astc: Enable parallel CPU astc decoding

Given the issues with GPU accelerated ASTC decoding with NVIDIA's latest drivers, parallelize astc decoding on the CPU.
Uses half the available threads in the system for astc decoding.

src/video_core/textures/astc.cpp

@@ -13,7 +13,9 @@
 
 #include <boost/container/static_vector.hpp>
 
+#include "common/alignment.h"
 #include "common/common_types.h"
+#include "common/thread_worker.h"
 #include "video_core/textures/astc.h"
 
 class InputBitStream {
@@ -1650,29 +1652,41 @@ static void DecompressBlock(std::span<const u8, 16> inBuf, const u32 blockWidth,
 
 void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
                 uint32_t block_width, uint32_t block_height, std::span<uint8_t> output) {
-    u32 block_index = 0;
+    const u32 rows = Common::DivideUp(height, block_height);
-    std::size_t depth_offset = 0;
+    const u32 cols = Common::DivideUp(width, block_width);
-    for (u32 z = 0; z < depth; z++) {
+
-        for (u32 y = 0; y < height; y += block_height) {
+    Common::ThreadWorker workers{std::max(std::thread::hardware_concurrency(), 2U) / 2,
-            for (u32 x = 0; x < width; x += block_width) {
+                                 "yuzu:ASTCDecompress"};
-                const std::span<const u8, 16> blockPtr{data.subspan(block_index * 16, 16)};
+
-
+    for (u32 z = 0; z < depth; ++z) {
-                // Blocks can be at most 12x12
+        const u32 depth_offset = z * height * width * 4;
-                std::array<u32, 12 * 12> uncompData;
+        for (u32 y_index = 0; y_index < rows; ++y_index) {
-                DecompressBlock(blockPtr, block_width, block_height, uncompData);
+            auto decompress_stride = [data, width, height, depth, block_width, block_height, output,
-
+                                      rows, cols, z, depth_offset, y_index] {
-                u32 decompWidth = std::min(block_width, width - x);
+                const u32 y = y_index * block_height;
-                u32 decompHeight = std::min(block_height, height - y);
+                for (u32 x_index = 0; x_index < cols; ++x_index) {
-
+                    const u32 block_index = (z * rows * cols) + (y_index * cols) + x_index;
-                const std::span<u8> outRow = output.subspan(depth_offset + (y * width + x) * 4);
+                    const u32 x = x_index * block_width;
-                for (u32 jj = 0; jj < decompHeight; jj++) {
+
-                    std::memcpy(outRow.data() + jj * width * 4,
+                    const std::span<const u8, 16> blockPtr{data.subspan(block_index * 16, 16)};
-                                uncompData.data() + jj * block_width, decompWidth * 4);
+
+                    // Blocks can be at most 12x12
+                    std::array<u32, 12 * 12> uncompData;
+                    DecompressBlock(blockPtr, block_width, block_height, uncompData);
+
+                    u32 decompWidth = std::min(block_width, width - x);
+                    u32 decompHeight = std::min(block_height, height - y);
+
+                    const std::span<u8> outRow = output.subspan(depth_offset + (y * width + x) * 4);
+                    for (u32 h = 0; h < decompHeight; ++h) {
+                        std::memcpy(outRow.data() + h * width * 4,
+                                    uncompData.data() + h * block_width, decompWidth * 4);
+                    }
                 }
-                ++block_index;
+            };
-            }
+            workers.QueueWork(std::move(decompress_stride));
         }
-        depth_offset += height * width * 4;
+        workers.WaitForRequests();
     }
 }