256 - args->uniforms->ShadeDesaturate(), 256 - args->uniforms->ShadeDesaturate(), 256 - args->uniforms->ShadeDesaturate()); + shade_fade = _mm256_set_epi16( + args->uniforms->ShadeFadeAlpha(), args->uniforms->ShadeFadeRed(), args->uniforms->ShadeFadeGreen(), args->uniforms->ShadeFadeBlue(), + args->uniforms->ShadeFadeAlpha(), args->uniforms->ShadeFadeRed(), args->uniforms->ShadeFadeGreen(), args->uniforms->ShadeFadeBlue(), + args->uniforms->ShadeFadeAlpha(), args->uniforms->ShadeFadeRed(), args->uniforms->ShadeFadeGreen(), args->uniforms->ShadeFadeBlue(), + args->uniforms->ShadeFadeAlpha(), args->uniforms->ShadeFadeRed(), args->uniforms->ShadeFadeGreen(), args->uniforms->ShadeFadeBlue()); + shade_light = _mm256_set_epi16( + args->uniforms->ShadeLightAlpha(), args->uniforms->ShadeLightRed(), args->uniforms->ShadeLightGreen(), args->uniforms->ShadeLightBlue(), + args->uniforms->ShadeLightAlpha(), args->uniforms->ShadeLightRed(), args->uniforms->ShadeLightGreen(), args->uniforms->ShadeLightBlue(), + args->uniforms->ShadeLightAlpha(), args->uniforms->ShadeLightRed(), args->uniforms->ShadeLightGreen(), args->uniforms->ShadeLightBlue(), + args->uniforms->ShadeLightAlpha(), args->uniforms->ShadeLightRed(), args->uniforms->ShadeLightGreen(), args->uniforms->ShadeLightBlue()); + desaturate = _mm256_sub_epi16(_mm256_set1_epi16(256), inv_desaturate); + } + else + { + inv_desaturate = _mm256_setzero_si256(); + shade_fade = _mm256_setzero_si256(); + shade_fade = _mm256_setzero_si256(); + shade_light = _mm256_setzero_si256(); + desaturate = _mm256_setzero_si256(); + } + + if (mask0 == 0xffffffff && mask1 == 0xffffffff) + { + for (int y = 0; y < 8; y++) + { + __m128 blockPosX = _mm_add_ps(blockPosY, gradientX); + __m128 W = _mm_shuffle_ps(blockPosY, blockPosX, _MM_SHUFFLE(0, 0, 0, 0)); + __m128 rcpW = _mm_div_ps(_mm_set1_ps((float)0x01000000), W); + __m128i posUV = _mm_cvtps_epi32(_mm_mul_ps(_mm_shuffle_ps(blockPosY, blockPosX, _MM_SHUFFLE(2, 1, 2, 1)), rcpW)); + + __m128 vis = _mm_mul_ps(_mm_shuffle_ps(blockPosY, blockPosX, _MM_SHUFFLE(3, 3, 3, 3)), W); + __m128i lightpospair = _mm_sub_epi32( + _mm_set1_epi32(FRACUNIT), + _mm_cvtps_epi32(_mm_mul_ps( + _mm_max_ps(_mm_sub_ps(_mm_set1_ps(shade), _mm_min_ps(_mm_set1_ps(24.0f / 32.0f), vis)), _mm_setzero_ps()), + _mm_set1_ps((float)FRACUNIT)))); + lightpospair = _mm_or_si128(_mm_and_si128(lightmask, lightpospair), _mm_andnot_si128(lightmask, fixedlight)); + + int32_t posU = _mm_cvtsi128_si32(posUV); + int32_t posV = _mm_cvtsi128_si32(_mm_srli_si128(posUV, 4)); + int32_t nextU = _mm_cvtsi128_si32(_mm_srli_si128(posUV, 8)); + int32_t nextV = _mm_cvtsi128_si32(_mm_srli_si128(posUV, 12)); + int32_t lightpos = _mm_cvtsi128_si32(lightpospair); + int32_t lightnext = _mm_cvtsi128_si32(_mm_srli_si128(lightpospair, 8)); + int32_t stepU = (nextU - posU) >> 3; + int32_t stepV = (nextV - posV) >> 3; + fixed_t lightstep = (lightnext - lightpos) >> 3; + + for (int ix = 0; ix < 2; ix++) + { + // Load bgcolor + __m256i bgcolor; + if (BlendT::Mode != (int)BlendModes::Opaque) + { + __m128i bgpacked = _mm_loadu_si128((__m128i*)dest); + bgcolor = _mm256_set_m128i(_mm_unpackhi_epi8(bgpacked, _mm_setzero_si128()), _mm_unpacklo_epi8(bgpacked, _mm_setzero_si128())); + } + else + bgcolor = _mm256_setzero_si256(); + + // Sample fgcolor + unsigned int ifgcolor0 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade0 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + unsigned int ifgcolor1 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade1 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + unsigned int ifgcolor2 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade2 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + unsigned int ifgcolor3 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade3 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + // Setup light + int lightpos0 = lightpos >> 8; + lightpos += lightstep; + int lightpos1 = lightpos >> 8; + lightpos += lightstep; + int lightpos2 = lightpos >> 8; + lightpos += lightstep; + int lightpos3 = lightpos >> 8; + lightpos += lightstep; + __m256i mlight = _mm256_set_epi16( + 256, lightpos3, lightpos3, lightpos3, + 256, lightpos2, lightpos2, lightpos2, + 256, lightpos1, lightpos1, lightpos1, + 256, lightpos0, lightpos0, lightpos0); + + __m256i shade_fade_lit; + if (ShadeModeT::Mode == (int)ShadeMode::Advanced) + { + __m256i inv_light = _mm256_sub_epi16(_mm256_set_epi16(0, 256, 256, 256, 0, 256, 256, 256, 0, 256, 256, 256, 0, 256, 256, 256), mlight); + shade_fade_lit = _mm256_mullo_epi16(shade_fade, inv_light); + } + else + { + shade_fade_lit = _mm256_setzero_si256(); + } + + // Shade and blend + __m128i fgpacked = _mm_set_epi32(ifgcolor3, ifgcolor2, ifgcolor1, ifgcolor0); + __m128i shadepacked = _mm_set_epi32(ifgshade3, ifgshade2, ifgshade1, ifgshade0); + __m256i mifgcolor = _mm256_set_m128i(_mm_unpackhi_epi8(fgpacked, _mm_setzero_si128()), _mm_unpacklo_epi8(fgpacked, _mm_setzero_si128())); + __m256i mifgshade = _mm256_set_m128i(_mm_unpackhi_epi32(shadepacked, shadepacked), _mm_unpacklo_epi32(shadepacked, shadepacked)); + __m256i fgcolor = mifgcolor; + fgcolor = Shade32_AVX2(fgcolor, mlight, desaturate, inv_desaturate, shade_fade_lit, shade_light); + __m256i outcolor = Blend32_AVX2(fgcolor, bgcolor, mifgcolor, mifgshade, srcalpha, destalpha); + + // Store result + _mm_storeu_si128((__m128i*)dest, _mm_or_si128(_mm256_extracti128_si256(outcolor, 0), _mm_slli_si128(_mm256_extracti128_si256(outcolor, 1), 8))); + dest += 4; + } + + blockPosY = _mm_add_ps(blockPosY, gradientY); + + dest += offset_next_line; + } + } + else + { + // mask0 loop: + for (int y = 0; y < 4; y++) + { + __m128 blockPosX = _mm_add_ps(blockPosY, gradientX); + __m128 W = _mm_shuffle_ps(blockPosY, blockPosX, _MM_SHUFFLE(0, 0, 0, 0)); + __m128 rcpW = _mm_div_ps(_mm_set1_ps((float)0x01000000), W); + __m128i posUV = _mm_cvtps_epi32(_mm_mul_ps(_mm_shuffle_ps(blockPosY, blockPosX, _MM_SHUFFLE(2, 1, 2, 1)), rcpW)); + + __m128 vis = _mm_mul_ps(_mm_shuffle_ps(blockPosY, blockPosX, _MM_SHUFFLE(3, 3, 3, 3)), W); + __m128i lightpospair = _mm_sub_epi32( + _mm_set1_epi32(FRACUNIT), + _mm_cvtps_epi32(_mm_mul_ps( + _mm_max_ps(_mm_sub_ps(_mm_set1_ps(shade), _mm_min_ps(_mm_set1_ps(24.0f / 32.0f), vis)), _mm_setzero_ps()), + _mm_set1_ps((float)FRACUNIT)))); + lightpospair = _mm_or_si128(_mm_and_si128(lightmask, lightpospair), _mm_andnot_si128(lightmask, fixedlight)); + + int32_t posU = _mm_cvtsi128_si32(posUV); + int32_t posV = _mm_cvtsi128_si32(_mm_srli_si128(posUV, 4)); + int32_t nextU = _mm_cvtsi128_si32(_mm_srli_si128(posUV, 8)); + int32_t nextV = _mm_cvtsi128_si32(_mm_srli_si128(posUV, 12)); + int32_t lightpos = _mm_cvtsi128_si32(lightpospair); + int32_t lightnext = _mm_cvtsi128_si32(_mm_srli_si128(lightpospair, 8)); + int32_t stepU = (nextU - posU) >> 3; + int32_t stepV = (nextV - posV) >> 3; + fixed_t lightstep = (lightnext - lightpos) >> 3; + + for (int x = 0; x < 2; x++) + { + // Load bgcolor + uint32_t desttmp[4]; + __m256i bgcolor; + if (BlendT::Mode != (int)BlendModes::Opaque) + { + if (mask0 & (1 << 31)) desttmp[0] = dest[0]; + if (mask0 & (1 << 30)) desttmp[1] = dest[1]; + if (mask0 & (1 << 29)) desttmp[2] = dest[2]; + if (mask0 & (1 << 28)) desttmp[3] = dest[3]; + + __m128i bgpacked = _mm_loadu_si128((__m128i*)(desttmp)); + bgcolor = _mm256_set_m128i(_mm_unpackhi_epi8(bgpacked, _mm_setzero_si128()), _mm_unpacklo_epi8(bgpacked, _mm_setzero_si128())); + } + else + bgcolor = _mm256_setzero_si256(); + + // Sample fgcolor + unsigned int ifgcolor0 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade0 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + unsigned int ifgcolor1 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade1 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + unsigned int ifgcolor2 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade2 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + unsigned int ifgcolor3 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade3 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + // Setup light + int lightpos0 = lightpos >> 8; + lightpos += lightstep; + int lightpos1 = lightpos >> 8; + lightpos += lightstep; + int lightpos2 = lightpos >> 8; + lightpos += lightstep; + int lightpos3 = lightpos >> 8; + lightpos += lightstep; + __m256i mlight = _mm256_set_epi16( + 256, lightpos3, lightpos3, lightpos3, + 256, lightpos2, lightpos2, lightpos2, + 256, lightpos1, lightpos1, lightpos1, + 256, lightpos0, lightpos0, lightpos0); + + __m256i shade_fade_lit; + if (ShadeModeT::Mode == (int)ShadeMode::Advanced) + { + __m256i inv_light = _mm256_sub_epi16(_mm256_set_epi16(0, 256, 256, 256, 0, 256, 256, 256, 0, 256, 256, 256, 0, 256, 256, 256), mlight); + shade_fade_lit = _mm256_mullo_epi16(shade_fade, inv_light); + } + else + { + shade_fade_lit = _mm256_setzero_si256(); + } + + // Shade and blend + __m128i fgpacked = _mm_set_epi32(ifgcolor3, ifgcolor2, ifgcolor1, ifgcolor0); + __m128i shadepacked = _mm_set_epi32(ifgshade3, ifgshade2, ifgshade1, ifgshade0); + __m256i mifgcolor = _mm256_set_m128i(_mm_unpackhi_epi8(fgpacked, _mm_setzero_si128()), _mm_unpacklo_epi8(fgpacked, _mm_setzero_si128())); + __m256i mifgshade = _mm256_set_m128i(_mm_unpackhi_epi32(shadepacked, shadepacked), _mm_unpacklo_epi32(shadepacked, shadepacked)); + __m256i fgcolor = mifgcolor; + fgcolor = Shade32_AVX2(fgcolor, mlight, desaturate, inv_desaturate, shade_fade_lit, shade_light); + __m256i outcolor = Blend32_AVX2(fgcolor, bgcolor, mifgcolor, mifgshade, srcalpha, destalpha); + + // Store result + _mm_storeu_si128((__m128i*)desttmp, _mm_or_si128(_mm256_extracti128_si256(outcolor, 0), _mm_slli_si128(_mm256_extracti128_si256(outcolor, 1), 8))); + if (mask0 & (1 << 31)) dest[0] = desttmp[0]; + if (mask0 & (1 << 30)) dest[1] = desttmp[1]; + if (mask0 & (1 << 29)) dest[2] = desttmp[2]; + if (mask0 & (1 << 28)) dest[3] = desttmp[3]; + + mask0 <<= 4; + dest += 4; + } + + blockPosY = _mm_add_ps(blockPosY, gradientY); + + dest += offset_next_line; + } + + // mask1 loop: + for (int y = 0; y < 4; y++) + { + __m128 blockPosX = _mm_add_ps(blockPosY, gradientX); + __m128 W = _mm_shuffle_ps(blockPosY, blockPosX, _MM_SHUFFLE(0, 0, 0, 0)); + __m128 rcpW = _mm_div_ps(_mm_set1_ps((float)0x01000000), W); + __m128i posUV = _mm_cvtps_epi32(_mm_mul_ps(_mm_shuffle_ps(blockPosY, blockPosX, _MM_SHUFFLE(2, 1, 2, 1)), rcpW)); + + __m128 vis = _mm_mul_ps(_mm_shuffle_ps(blockPosY, blockPosX, _MM_SHUFFLE(3, 3, 3, 3)), W); + __m128i lightpospair = _mm_sub_epi32( + _mm_set1_epi32(FRACUNIT), + _mm_cvtps_epi32(_mm_mul_ps( + _mm_max_ps(_mm_sub_ps(_mm_set1_ps(shade), _mm_min_ps(_mm_set1_ps(24.0f / 32.0f), vis)), _mm_setzero_ps()), + _mm_set1_ps((float)FRACUNIT)))); + lightpospair = _mm_or_si128(_mm_and_si128(lightmask, lightpospair), _mm_andnot_si128(lightmask, fixedlight)); + + int32_t posU = _mm_cvtsi128_si32(posUV); + int32_t posV = _mm_cvtsi128_si32(_mm_srli_si128(posUV, 4)); + int32_t nextU = _mm_cvtsi128_si32(_mm_srli_si128(posUV, 8)); + int32_t nextV = _mm_cvtsi128_si32(_mm_srli_si128(posUV, 12)); + int32_t lightpos = _mm_cvtsi128_si32(lightpospair); + int32_t lightnext = _mm_cvtsi128_si32(_mm_srli_si128(lightpospair, 8)); + int32_t stepU = (nextU - posU) >> 3; + int32_t stepV = (nextV - posV) >> 3; + fixed_t lightstep = (lightnext - lightpos) >> 3; + + for (int x = 0; x < 2; x++) + { + // Load bgcolor + uint32_t desttmp[4]; + __m256i bgcolor; + if (BlendT::Mode != (int)BlendModes::Opaque) + { + if (mask1 & (1 << 31)) desttmp[0] = dest[0]; + if (mask1 & (1 << 30)) desttmp[1] = dest[1]; + if (mask1 & (1 << 29)) desttmp[2] = dest[2]; + if (mask1 & (1 << 28)) desttmp[3] = dest[3]; + + __m128i bgpacked = _mm_loadu_si128((__m128i*)(desttmp)); + bgcolor = _mm256_set_m128i(_mm_unpackhi_epi8(bgpacked, _mm_setzero_si128()), _mm_unpacklo_epi8(bgpacked, _mm_setzero_si128())); + } + else + bgcolor = _mm256_setzero_si256(); + + // Sample fgcolor + unsigned int ifgcolor0 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade0 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + unsigned int ifgcolor1 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade1 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + unsigned int ifgcolor2 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade2 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + unsigned int ifgcolor3 = Sample32_AVX2(posU, posV, texPixels, texWidth, texHeight, oneU, oneV, color, translation); + unsigned int ifgshade3 = SampleShade32_AVX2(posU, posV, texPixels, texWidth, texHeight, fuzzpos); + posU += stepU; + posV += stepV; + + // Setup light + int lightpos0 = lightpos >> 8; + lightpos += lightstep; + int lightpos1 = lightpos >> 8; + lightpos += lightstep; + int lightpos2 = lightpos >> 8; + lightpos += lightstep; + int lightpos3 = lightpos >> 8; + lightpos += lightstep; + __m256i mlight = _mm256_set_epi16( + 256, lightpos3, lightpos3, lightpos3, + 256, lightpos2, lightpos2, lightpos2, + 256, lightpos1, lightpos1, lightpos1, + 256, lightpos0, lightpos0, lightpos0); + + __m256i shade_fade_lit; + if (ShadeModeT::Mode == (int)ShadeMode::Advanced) + { + __m256i inv_light = _mm256_sub_epi16(_mm256_set_epi16(0, 256, 256, 256, 0, 256, 256, 256, 0, 256, 256, 256, 0, 256, 256, 256), mlight); + shade_fade_lit = _mm256_mullo_epi16(shade_fade, inv_light); + } + else + { + shade_fade_lit = _mm256_setzero_si256(); + } + + // Shade and blend + __m128i fgpacked = _mm_set_epi32(ifgcolor3, ifgcolor2, ifgcolor1, ifgcolor0); + __m128i shadepacked = _mm_set_epi32(ifgshade3, ifgshade2, ifgshade1, ifgshade0); + __m256i mifgcolor = _mm256_set_m128i(_mm_unpackhi_epi8(fgpacked, _mm_setzero_si128()), _mm_unpacklo_epi8(fgpacked, _mm_setzero_si128())); + __m256i mifgshade = _mm256_set_m128i(_mm_unpackhi_epi32(shadepacked, shadepacked), _mm_unpacklo_epi32(shadepacked, shadepacked)); + __m256i fgcolor = mifgcolor; + fgcolor = Shade32_AVX2(fgcolor, mlight, desaturate, inv_desaturate, shade_fade_lit, shade_light); + __m256i outcolor = Blend32_AVX2(fgcolor, bgcolor, mifgcolor, mifgshade, srcalpha, destalpha); + + // Store result + _mm_storeu_si128((__m128i*)desttmp, _mm_or_si128(_mm256_extracti128_si256(outcolor, 0), _mm_slli_si128(_mm256_extracti128_si256(outcolor, 1), 8))); + if (mask1 & (1 << 31)) dest[0] = desttmp[0]; + if (mask1 & (1 << 30)) dest[1] = desttmp[1]; + if (mask1 & (1 << 29)) dest[2] = desttmp[2]; + if (mask1 & (1 << 28)) dest[3] = desttmp[3]; + + mask1 <<= 4; + dest += 4; + } + + blockPosY = _mm_add_ps(blockPosY, gradientY); + + dest += offset_next_line; + } + } + } +};