3 files changed, 179 insertions, 69 deletions

diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index 3d5476e5d..18ab723f7 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -4,6 +4,7 @@
 
 #include <cmath>
 #include <cstring>
+#include "common/alignment.h"
 #include "common/assert.h"
 #include "core/memory.h"
 #include "video_core/gpu.h"
@@ -39,72 +40,146 @@ 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) {
+/**
+ * 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 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<u8*, 2> 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);
+    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_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_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;
+                std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel);
+            }
+            pixel_base += stride_x;
+            if ((y + 1) % gob_size_y == 0)
+                y_address += gob_size;
         }
+        z_address += xy_block_size;
     }
 }
 
-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) {
+/**
+ * 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 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<u8*, 2> 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);
+    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;
+    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_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_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;
+                data_ptrs[!unswizzle] = unswizzled_data + pixel_index;
+                std::memcpy(data_ptrs[0], data_ptrs[1], copy_size);
+            }
+            pixel_base += stride_x;
+            if ((y + 1) % gob_size_y == 0)
+                y_address += gob_size;
         }
+        z_address += xy_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) {
+/**
+ * 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 '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
+ */
+template <bool fast>
+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;
+    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);
+                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;
+            }
+        }
+    }
+}
+
+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) {
-        FastSwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data,
-                        unswizzled_data, unswizzle, block_height);
+        SwizzledData<true>(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
+                           bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth);
     } else {
-        LegacySwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data,
-                          unswizzled_data, unswizzle, block_height);
+        SwizzledData<false>(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
+                            bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth);
     }
 }
 
@@ -125,7 +200,9 @@ u32 BytesPerPixel(TextureFormat format) {
     case TextureFormat::R32_G32_B32:
         return 12;
     case TextureFormat::ASTC_2D_4X4:
+    case TextureFormat::ASTC_2D_5X4:
     case TextureFormat::ASTC_2D_8X8:
+    case TextureFormat::ASTC_2D_8X5:
     case TextureFormat::A8R8G8B8:
     case TextureFormat::A2B10G10R10:
     case TextureFormat::BF10GF11RF11:
@@ -152,10 +229,11 @@ u32 BytesPerPixel(TextureFormat format) {
 }
 
 std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width,
-                                 u32 height, u32 block_height) {
-    std::vector<u8> 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);
+                                 u32 height, u32 depth, u32 block_height, u32 block_depth) {
+    std::vector<u8> 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);
     return unswizzled_data;
 }
 
@@ -199,4 +277,19 @@ std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat
     return rgba_data;
 }
 
+std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height, u32 depth,
+                          u32 block_height, u32 block_depth) {
+    if (tiled) {
+        const u32 gobs_in_x = 64 / bytes_per_pixel;
+        const u32 gobs_in_y = 8;
+        const u32 gobs_in_z = 1;
+        const u32 aligned_width = Common::AlignUp(width, gobs_in_x);
+        const u32 aligned_height = Common::AlignUp(height, gobs_in_y * block_height);
+        const u32 aligned_depth = Common::AlignUp(depth, gobs_in_z * block_depth);
+        return aligned_width * aligned_height * aligned_depth * bytes_per_pixel;
+    } else {
+        return width * height * depth * bytes_per_pixel;
+    }
+}
+
 } // namespace Tegra::Texture
diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h
index 1f7b731be..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<u8> 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<u8> 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.
@@ -32,4 +29,10 @@ void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_
 std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat format, u32 width,
                               u32 height);
 
+/**
+ * This function calculates the correct size of a texture depending if it's tiled or not.
+ */
+std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height, u32 depth,
+                          u32 block_height, u32 block_depth);
+
 } // namespace Tegra::Texture
diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h
index 8f31d825a..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;
@@ -161,7 +162,9 @@ struct TICEntry {
         BitField<21, 3, TICHeaderVersion> header_version;
     };
     union {
+        BitField<0, 3, u32> block_width;
         BitField<3, 3, u32> block_height;
+        BitField<6, 3, u32> block_depth;
 
         // High 16 bits of the pitch value
         BitField<0, 16, u32> pitch_high;
@@ -202,13 +205,24 @@ struct TICEntry {
         return depth_minus_1 + 1;
     }
 
+    u32 BlockWidth() const {
+        ASSERT(IsTiled());
+        // The block height is stored in log2 format.
+        return 1 << block_width;
+    }
+
     u32 BlockHeight() const {
-        ASSERT(header_version == TICHeaderVersion::BlockLinear ||
-               header_version == TICHeaderVersion::BlockLinearColorKey);
+        ASSERT(IsTiled());
         // The block height is stored in log2 format.
         return 1 << block_height;
     }
 
+    u32 BlockDepth() const {
+        ASSERT(IsTiled());
+        // The block height is stored in log2 format.
+        return 1 << block_depth;
+    }
+
     bool IsTiled() const {
         return header_version == TICHeaderVersion::BlockLinear ||
                header_version == TICHeaderVersion::BlockLinearColorKey;