fix etc2/bc7

1 month ago · d44ce7b93b
4 changed files with 57 additions and 79 deletions
--- a/src/qt_common/config/shared_translation.cpp
+++ b/src/qt_common/config/shared_translation.cpp
@ -482,8 +482,10 @@ std::unique_ptr<ComboboxTranslationMap> ComboboxEnumeration(QObject* parent)
        {Settings::EnumMetadata<Settings::AstcRecompression>::Index(),
         {
             PAIR(AstcRecompression, Uncompressed, tr("Uncompressed (Best quality)")),
-             PAIR(AstcRecompression, Bc1, tr("BC1 (Low quality)")),
-             PAIR(AstcRecompression, Bc3, tr("BC3 (Medium quality)")),
+             PAIR(AstcRecompression, Bc1, tr("BC1")),
+             PAIR(AstcRecompression, Bc3, tr("BC3")),
+             PAIR(AstcRecompression, Bc7, tr("BC7")),
+             PAIR(AstcRecompression, Etc2, tr("ETC2")),
         }});
    translations->insert({Settings::EnumMetadata<Settings::VramUsageMode>::Index(),
                          {
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@ -437,8 +437,7 @@ void DecodeIntegerSequence(uint max_range, uint num_values) {
    EncodingData val = EncodingData(encoding_values[max_range]);
    const uint encoding = Encoding(val);
    const uint num_bits = NumBits(val);
-    uint vals_decoded = 0;
-    while (vals_decoded < num_values && !result_limit_reached) {
+    for (uint vals_decoded = 0; vals_decoded < num_values && !result_limit_reached; ) {
        switch (encoding) {
        case QUINT:
            DecodeQuintBlock(num_bits);
@ -571,10 +570,8 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, o

 ivec2 BitTransferSigned(int a, int b) {
    ivec2 transferred;
-    transferred.y = b >> 1;
-    transferred.y |= a & 0x80;
-    transferred.x = a >> 1;
-    transferred.x &= 0x3F;
+    transferred.x = (a >> 1) & 0x3F;
+    transferred.y = (b >> 1) | (a & 0x80);
    if ((transferred.x & 0x20) > 0) {
        transferred.x -= 0x40;
    }
@ -831,38 +828,6 @@ uvec4 GetUnquantizedWeightVector(uint t, uint s, uvec2 size, uint plane_index, b
    return weight_vec;
 }

-int FindLayout(uint mode) {
-    if ((mode & 3) != 0) {
-        if ((mode & 8) != 0) {
-            if ((mode & 4) != 0) {
-                if ((mode & 0x100) != 0) {
-                    return 4;
-                }
-                return 3;
-            }
-            return 2;
-        }
-        if ((mode & 4) != 0) {
-            return 1;
-        }
-        return 0;
-    }
-    if ((mode & 0x100) != 0) {
-        if ((mode & 0x80) != 0) {
-            if ((mode & 0x20) != 0) {
-                return 8;
-            }
-            return 7;
-        }
-        return 9;
-    }
-    if ((mode & 0x80) != 0) {
-        return 6;
-    }
-    return 5;
-}
-
-
 void FillError(ivec3 coord) {
    for (uint j = 0; j < block_dims.y; j++) {
        for (uint i = 0; i < block_dims.x; i++) {
@ -908,9 +873,40 @@ bool IsError(uint mode) {
    return false;
 }

-uvec2 DecodeBlockSize(uint mode) {
+int FindLayout(uint mode) {
+    if ((mode & 3) != 0) {
+        if ((mode & 8) != 0) {
+            if ((mode & 4) != 0) {
+                if ((mode & 0x100) != 0) {
+                    return 4;
+                }
+                return 3;
+            }
+            return 2;
+        }
+        if ((mode & 4) != 0) {
+            return 1;
+        }
+        return 0;
+    }
+    if ((mode & 0x100) != 0) {
+        if ((mode & 0x80) != 0) {
+            if ((mode & 0x20) != 0) {
+                return 8;
+            }
+            return 7;
+        }
+        return 9;
+    }
+    if ((mode & 0x80) != 0) {
+        return 6;
+    }
+    return 5;
+}
+
+uvec2 DecodeBlockSize(uint layout, uint mode) {
    uint A, B;
-    switch (FindLayout(mode)) {
+    switch (layout) {
    case 0:
        A = (mode >> 5) & 0x3;
        B = (mode >> 7) & 0x3;
@ -967,6 +963,11 @@ uint DecodeMaxWeight(uint mode) {

 void DecompressBlock(ivec3 coord) {
    uint mode = StreamBits(11);
+    const uint num_partitions = StreamBits(2) + 1;
+    const uint mode_layout = FindLayout(mode);
+    const bool dual_plane = (mode_layout != 9) && ((mode & 0x400) != 0);
+    const uvec2 size_params = DecodeBlockSize(mode_layout, mode);
+
    if (IsError(mode)) {
        FillError(coord);
        return;
@ -976,15 +977,8 @@ void DecompressBlock(ivec3 coord) {
        FillVoidExtentLDR(coord);
        return;
    }
-    const uvec2 size_params = DecodeBlockSize(mode);
-    if ((size_params.x > block_dims.x) || (size_params.y > block_dims.y)) {
-        FillError(coord);
-        return;
-    }
-    const uint num_partitions = StreamBits(2) + 1;
-    const uint mode_layout = FindLayout(mode);
-    const bool dual_plane = (mode_layout != 9) && ((mode & 0x400) != 0);
-    if (num_partitions > 4 || (num_partitions == 4 && dual_plane)) {
+    if (((size_params.x > block_dims.x) || (size_params.y > block_dims.y))
+    || (num_partitions > 4 || (num_partitions == 4 && dual_plane))) {
        FillError(coord);
        return;
    }
@ -1002,25 +996,10 @@ void DecompressBlock(ivec3 coord) {
    const uint base_mode = base_cem & 3;
    const uint max_weight = DecodeMaxWeight(mode);
    const uint weight_bits = GetPackedBitSize(size_params, dual_plane, max_weight);
-    uint remaining_bits = 128 - weight_bits - total_bitsread;
-    uint extra_cem_bits = 0;
-    if (base_mode > 0) {
-        switch (num_partitions) {
-        case 2:
-            extra_cem_bits += 2;
-            break;
-        case 3:
-            extra_cem_bits += 5;
-            break;
-        case 4:
-            extra_cem_bits += 8;
-            break;
-        default:
-            return;
-        }
-    }
-    remaining_bits -= extra_cem_bits;
+    const uint extra_cem_bits = base_mode > 0 ? ((0x85200 >> (num_partitions * 4)) & 0x0f) : 0;
    const uint plane_selector_bits = dual_plane ? 2 : 0;
+    const uint remaining_bits = 128 - weight_bits - total_bitsread;
+    remaining_bits -= extra_cem_bits;
    remaining_bits -= plane_selector_bits;
    if (remaining_bits > 128) {
        // Bad data, more remaining bits than 4 bytes
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@ -538,16 +538,16 @@ ASTCDecoderPass::ASTCDecoderPass(const Device& device_, Scheduler& scheduler_,
                                 ComputePassDescriptorQueue& compute_pass_descriptor_queue_,
                                 MemoryAllocator& memory_allocator_)
    : ComputePass(device_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS,
-                  ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO,
-                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(AstcPushConstants)>, ASTC_DECODER_COMP_SPV),
-      scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
-      compute_pass_descriptor_queue{compute_pass_descriptor_queue_}, memory_allocator{
-                                                                         memory_allocator_} {}
+        ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO,
+        COMPUTE_PUSH_CONSTANT_RANGE<sizeof(AstcPushConstants)>, ASTC_DECODER_COMP_SPV)
+    , scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}
+    , compute_pass_descriptor_queue{compute_pass_descriptor_queue_}, memory_allocator{memory_allocator_}
+{}

 ASTCDecoderPass::~ASTCDecoderPass() = default;

-void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
-                               std::span<const VideoCommon::SwizzleParameters> swizzles) {
+void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    ASSERT(image.info.size.depth == 1);
    using namespace VideoCommon::Accelerated;
    const std::array<u32, 2> block_dims{
        VideoCore::Surface::DefaultBlockWidth(image.info.format),
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@ -2437,16 +2437,13 @@ void TextureCacheRuntime::AccelerateImageUpload(
    if (IsPixelFormatASTC(image.info.format)) {
        return astc_decoder_pass->Assemble(image, map, swizzles);
    }
-
    if (bl3d_unswizzle_pass &&
        IsPixelFormatBCn(image.info.format) &&
        image.info.type == ImageType::e3D &&
        image.info.resources.levels == 1 &&
        image.info.resources.layers == 1) {
-
        return bl3d_unswizzle_pass->Unswizzle(image, map, swizzles, z_start, z_count);
    }
-
    ASSERT(false);
 }