summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorbunnei <bunneidev@gmail.com>2022-09-19 21:17:51 +0200
committerGitHub <noreply@github.com>2022-09-19 21:17:51 +0200
commit8d4458ef24e473e57b9931d7a9d1442b51fb0b1a (patch)
treeeacbdc37ea2879612238e15cf3c47c82f297d756
parentMerge pull request #8915 from vonchenplus/opus_multi_stream (diff)
parentastc: Enable parallel CPU astc decoding (diff)
downloadyuzu-8d4458ef24e473e57b9931d7a9d1442b51fb0b1a.tar
yuzu-8d4458ef24e473e57b9931d7a9d1442b51fb0b1a.tar.gz
yuzu-8d4458ef24e473e57b9931d7a9d1442b51fb0b1a.tar.bz2
yuzu-8d4458ef24e473e57b9931d7a9d1442b51fb0b1a.tar.lz
yuzu-8d4458ef24e473e57b9931d7a9d1442b51fb0b1a.tar.xz
yuzu-8d4458ef24e473e57b9931d7a9d1442b51fb0b1a.tar.zst
yuzu-8d4458ef24e473e57b9931d7a9d1442b51fb0b1a.zip
Diffstat (limited to '')
-rw-r--r--src/video_core/textures/astc.cpp56
1 files changed, 35 insertions, 21 deletions
diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp
index e3f3d3c5d..b159494c5 100644
--- a/src/video_core/textures/astc.cpp
+++ b/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;
- std::size_t depth_offset = 0;
- for (u32 z = 0; z < depth; z++) {
- for (u32 y = 0; y < height; y += block_height) {
- for (u32 x = 0; x < width; x += block_width) {
- const std::span<const u8, 16> blockPtr{data.subspan(block_index * 16, 16)};
-
- // 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 jj = 0; jj < decompHeight; jj++) {
- std::memcpy(outRow.data() + jj * width * 4,
- uncompData.data() + jj * block_width, decompWidth * 4);
+ const u32 rows = Common::DivideUp(height, block_height);
+ const u32 cols = Common::DivideUp(width, block_width);
+
+ Common::ThreadWorker workers{std::max(std::thread::hardware_concurrency(), 2U) / 2,
+ "yuzu:ASTCDecompress"};
+
+ for (u32 z = 0; z < depth; ++z) {
+ const u32 depth_offset = z * height * width * 4;
+ for (u32 y_index = 0; y_index < rows; ++y_index) {
+ auto decompress_stride = [data, width, height, depth, block_width, block_height, output,
+ rows, cols, z, depth_offset, y_index] {
+ const u32 y = y_index * block_height;
+ for (u32 x_index = 0; x_index < cols; ++x_index) {
+ const u32 block_index = (z * rows * cols) + (y_index * cols) + x_index;
+ const u32 x = x_index * block_width;
+
+ const std::span<const u8, 16> blockPtr{data.subspan(block_index * 16, 16)};
+
+ // 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();
}
}