From 736db284d2d98a0603620a205d90942810fa531f Mon Sep 17 00:00:00 2001 From: FernandoS27 Date: Wed, 10 Oct 2018 22:03:38 -0400 Subject: [PATCH 1/6] Implement Fast 3D Swizzle --- src/video_core/textures/decoders.cpp | 76 +++++++++++++++++++++++++++- 1 file changed, 74 insertions(+), 2 deletions(-) diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 0d2456b56..8ae6dcf9a 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -98,11 +98,83 @@ static void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_ } } +void Fast3DProcessGobs(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, const u32 x_start, + const u32 y_start, const u32 z_start, const u32 x_end, const u32 y_end, + const u32 z_end, const u32 tile_offset, const u32 xy_block_size, + const u32 layer_z, const u32 stride_x, const u32 bytes_per_pixel, + const u32 out_bytes_per_pixel) { + std::array data_ptrs; + u32 z_adress = tile_offset; + const u32 x_startb = x_start * bytes_per_pixel; + const u32 x_endb = x_end * bytes_per_pixel; + const u32 copy_size = 16; + const u32 gob_size = 64 * 8 * 1; + for (u32 z = z_start; z < z_end; z++) { + u32 y_adress = z_adress; + u32 pixel_base = layer_z * z + y_start * stride_x; + for (u32 y = y_start; y < y_end; y++) { + const auto& table = fast_swizzle_table[y % 8]; + for (u32 xb = x_startb; xb < x_endb; xb += copy_size) { + const u32 swizzle_offset{y_adress + table[(xb / 16) % 4]}; + const u32 out_x = xb * out_bytes_per_pixel / bytes_per_pixel; + const u32 pixel_index{out_x + pixel_base}; + data_ptrs[unswizzle] = swizzled_data + swizzle_offset; + data_ptrs[!unswizzle] = unswizzled_data + pixel_index; + std::memcpy(data_ptrs[0], data_ptrs[1], copy_size); + } + pixel_base += stride_x; + if ((y + 1) % 8 == 0) + y_adress += gob_size; + } + z_adress += xy_block_size; + } +} + +void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 width, + u32 height, u32 depth, u32 bytes_per_pixel, u32 out_bytes_per_pixel, + u32 block_height, u32 block_depth) { + auto div_ceil = [](u32 x, u32 y) { return ((x + y - 1) / y); }; + + const u32 stride_x = width * out_bytes_per_pixel; + const u32 layer_z = height * stride_x; + const u32 gob_x_bytes = 64; + const u32 gob_elements_x = gob_x_bytes / bytes_per_pixel; + const u32 gob_elements_y = 8; + const u32 gob_elements_z = 1; + const u32 block_x_elements = gob_elements_x; + const u32 block_y_elements = gob_elements_y * block_height; + const u32 block_z_elements = gob_elements_z * block_depth; + const u32 blocks_on_x = div_ceil(width, block_x_elements); + const u32 blocks_on_y = div_ceil(height, block_y_elements); + const u32 blocks_on_z = div_ceil(depth, block_z_elements); + const u32 blocks = blocks_on_x * blocks_on_y * blocks_on_z; + const u32 gob_size = 64 * 8 * 1; + const u32 xy_block_size = gob_size * block_height; + const u32 block_size = xy_block_size * block_depth; + u32 tile_offset = 0; + for (u32 zb = 0; zb < blocks_on_z; zb++) { + const u32 z_start = zb * block_z_elements; + const u32 z_end = std::min(depth, z_start + block_z_elements); + for (u32 yb = 0; yb < blocks_on_y; yb++) { + const u32 y_start = yb * block_y_elements; + const u32 y_end = std::min(height, y_start + block_y_elements); + for (u32 xb = 0; xb < blocks_on_x; xb++) { + const u32 x_start = xb * block_x_elements; + const u32 x_end = std::min(width, x_start + block_x_elements); + Fast3DProcessGobs(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, + z_start, x_end, y_end, z_end, tile_offset, xy_block_size, layer_z, + stride_x, bytes_per_pixel, out_bytes_per_pixel); + tile_offset += block_size; + } + } + } +} + void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height) { if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) { - FastSwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data, - unswizzled_data, unswizzle, block_height); + Fast3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, 1U, + bytes_per_pixel, out_bytes_per_pixel, block_height, 1U); } else { LegacySwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data, unswizzled_data, unswizzle, block_height); From 4d959c6bdca6a5545a6d757485ed3824dcb6660b Mon Sep 17 00:00:00 2001 From: FernandoS27 Date: Thu, 11 Oct 2018 14:33:50 -0400 Subject: [PATCH 2/6] Implement Precise 3D Swizzle --- src/video_core/textures/decoders.cpp | 74 ++++++++++++++++++++++++++-- 1 file changed, 71 insertions(+), 3 deletions(-) diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 8ae6dcf9a..d6750b174 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -98,6 +98,74 @@ static void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_ } } +void Precise3DProcessGobs(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, const u32 x_start, + const u32 y_start, const u32 z_start, const u32 x_end, const u32 y_end, + const u32 z_end, const u32 tile_offset, const u32 xy_block_size, + const u32 layer_z, const u32 stride_x, const u32 bytes_per_pixel, + const u32 out_bytes_per_pixel) { + std::array data_ptrs; + u32 z_adress = tile_offset; + const u32 gob_size = 64 * 8 * 1; + for (u32 z = z_start; z < z_end; z++) { + u32 y_adress = z_adress; + u32 pixel_base = layer_z * z + y_start * stride_x; + for (u32 y = y_start; y < y_end; y++) { + const auto& table = legacy_swizzle_table[y % 8]; + for (u32 x = x_start; x < x_end; x++) { + const u32 swizzle_offset{y_adress + table[x * bytes_per_pixel % 64]}; + const u32 pixel_index{x * out_bytes_per_pixel + pixel_base}; + data_ptrs[unswizzle] = swizzled_data + swizzle_offset; + data_ptrs[!unswizzle] = unswizzled_data + pixel_index; + std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); + } + pixel_base += stride_x; + if ((y + 1) % 8 == 0) + y_adress += gob_size; + } + z_adress += xy_block_size; + } +} + +void Precise3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 width, + u32 height, u32 depth, u32 bytes_per_pixel, u32 out_bytes_per_pixel, + u32 block_height, u32 block_depth) { + auto div_ceil = [](u32 x, u32 y) { return ((x + y - 1) / y); }; + + const u32 stride_x = width * out_bytes_per_pixel; + const u32 layer_z = height * stride_x; + const u32 gob_x_bytes = 64; + const u32 gob_elements_x = gob_x_bytes / bytes_per_pixel; + const u32 gob_elements_y = 8; + const u32 gob_elements_z = 1; + const u32 block_x_elements = gob_elements_x; + const u32 block_y_elements = gob_elements_y * block_height; + const u32 block_z_elements = gob_elements_z * block_depth; + const u32 blocks_on_x = div_ceil(width, block_x_elements); + const u32 blocks_on_y = div_ceil(height, block_y_elements); + const u32 blocks_on_z = div_ceil(depth, block_z_elements); + const u32 blocks = blocks_on_x * blocks_on_y * blocks_on_z; + const u32 gob_size = gob_x_bytes * gob_elements_y * gob_elements_z; + const u32 xy_block_size = gob_size * block_height; + const u32 block_size = xy_block_size * block_depth; + u32 tile_offset = 0; + for (u32 zb = 0; zb < blocks_on_z; zb++) { + const u32 z_start = zb * block_z_elements; + const u32 z_end = std::min(depth, z_start + block_z_elements); + for (u32 yb = 0; yb < blocks_on_y; yb++) { + const u32 y_start = yb * block_y_elements; + const u32 y_end = std::min(height, y_start + block_y_elements); + for (u32 xb = 0; xb < blocks_on_x; xb++) { + const u32 x_start = xb * block_x_elements; + const u32 x_end = std::min(width, x_start + block_x_elements); + Precise3DProcessGobs(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, + z_start, x_end, y_end, z_end, tile_offset, xy_block_size, layer_z, + stride_x, bytes_per_pixel, out_bytes_per_pixel); + tile_offset += block_size; + } + } + } +} + void Fast3DProcessGobs(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, const u32 y_end, const u32 z_end, const u32 tile_offset, const u32 xy_block_size, @@ -148,7 +216,7 @@ void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, const u32 blocks_on_y = div_ceil(height, block_y_elements); const u32 blocks_on_z = div_ceil(depth, block_z_elements); const u32 blocks = blocks_on_x * blocks_on_y * blocks_on_z; - const u32 gob_size = 64 * 8 * 1; + const u32 gob_size = gob_x_bytes * gob_elements_y * gob_elements_z; const u32 xy_block_size = gob_size * block_height; const u32 block_size = xy_block_size * block_depth; u32 tile_offset = 0; @@ -176,8 +244,8 @@ void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_ Fast3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, 1U, bytes_per_pixel, out_bytes_per_pixel, block_height, 1U); } else { - LegacySwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data, - unswizzled_data, unswizzle, block_height); + Precise3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, 1U, + bytes_per_pixel, out_bytes_per_pixel, block_height, 1U); } } From d4ae43f9c1dd1b366cf71520841d5f2f051ce69d Mon Sep 17 00:00:00 2001 From: FernandoS27 Date: Thu, 11 Oct 2018 17:08:48 -0400 Subject: [PATCH 3/6] Remove old Swizzle algorithms and use 3d Swizzle --- src/video_core/textures/decoders.cpp | 162 ++++++++++++--------------- 1 file changed, 69 insertions(+), 93 deletions(-) diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index d6750b174..5e2d3ac32 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -40,97 +40,56 @@ struct alignas(64) SwizzleTable { constexpr auto legacy_swizzle_table = SwizzleTable<8, 64, 1>(); constexpr auto fast_swizzle_table = SwizzleTable<8, 4, 16>(); -static void LegacySwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u8* swizzled_data, u8* unswizzled_data, bool unswizzle, - u32 block_height) { - std::array data_ptrs; - const std::size_t stride = width * bytes_per_pixel; - const std::size_t gobs_in_x = 64; - const std::size_t gobs_in_y = 8; - const std::size_t gobs_size = gobs_in_x * gobs_in_y; - const std::size_t image_width_in_gobs{(stride + gobs_in_x - 1) / gobs_in_x}; - for (std::size_t y = 0; y < height; ++y) { - const std::size_t gob_y_address = - (y / (gobs_in_y * block_height)) * gobs_size * block_height * image_width_in_gobs + - (y % (gobs_in_y * block_height) / gobs_in_y) * gobs_size; - const auto& table = legacy_swizzle_table[y % gobs_in_y]; - for (std::size_t x = 0; x < width; ++x) { - const std::size_t gob_address = - gob_y_address + (x * bytes_per_pixel / gobs_in_x) * gobs_size * block_height; - const std::size_t x2 = x * bytes_per_pixel; - const std::size_t swizzle_offset = gob_address + table[x2 % gobs_in_x]; - const std::size_t pixel_index = (x + y * width) * out_bytes_per_pixel; - - data_ptrs[unswizzle] = swizzled_data + swizzle_offset; - data_ptrs[!unswizzle] = unswizzled_data + pixel_index; - - std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); - } - } -} - -static void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u8* swizzled_data, u8* unswizzled_data, bool unswizzle, - u32 block_height) { - std::array data_ptrs; - const std::size_t stride{width * bytes_per_pixel}; - const std::size_t gobs_in_x = 64; - const std::size_t gobs_in_y = 8; - const std::size_t gobs_size = gobs_in_x * gobs_in_y; - const std::size_t image_width_in_gobs{(stride + gobs_in_x - 1) / gobs_in_x}; - const std::size_t copy_size{16}; - for (std::size_t y = 0; y < height; ++y) { - const std::size_t initial_gob = - (y / (gobs_in_y * block_height)) * gobs_size * block_height * image_width_in_gobs + - (y % (gobs_in_y * block_height) / gobs_in_y) * gobs_size; - const std::size_t pixel_base{y * width * out_bytes_per_pixel}; - const auto& table = fast_swizzle_table[y % gobs_in_y]; - for (std::size_t xb = 0; xb < stride; xb += copy_size) { - const std::size_t gob_address{initial_gob + - (xb / gobs_in_x) * gobs_size * block_height}; - const std::size_t swizzle_offset{gob_address + table[(xb / 16) % 4]}; - const std::size_t out_x = xb * out_bytes_per_pixel / bytes_per_pixel; - const std::size_t pixel_index{out_x + pixel_base}; - data_ptrs[unswizzle] = swizzled_data + swizzle_offset; - data_ptrs[!unswizzle] = unswizzled_data + pixel_index; - std::memcpy(data_ptrs[0], data_ptrs[1], copy_size); - } - } -} - -void Precise3DProcessGobs(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, const u32 x_start, - const u32 y_start, const u32 z_start, const u32 x_end, const u32 y_end, - const u32 z_end, const u32 tile_offset, const u32 xy_block_size, - const u32 layer_z, const u32 stride_x, const u32 bytes_per_pixel, - const u32 out_bytes_per_pixel) { +/** + * This function manages ALL the GOBs(Group of Bytes) Inside a single block. + * Instead of going gob by gob, we map the coordinates inside a block and manage from + * those. Block_Width is assumed to be 1. + */ +void Precise3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, + const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, + const u32 y_end, const u32 z_end, const u32 tile_offset, + const u32 xy_block_size, const u32 layer_z, const u32 stride_x, + const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { std::array data_ptrs; u32 z_adress = tile_offset; - const u32 gob_size = 64 * 8 * 1; + const u32 gob_size_x = 64; + const u32 gob_size_y = 8; + const u32 gob_size_z = 1; + const u32 gob_size = gob_size_x * gob_size_y * gob_size_z; for (u32 z = z_start; z < z_end; z++) { u32 y_adress = z_adress; u32 pixel_base = layer_z * z + y_start * stride_x; for (u32 y = y_start; y < y_end; y++) { - const auto& table = legacy_swizzle_table[y % 8]; + const auto& table = legacy_swizzle_table[y % gob_size_y]; for (u32 x = x_start; x < x_end; x++) { - const u32 swizzle_offset{y_adress + table[x * bytes_per_pixel % 64]}; + const u32 swizzle_offset{y_adress + table[x * bytes_per_pixel % gob_size_x]}; const u32 pixel_index{x * out_bytes_per_pixel + pixel_base}; data_ptrs[unswizzle] = swizzled_data + swizzle_offset; data_ptrs[!unswizzle] = unswizzled_data + pixel_index; std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); } pixel_base += stride_x; - if ((y + 1) % 8 == 0) + if ((y + 1) % gob_size_y == 0) y_adress += gob_size; } z_adress += xy_block_size; } } -void Precise3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 width, - u32 height, u32 depth, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u32 block_height, u32 block_depth) { - auto div_ceil = [](u32 x, u32 y) { return ((x + y - 1) / y); }; - +/** + * This function unswizzles or swizzles a texture by mapping Linear to BlockLinear Textue. + * The body of this function takes care of splitting the swizzled texture into blocks, + * and managing the extents of it. Once all the parameters of a single block are obtained, + * the function calls '3DProcessBlock' to process that particular Block. + * + * Documentation for the memory layout and decoding can be found at: + * https://envytools.readthedocs.io/en/latest/hw/memory/g80-surface.html#blocklinear-surfaces + */ +void Precise3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, + const u32 width, const u32 height, const u32 depth, + const u32 bytes_per_pixel, const u32 out_bytes_per_pixel, + const u32 block_height, const u32 block_depth) { + auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); }; const u32 stride_x = width * out_bytes_per_pixel; const u32 layer_z = height * stride_x; const u32 gob_x_bytes = 64; @@ -157,33 +116,41 @@ void Precise3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, bool unswizzl for (u32 xb = 0; xb < blocks_on_x; xb++) { const u32 x_start = xb * block_x_elements; const u32 x_end = std::min(width, x_start + block_x_elements); - Precise3DProcessGobs(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, - z_start, x_end, y_end, z_end, tile_offset, xy_block_size, layer_z, - stride_x, bytes_per_pixel, out_bytes_per_pixel); + Precise3DProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, + z_start, x_end, y_end, z_end, tile_offset, xy_block_size, + layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel); tile_offset += block_size; } } } } -void Fast3DProcessGobs(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, const u32 x_start, - const u32 y_start, const u32 z_start, const u32 x_end, const u32 y_end, - const u32 z_end, const u32 tile_offset, const u32 xy_block_size, - const u32 layer_z, const u32 stride_x, const u32 bytes_per_pixel, - const u32 out_bytes_per_pixel) { +/** + * This function manages ALL the GOBs(Group of Bytes) Inside a single block. + * Instead of going gob by gob, we map the coordinates inside a block and manage from + * those. Block_Width is assumed to be 1. + */ +void Fast3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, + const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, + const u32 y_end, const u32 z_end, const u32 tile_offset, + const u32 xy_block_size, const u32 layer_z, const u32 stride_x, + const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { std::array data_ptrs; u32 z_adress = tile_offset; const u32 x_startb = x_start * bytes_per_pixel; const u32 x_endb = x_end * bytes_per_pixel; const u32 copy_size = 16; - const u32 gob_size = 64 * 8 * 1; + const u32 gob_size_x = 64; + const u32 gob_size_y = 8; + const u32 gob_size_z = 1; + const u32 gob_size = gob_size_x * gob_size_y * gob_size_z; for (u32 z = z_start; z < z_end; z++) { u32 y_adress = z_adress; u32 pixel_base = layer_z * z + y_start * stride_x; for (u32 y = y_start; y < y_end; y++) { - const auto& table = fast_swizzle_table[y % 8]; + const auto& table = fast_swizzle_table[y % gob_size_y]; for (u32 xb = x_startb; xb < x_endb; xb += copy_size) { - const u32 swizzle_offset{y_adress + table[(xb / 16) % 4]}; + const u32 swizzle_offset{y_adress + table[(xb / copy_size) % 4]}; const u32 out_x = xb * out_bytes_per_pixel / bytes_per_pixel; const u32 pixel_index{out_x + pixel_base}; data_ptrs[unswizzle] = swizzled_data + swizzle_offset; @@ -191,18 +158,27 @@ void Fast3DProcessGobs(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, c std::memcpy(data_ptrs[0], data_ptrs[1], copy_size); } pixel_base += stride_x; - if ((y + 1) % 8 == 0) + if ((y + 1) % gob_size_y == 0) y_adress += gob_size; } z_adress += xy_block_size; } } -void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 width, - u32 height, u32 depth, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u32 block_height, u32 block_depth) { - auto div_ceil = [](u32 x, u32 y) { return ((x + y - 1) / y); }; - +/** + * This function unswizzles or swizzles a texture by mapping Linear to BlockLinear Textue. + * The body of this function takes care of splitting the swizzled texture into blocks, + * and managing the extents of it. Once all the parameters of a single block are obtained, + * the function calls '3DProcessBlock' to process that particular Block. + * + * Documentation for the memory layout and decoding can be found at: + * https://envytools.readthedocs.io/en/latest/hw/memory/g80-surface.html#blocklinear-surfaces + */ +void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, + const u32 width, const u32 height, const u32 depth, + const u32 bytes_per_pixel, const u32 out_bytes_per_pixel, + const u32 block_height, const u32 block_depth) { + auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); }; const u32 stride_x = width * out_bytes_per_pixel; const u32 layer_z = height * stride_x; const u32 gob_x_bytes = 64; @@ -229,9 +205,9 @@ void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, bool unswizzle, for (u32 xb = 0; xb < blocks_on_x; xb++) { const u32 x_start = xb * block_x_elements; const u32 x_end = std::min(width, x_start + block_x_elements); - Fast3DProcessGobs(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, - z_start, x_end, y_end, z_end, tile_offset, xy_block_size, layer_z, - stride_x, bytes_per_pixel, out_bytes_per_pixel); + Fast3DProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, + z_start, x_end, y_end, z_end, tile_offset, xy_block_size, + layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel); tile_offset += block_size; } } @@ -245,7 +221,7 @@ void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_ bytes_per_pixel, out_bytes_per_pixel, block_height, 1U); } else { Precise3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, 1U, - bytes_per_pixel, out_bytes_per_pixel, block_height, 1U); + bytes_per_pixel, out_bytes_per_pixel, block_height, 1U); } } From e0ca938b222584cdbf9e03460171c03484882bd4 Mon Sep 17 00:00:00 2001 From: FernandoS27 Date: Thu, 11 Oct 2018 19:11:47 -0400 Subject: [PATCH 4/6] Propagate depth and depth_block on modules using decoders --- src/video_core/engines/fermi_2d.cpp | 14 ++++--- src/video_core/engines/maxwell_dma.cpp | 10 +++-- src/video_core/engines/maxwell_dma.h | 4 ++ .../renderer_opengl/gl_rasterizer_cache.cpp | 32 ++++++++------- src/video_core/textures/decoders.cpp | 40 ++++++++++--------- src/video_core/textures/decoders.h | 15 +++---- src/video_core/textures/texture.h | 1 + .../debugger/graphics/graphics_surface.cpp | 5 ++- 8 files changed, 67 insertions(+), 54 deletions(-) diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp index 912e785b9..597b279b9 100644 --- a/src/video_core/engines/fermi_2d.cpp +++ b/src/video_core/engines/fermi_2d.cpp @@ -62,14 +62,16 @@ void Fermi2D::HandleSurfaceCopy() { u8* dst_buffer = Memory::GetPointer(dest_cpu); if (!regs.src.linear && regs.dst.linear) { // If the input is tiled and the output is linear, deswizzle the input and copy it over. - Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel, - dst_bytes_per_pixel, src_buffer, dst_buffer, true, - regs.src.BlockHeight()); + Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth, + src_bytes_per_pixel, dst_bytes_per_pixel, src_buffer, + dst_buffer, true, regs.src.BlockHeight(), + regs.src.BlockDepth()); } else { // If the input is linear and the output is tiled, swizzle the input and copy it over. - Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel, - dst_bytes_per_pixel, dst_buffer, src_buffer, false, - regs.dst.BlockHeight()); + Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth, + src_bytes_per_pixel, dst_bytes_per_pixel, dst_buffer, + src_buffer, false, regs.dst.BlockHeight(), + regs.dst.BlockDepth()); } } } diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index aa7481b8c..bf2a21bb6 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -68,12 +68,14 @@ void MaxwellDMA::HandleCopy() { if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) { // If the input is tiled and the output is linear, deswizzle the input and copy it over. - Texture::CopySwizzledData(regs.src_params.size_x, regs.src_params.size_y, 1, 1, src_buffer, - dst_buffer, true, regs.src_params.BlockHeight()); + Texture::CopySwizzledData(regs.src_params.size_x, regs.src_params.size_y, + regs.src_params.size_z, 1, 1, src_buffer, dst_buffer, true, + regs.src_params.BlockHeight(), regs.src_params.BlockDepth()); } else { // If the input is linear and the output is tiled, swizzle the input and copy it over. - Texture::CopySwizzledData(regs.dst_params.size_x, regs.dst_params.size_y, 1, 1, dst_buffer, - src_buffer, false, regs.dst_params.BlockHeight()); + Texture::CopySwizzledData(regs.dst_params.size_x, regs.dst_params.size_y, + regs.dst_params.size_z, 1, 1, dst_buffer, src_buffer, false, + regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth()); } } diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index 311ccb616..df19e02e2 100644 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -43,6 +43,10 @@ public: u32 BlockHeight() const { return 1 << block_height; } + + u32 BlockDepth() const { + return 1 << block_depth; + } }; static_assert(sizeof(Parameters) == 24, "Parameters has wrong size"); diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index 65a220c41..86cb62dae 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -323,8 +323,8 @@ static bool IsFormatBCn(PixelFormat format) { } template -void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::size_t gl_buffer_size, - VAddr addr) { +void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth, u8* gl_buffer, + std::size_t gl_buffer_size, VAddr addr) { constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / CHAR_BIT; constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format); @@ -333,7 +333,7 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::si // pixel values. const u32 tile_size{IsFormatBCn(format) ? 4U : 1U}; const std::vector data = Tegra::Texture::UnswizzleTexture( - addr, tile_size, bytes_per_pixel, stride, height, block_height); + addr, tile_size, bytes_per_pixel, stride, height, depth, block_height, block_depth); const std::size_t size_to_copy{std::min(gl_buffer_size, data.size())}; memcpy(gl_buffer, data.data(), size_to_copy); } else { @@ -345,7 +345,7 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::si } } -static constexpr std::array morton_to_gl_fns = { // clang-format off @@ -403,7 +403,7 @@ static constexpr std::array gl_to_morton_fns = { // clang-format off @@ -827,25 +827,27 @@ void CachedSurface::LoadGLBuffer() { if (params.is_tiled) { gl_buffer.resize(total_size); + u32 depth = params.depth; + u32 block_depth = params.block_depth; ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}", params.block_width, static_cast(params.target)); - ASSERT_MSG(params.block_depth == 1, "Block depth is defined as {} on texture type {}", - params.block_depth, static_cast(params.target)); - // TODO(bunnei): This only unswizzles and copies a 2D texture - we do not yet know how to do - // this for 3D textures, etc. switch (params.target) { case SurfaceParams::SurfaceTarget::Texture2D: - // Pass impl. to the fallback code below + // TODO(Blinkhawk): Eliminate this condition once all texture types are implemented. + depth = 1U; + block_depth = 1U; break; case SurfaceParams::SurfaceTarget::Texture2DArray: case SurfaceParams::SurfaceTarget::TextureCubemap: + depth = 1U; + block_depth = 1U; for (std::size_t index = 0; index < params.depth; ++index) { const std::size_t offset{index * copy_size}; morton_to_gl_fns[static_cast(params.pixel_format)]( - params.width, params.block_height, params.height, gl_buffer.data() + offset, - copy_size, params.addr + offset); + params.width, params.block_height, params.height, 1U, 1U, + gl_buffer.data() + offset, copy_size, params.addr + offset); } break; default: @@ -854,9 +856,11 @@ void CachedSurface::LoadGLBuffer() { UNREACHABLE(); } + const std::size_t size = copy_size * depth; + morton_to_gl_fns[static_cast(params.pixel_format)]( - params.width, params.block_height, params.height, gl_buffer.data(), copy_size, - params.addr); + params.width, params.block_height, params.height, block_depth, depth, gl_buffer.data(), + size, params.addr); } else { const u8* const texture_src_data_end{texture_src_data + total_size}; gl_buffer.assign(texture_src_data, texture_src_data_end); diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 5e2d3ac32..6693067f8 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -51,18 +51,18 @@ void Precise3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool un const u32 xy_block_size, const u32 layer_z, const u32 stride_x, const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { std::array data_ptrs; - u32 z_adress = tile_offset; + u32 z_address = tile_offset; const u32 gob_size_x = 64; const u32 gob_size_y = 8; const u32 gob_size_z = 1; const u32 gob_size = gob_size_x * gob_size_y * gob_size_z; for (u32 z = z_start; z < z_end; z++) { - u32 y_adress = z_adress; + u32 y_address = z_address; u32 pixel_base = layer_z * z + y_start * stride_x; for (u32 y = y_start; y < y_end; y++) { const auto& table = legacy_swizzle_table[y % gob_size_y]; for (u32 x = x_start; x < x_end; x++) { - const u32 swizzle_offset{y_adress + table[x * bytes_per_pixel % gob_size_x]}; + const u32 swizzle_offset{y_address + table[x * bytes_per_pixel % gob_size_x]}; const u32 pixel_index{x * out_bytes_per_pixel + pixel_base}; data_ptrs[unswizzle] = swizzled_data + swizzle_offset; data_ptrs[!unswizzle] = unswizzled_data + pixel_index; @@ -70,9 +70,9 @@ void Precise3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool un } pixel_base += stride_x; if ((y + 1) % gob_size_y == 0) - y_adress += gob_size; + y_address += gob_size; } - z_adress += xy_block_size; + z_address += xy_block_size; } } @@ -136,7 +136,7 @@ void Fast3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswi const u32 xy_block_size, const u32 layer_z, const u32 stride_x, const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { std::array data_ptrs; - u32 z_adress = tile_offset; + u32 z_address = tile_offset; const u32 x_startb = x_start * bytes_per_pixel; const u32 x_endb = x_end * bytes_per_pixel; const u32 copy_size = 16; @@ -145,12 +145,12 @@ void Fast3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswi const u32 gob_size_z = 1; const u32 gob_size = gob_size_x * gob_size_y * gob_size_z; for (u32 z = z_start; z < z_end; z++) { - u32 y_adress = z_adress; + u32 y_address = z_address; u32 pixel_base = layer_z * z + y_start * stride_x; for (u32 y = y_start; y < y_end; y++) { const auto& table = fast_swizzle_table[y % gob_size_y]; for (u32 xb = x_startb; xb < x_endb; xb += copy_size) { - const u32 swizzle_offset{y_adress + table[(xb / copy_size) % 4]}; + const u32 swizzle_offset{y_address + table[(xb / copy_size) % 4]}; const u32 out_x = xb * out_bytes_per_pixel / bytes_per_pixel; const u32 pixel_index{out_x + pixel_base}; data_ptrs[unswizzle] = swizzled_data + swizzle_offset; @@ -159,9 +159,9 @@ void Fast3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswi } pixel_base += stride_x; if ((y + 1) % gob_size_y == 0) - y_adress += gob_size; + y_address += gob_size; } - z_adress += xy_block_size; + z_address += xy_block_size; } } @@ -214,14 +214,15 @@ void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswi } } -void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height) { +void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel, + u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, + bool unswizzle, u32 block_height, u32 block_depth) { if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) { - Fast3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, 1U, - bytes_per_pixel, out_bytes_per_pixel, block_height, 1U); + Fast3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, depth, + bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth); } else { - Precise3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, 1U, - bytes_per_pixel, out_bytes_per_pixel, block_height, 1U); + Precise3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, depth, + bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth); } } @@ -269,10 +270,11 @@ u32 BytesPerPixel(TextureFormat format) { } std::vector UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width, - u32 height, u32 block_height) { + u32 height, u32 depth, u32 block_height, u32 block_depth) { std::vector unswizzled_data(width * height * bytes_per_pixel); - CopySwizzledData(width / tile_size, height / tile_size, bytes_per_pixel, bytes_per_pixel, - Memory::GetPointer(address), unswizzled_data.data(), true, block_height); + CopySwizzledData(width / tile_size, height / tile_size, depth, bytes_per_pixel, bytes_per_pixel, + Memory::GetPointer(address), unswizzled_data.data(), true, block_height, + block_depth); return unswizzled_data; } diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h index 234d250af..aaf316947 100644 --- a/src/video_core/textures/decoders.h +++ b/src/video_core/textures/decoders.h @@ -14,17 +14,14 @@ namespace Tegra::Texture { * Unswizzles a swizzled texture without changing its format. */ std::vector UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width, - u32 height, u32 block_height = TICEntry::DefaultBlockHeight); - -/** - * Unswizzles a swizzled depth texture without changing its format. - */ -std::vector UnswizzleDepthTexture(VAddr address, DepthFormat format, u32 width, u32 height, - u32 block_height = TICEntry::DefaultBlockHeight); + u32 height, u32 depth, + u32 block_height = TICEntry::DefaultBlockHeight, + u32 block_depth = TICEntry::DefaultBlockHeight); /// Copies texture data from a buffer and performs swizzling/unswizzling as necessary. -void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height); +void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel, + u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, + bool unswizzle, u32 block_height, u32 block_depth); /** * Decodes an unswizzled texture into a A8R8G8B8 texture. diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h index 58d17abcb..5947bd2b9 100644 --- a/src/video_core/textures/texture.h +++ b/src/video_core/textures/texture.h @@ -141,6 +141,7 @@ static_assert(sizeof(TextureHandle) == 4, "TextureHandle has wrong size"); struct TICEntry { static constexpr u32 DefaultBlockHeight = 16; + static constexpr u32 DefaultBlockDepth = 1; union { u32 raw; diff --git a/src/yuzu/debugger/graphics/graphics_surface.cpp b/src/yuzu/debugger/graphics/graphics_surface.cpp index cbcd5dd5f..44d423da2 100644 --- a/src/yuzu/debugger/graphics/graphics_surface.cpp +++ b/src/yuzu/debugger/graphics/graphics_surface.cpp @@ -386,8 +386,9 @@ void GraphicsSurfaceWidget::OnUpdate() { // TODO(bunnei): Will not work with BCn formats that swizzle 4x4 tiles. // Needs to be fixed if we plan to use this feature more, otherwise we may remove it. - auto unswizzled_data = Tegra::Texture::UnswizzleTexture( - *address, 1, Tegra::Texture::BytesPerPixel(surface_format), surface_width, surface_height); + auto unswizzled_data = + Tegra::Texture::UnswizzleTexture(*address, 1, Tegra::Texture::BytesPerPixel(surface_format), + surface_width, surface_height, 1U); auto texture_data = Tegra::Texture::DecodeTexture(unswizzled_data, surface_format, surface_width, surface_height); From 1ff20d8538e7ed6a9121882afef4d73503667842 Mon Sep 17 00:00:00 2001 From: FernandoS27 Date: Sat, 13 Oct 2018 16:11:11 -0400 Subject: [PATCH 5/6] Fix a Crash on Zelda BotW and Splatoon 2, and simplified LoadGLBuffer --- .../renderer_opengl/gl_rasterizer_cache.cpp | 19 +------------------ src/video_core/textures/decoders.cpp | 2 +- 2 files changed, 2 insertions(+), 19 deletions(-) diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index 86cb62dae..8fdb59b5b 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -833,27 +833,10 @@ void CachedSurface::LoadGLBuffer() { ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}", params.block_width, static_cast(params.target)); - switch (params.target) { - case SurfaceParams::SurfaceTarget::Texture2D: + if (params.target == SurfaceParams::SurfaceTarget::Texture2D) { // TODO(Blinkhawk): Eliminate this condition once all texture types are implemented. depth = 1U; block_depth = 1U; - break; - case SurfaceParams::SurfaceTarget::Texture2DArray: - case SurfaceParams::SurfaceTarget::TextureCubemap: - depth = 1U; - block_depth = 1U; - for (std::size_t index = 0; index < params.depth; ++index) { - const std::size_t offset{index * copy_size}; - morton_to_gl_fns[static_cast(params.pixel_format)]( - params.width, params.block_height, params.height, 1U, 1U, - gl_buffer.data() + offset, copy_size, params.addr + offset); - } - break; - default: - LOG_CRITICAL(HW_GPU, "Unimplemented tiled load for target={}", - static_cast(params.target)); - UNREACHABLE(); } const std::size_t size = copy_size * depth; diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 6693067f8..5dab47886 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -271,7 +271,7 @@ u32 BytesPerPixel(TextureFormat format) { std::vector UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth) { - std::vector unswizzled_data(width * height * bytes_per_pixel); + std::vector unswizzled_data(width * height * depth * bytes_per_pixel); CopySwizzledData(width / tile_size, height / tile_size, depth, bytes_per_pixel, bytes_per_pixel, Memory::GetPointer(address), unswizzled_data.data(), true, block_height, block_depth); From 331ce2942c4906945b4d42f1ebe8b9b6e453c6ee Mon Sep 17 00:00:00 2001 From: FernandoS27 Date: Sat, 13 Oct 2018 20:42:38 -0400 Subject: [PATCH 6/6] Shorten the implementation of 3D swizzle to only 3 functions --- src/video_core/textures/decoders.cpp | 97 ++++++++-------------------- 1 file changed, 27 insertions(+), 70 deletions(-) diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 5dab47886..0fe91622e 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -45,11 +45,11 @@ constexpr auto fast_swizzle_table = SwizzleTable<8, 4, 16>(); * Instead of going gob by gob, we map the coordinates inside a block and manage from * those. Block_Width is assumed to be 1. */ -void Precise3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, - const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, - const u32 y_end, const u32 z_end, const u32 tile_offset, - const u32 xy_block_size, const u32 layer_z, const u32 stride_x, - const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { +void PreciseProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, + const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, + const u32 y_end, const u32 z_end, const u32 tile_offset, + const u32 xy_block_size, const u32 layer_z, const u32 stride_x, + const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { std::array data_ptrs; u32 z_address = tile_offset; const u32 gob_size_x = 64; @@ -76,65 +76,16 @@ void Precise3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool un } } -/** - * This function unswizzles or swizzles a texture by mapping Linear to BlockLinear Textue. - * The body of this function takes care of splitting the swizzled texture into blocks, - * and managing the extents of it. Once all the parameters of a single block are obtained, - * the function calls '3DProcessBlock' to process that particular Block. - * - * Documentation for the memory layout and decoding can be found at: - * https://envytools.readthedocs.io/en/latest/hw/memory/g80-surface.html#blocklinear-surfaces - */ -void Precise3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, - const u32 width, const u32 height, const u32 depth, - const u32 bytes_per_pixel, const u32 out_bytes_per_pixel, - const u32 block_height, const u32 block_depth) { - auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); }; - const u32 stride_x = width * out_bytes_per_pixel; - const u32 layer_z = height * stride_x; - const u32 gob_x_bytes = 64; - const u32 gob_elements_x = gob_x_bytes / bytes_per_pixel; - const u32 gob_elements_y = 8; - const u32 gob_elements_z = 1; - const u32 block_x_elements = gob_elements_x; - const u32 block_y_elements = gob_elements_y * block_height; - const u32 block_z_elements = gob_elements_z * block_depth; - const u32 blocks_on_x = div_ceil(width, block_x_elements); - const u32 blocks_on_y = div_ceil(height, block_y_elements); - const u32 blocks_on_z = div_ceil(depth, block_z_elements); - const u32 blocks = blocks_on_x * blocks_on_y * blocks_on_z; - const u32 gob_size = gob_x_bytes * gob_elements_y * gob_elements_z; - const u32 xy_block_size = gob_size * block_height; - const u32 block_size = xy_block_size * block_depth; - u32 tile_offset = 0; - for (u32 zb = 0; zb < blocks_on_z; zb++) { - const u32 z_start = zb * block_z_elements; - const u32 z_end = std::min(depth, z_start + block_z_elements); - for (u32 yb = 0; yb < blocks_on_y; yb++) { - const u32 y_start = yb * block_y_elements; - const u32 y_end = std::min(height, y_start + block_y_elements); - for (u32 xb = 0; xb < blocks_on_x; xb++) { - const u32 x_start = xb * block_x_elements; - const u32 x_end = std::min(width, x_start + block_x_elements); - Precise3DProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, - z_start, x_end, y_end, z_end, tile_offset, xy_block_size, - layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel); - tile_offset += block_size; - } - } - } -} - /** * This function manages ALL the GOBs(Group of Bytes) Inside a single block. * Instead of going gob by gob, we map the coordinates inside a block and manage from * those. Block_Width is assumed to be 1. */ -void Fast3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, - const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, - const u32 y_end, const u32 z_end, const u32 tile_offset, - const u32 xy_block_size, const u32 layer_z, const u32 stride_x, - const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { +void FastProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, + const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, + const u32 y_end, const u32 z_end, const u32 tile_offset, + const u32 xy_block_size, const u32 layer_z, const u32 stride_x, + const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { std::array data_ptrs; u32 z_address = tile_offset; const u32 x_startb = x_start * bytes_per_pixel; @@ -169,15 +120,15 @@ void Fast3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswi * This function unswizzles or swizzles a texture by mapping Linear to BlockLinear Textue. * The body of this function takes care of splitting the swizzled texture into blocks, * and managing the extents of it. Once all the parameters of a single block are obtained, - * the function calls '3DProcessBlock' to process that particular Block. + * the function calls 'ProcessBlock' to process that particular Block. * * Documentation for the memory layout and decoding can be found at: * https://envytools.readthedocs.io/en/latest/hw/memory/g80-surface.html#blocklinear-surfaces */ -void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, - const u32 width, const u32 height, const u32 depth, - const u32 bytes_per_pixel, const u32 out_bytes_per_pixel, - const u32 block_height, const u32 block_depth) { +template +void SwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, const u32 width, + const u32 height, const u32 depth, const u32 bytes_per_pixel, + const u32 out_bytes_per_pixel, const u32 block_height, const u32 block_depth) { auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); }; const u32 stride_x = width * out_bytes_per_pixel; const u32 layer_z = height * stride_x; @@ -205,9 +156,15 @@ void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswi for (u32 xb = 0; xb < blocks_on_x; xb++) { const u32 x_start = xb * block_x_elements; const u32 x_end = std::min(width, x_start + block_x_elements); - Fast3DProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, - z_start, x_end, y_end, z_end, tile_offset, xy_block_size, - layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel); + if (fast) { + FastProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, + z_start, x_end, y_end, z_end, tile_offset, xy_block_size, + layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel); + } else { + PreciseProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, + z_start, x_end, y_end, z_end, tile_offset, xy_block_size, + layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel); + } tile_offset += block_size; } } @@ -218,11 +175,11 @@ void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel, u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height, u32 block_depth) { if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) { - Fast3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, depth, + SwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, depth, bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth); } else { - Precise3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, depth, - bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth); + SwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, depth, + bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth); } }