astc_decoder: Reduce workgroup size

This reduces the amount of over dispatching when there are odd dimensions (i.e. ASTC 8x5), which rarely evenly divide into 32x32.

src/common/settings.h

@@ -319,6 +319,7 @@ struct Values {
     // Renderer
     Setting<RendererBackend> renderer_backend{RendererBackend::OpenGL, "backend"};
     BasicSetting<bool> renderer_debug{false, "debug"};
+    BasicSetting<bool> renderer_shader_feedback{false, "shader_feedback"};
     BasicSetting<bool> enable_nsight_aftermath{false, "nsight_aftermath"};
     BasicSetting<bool> disable_shader_loop_safety_checks{false,
                                                          "disable_shader_loop_safety_checks"};

src/core/hle/api_version.h

@@ -12,9 +12,9 @@ namespace HLE::ApiVersion {
 
 // Horizon OS version constants.
 
-constexpr u8 HOS_VERSION_MAJOR = 11;
-constexpr u8 HOS_VERSION_MINOR = 0;
-constexpr u8 HOS_VERSION_MICRO = 1;
+constexpr u8 HOS_VERSION_MAJOR = 12;
+constexpr u8 HOS_VERSION_MINOR = 1;
+constexpr u8 HOS_VERSION_MICRO = 0;
 
 // NintendoSDK version constants.
 
@@ -22,15 +22,15 @@ constexpr u8 SDK_REVISION_MAJOR = 1;
 constexpr u8 SDK_REVISION_MINOR = 0;
 
 constexpr char PLATFORM_STRING[] = "NX";
-constexpr char VERSION_HASH[] = "69103fcb2004dace877094c2f8c29e6113be5dbf";
-constexpr char DISPLAY_VERSION[] = "11.0.1";
-constexpr char DISPLAY_TITLE[] = "NintendoSDK Firmware for NX 11.0.1-1.0";
+constexpr char VERSION_HASH[] = "76b10c2dab7d3aa73fc162f8dff1655e6a21caf4";
+constexpr char DISPLAY_VERSION[] = "12.1.0";
+constexpr char DISPLAY_TITLE[] = "NintendoSDK Firmware for NX 12.1.0-1.0";
 
 // Atmosphere version constants.
 
 constexpr u8 ATMOSPHERE_RELEASE_VERSION_MAJOR = 0;
 constexpr u8 ATMOSPHERE_RELEASE_VERSION_MINOR = 19;
-constexpr u8 ATMOSPHERE_RELEASE_VERSION_MICRO = 4;
+constexpr u8 ATMOSPHERE_RELEASE_VERSION_MICRO = 5;
 
 constexpr u32 GetTargetFirmware() {
     return u32{HOS_VERSION_MAJOR} << 24 | u32{HOS_VERSION_MINOR} << 16 |

src/core/hle/service/am/applets/applet_software_keyboard.cpp

@@ -377,7 +377,7 @@ void SoftwareKeyboard::SubmitForTextCheck(std::u16string submitted_text) {
 
     if (swkbd_config_common.use_utf8) {
         std::string utf8_submitted_text = Common::UTF16ToUTF8(current_text);
-        const u64 buffer_size = sizeof(u64) + utf8_submitted_text.size();
+        const u64 buffer_size = utf8_submitted_text.size();
 
         LOG_DEBUG(Service_AM, "\nBuffer Size: {}\nUTF-8 Submitted Text: {}", buffer_size,
                   utf8_submitted_text);
@@ -386,7 +386,7 @@ void SoftwareKeyboard::SubmitForTextCheck(std::u16string submitted_text) {
         std::memcpy(out_data.data() + sizeof(u64), utf8_submitted_text.data(),
                     utf8_submitted_text.size());
     } else {
-        const u64 buffer_size = sizeof(u64) + current_text.size() * sizeof(char16_t);
+        const u64 buffer_size = current_text.size() * sizeof(char16_t);
 
         LOG_DEBUG(Service_AM, "\nBuffer Size: {}\nUTF-16 Submitted Text: {}", buffer_size,
                   Common::UTF16ToUTF8(current_text));

src/core/hle/service/ns/language.cpp

@@ -339,6 +339,7 @@ std::optional<ApplicationLanguage> ConvertToApplicationLanguage(
     case Set::LanguageCode::FR_CA:
         return ApplicationLanguage::CanadianFrench;
     case Set::LanguageCode::PT:
+    case Set::LanguageCode::PT_BR:
         return ApplicationLanguage::Portuguese;
     case Set::LanguageCode::RU:
         return ApplicationLanguage::Russian;

src/core/hle/service/set/set.cpp

@@ -12,7 +12,7 @@
 
 namespace Service::Set {
 namespace {
-constexpr std::array<LanguageCode, 17> available_language_codes = {{
+constexpr std::array<LanguageCode, 18> available_language_codes = {{
     LanguageCode::JA,
     LanguageCode::EN_US,
     LanguageCode::FR,
@@ -30,6 +30,7 @@ constexpr std::array<LanguageCode, 17> available_language_codes = {{
     LanguageCode::ES_419,
     LanguageCode::ZH_HANS,
     LanguageCode::ZH_HANT,
+    LanguageCode::PT_BR,
 }};
 
 enum class KeyboardLayout : u64 {
@@ -50,7 +51,7 @@ enum class KeyboardLayout : u64 {
     ChineseTraditional = 14,
 };
 
-constexpr std::array<std::pair<LanguageCode, KeyboardLayout>, 17> language_to_layout{{
+constexpr std::array<std::pair<LanguageCode, KeyboardLayout>, 18> language_to_layout{{
     {LanguageCode::JA, KeyboardLayout::Japanese},
     {LanguageCode::EN_US, KeyboardLayout::EnglishUs},
     {LanguageCode::FR, KeyboardLayout::French},
@@ -68,10 +69,11 @@ constexpr std::array<std::pair<LanguageCode, KeyboardLayout>, 17> language_to_la
     {LanguageCode::ES_419, KeyboardLayout::SpanishLatin},
     {LanguageCode::ZH_HANS, KeyboardLayout::ChineseSimplified},
     {LanguageCode::ZH_HANT, KeyboardLayout::ChineseTraditional},
+    {LanguageCode::PT_BR, KeyboardLayout::Portuguese},
 }};
 
-constexpr std::size_t pre4_0_0_max_entries = 15;
-constexpr std::size_t post4_0_0_max_entries = 17;
+constexpr std::size_t PRE_4_0_0_MAX_ENTRIES = 0xF;
+constexpr std::size_t POST_4_0_0_MAX_ENTRIES = 0x40;
 
 constexpr ResultCode ERR_INVALID_LANGUAGE{ErrorModule::Settings, 625};
 
@@ -81,9 +83,9 @@ void PushResponseLanguageCode(Kernel::HLERequestContext& ctx, std::size_t num_la
     rb.Push(static_cast<u32>(num_language_codes));
 }
 
-void GetAvailableLanguageCodesImpl(Kernel::HLERequestContext& ctx, std::size_t max_size) {
+void GetAvailableLanguageCodesImpl(Kernel::HLERequestContext& ctx, std::size_t max_entries) {
     const std::size_t requested_amount = ctx.GetWriteBufferSize() / sizeof(LanguageCode);
-    const std::size_t copy_amount = std::min(requested_amount, max_size);
+    const std::size_t copy_amount = std::min(requested_amount, max_entries);
     const std::size_t copy_size = copy_amount * sizeof(LanguageCode);
 
     ctx.WriteBuffer(available_language_codes.data(), copy_size);
@@ -118,7 +120,7 @@ LanguageCode GetLanguageCodeFromIndex(std::size_t index) {
 void SET::GetAvailableLanguageCodes(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_SET, "called");
 
-    GetAvailableLanguageCodesImpl(ctx, pre4_0_0_max_entries);
+    GetAvailableLanguageCodesImpl(ctx, PRE_4_0_0_MAX_ENTRIES);
 }
 
 void SET::MakeLanguageCode(Kernel::HLERequestContext& ctx) {
@@ -140,19 +142,19 @@ void SET::MakeLanguageCode(Kernel::HLERequestContext& ctx) {
 void SET::GetAvailableLanguageCodes2(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_SET, "called");
 
-    GetAvailableLanguageCodesImpl(ctx, post4_0_0_max_entries);
+    GetAvailableLanguageCodesImpl(ctx, POST_4_0_0_MAX_ENTRIES);
 }
 
 void SET::GetAvailableLanguageCodeCount(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_SET, "called");
 
-    PushResponseLanguageCode(ctx, pre4_0_0_max_entries);
+    PushResponseLanguageCode(ctx, PRE_4_0_0_MAX_ENTRIES);
 }
 
 void SET::GetAvailableLanguageCodeCount2(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_SET, "called");
 
-    PushResponseLanguageCode(ctx, post4_0_0_max_entries);
+    PushResponseLanguageCode(ctx, POST_4_0_0_MAX_ENTRIES);
 }
 
 void SET::GetQuestFlag(Kernel::HLERequestContext& ctx) {

src/core/hle/service/set/set.h

@@ -31,6 +31,7 @@ enum class LanguageCode : u64 {
     ES_419 = 0x00003931342D7365,
     ZH_HANS = 0x00736E61482D687A,
     ZH_HANT = 0x00746E61482D687A,
+    PT_BR = 0x00000052422D7470,
 };
 LanguageCode GetLanguageCodeFromIndex(std::size_t idx);
 

src/video_core/CMakeLists.txt

@@ -106,6 +106,8 @@ add_library(video_core STATIC
     renderer_vulkan/maxwell_to_vk.cpp
     renderer_vulkan/maxwell_to_vk.h
     renderer_vulkan/pipeline_helper.h
+    renderer_vulkan/pipeline_statistics.cpp
+    renderer_vulkan/pipeline_statistics.h
     renderer_vulkan/renderer_vulkan.h
     renderer_vulkan/renderer_vulkan.cpp
     renderer_vulkan/vk_blit_screen.cpp

src/video_core/host_shaders/astc_decoder.comp

@@ -10,33 +10,27 @@
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)
 #define BINDING_INPUT_BUFFER 0
-#define BINDING_ENC_BUFFER 1
-#define BINDING_SWIZZLE_BUFFER 2
-#define BINDING_OUTPUT_IMAGE 3
+#define BINDING_OUTPUT_IMAGE 1
 
 #else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
 
 #define BEGIN_PUSH_CONSTANTS
 #define END_PUSH_CONSTANTS
 #define UNIFORM(n) layout(location = n) uniform
-#define BINDING_SWIZZLE_BUFFER 0
-#define BINDING_INPUT_BUFFER 1
-#define BINDING_ENC_BUFFER 2
+#define BINDING_INPUT_BUFFER 0
 #define BINDING_OUTPUT_IMAGE 0
 
 #endif
 
-layout(local_size_x = 32, local_size_y = 32, local_size_z = 1) in;
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
 
 BEGIN_PUSH_CONSTANTS
 UNIFORM(1) uvec2 block_dims;
-
-UNIFORM(2) uint bytes_per_block_log2;
-UNIFORM(3) uint layer_stride;
-UNIFORM(4) uint block_size;
-UNIFORM(5) uint x_shift;
-UNIFORM(6) uint block_height;
-UNIFORM(7) uint block_height_mask;
+UNIFORM(2) uint layer_stride;
+UNIFORM(3) uint block_size;
+UNIFORM(4) uint x_shift;
+UNIFORM(5) uint block_height;
+UNIFORM(6) uint block_height_mask;
 END_PUSH_CONSTANTS
 
 struct EncodingData {
@@ -55,45 +49,35 @@ struct TexelWeightParams {
     bool void_extent_hdr;
 };
 
-// Swizzle data
-layout(binding = BINDING_SWIZZLE_BUFFER, std430) readonly buffer SwizzleTable {
-    uint swizzle_table[];
-};
-
 layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU32 {
-    uint astc_data[];
-};
-
-// ASTC Encodings data
-layout(binding = BINDING_ENC_BUFFER, std430) readonly buffer EncodingsValues {
-    EncodingData encoding_values[];
+    uvec4 astc_data[];
 };
 
 layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly image2DArray dest_image;
 
-const uint GOB_SIZE_X = 64;
-const uint GOB_SIZE_Y = 8;
-const uint GOB_SIZE_Z = 1;
-const uint GOB_SIZE = GOB_SIZE_X * GOB_SIZE_Y * GOB_SIZE_Z;
-
 const uint GOB_SIZE_X_SHIFT = 6;
 const uint GOB_SIZE_Y_SHIFT = 3;
-const uint GOB_SIZE_Z_SHIFT = 0;
-const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHIFT;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT;
 
-const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1, GOB_SIZE_Y - 1);
-
-const int BLOCK_SIZE_IN_BYTES = 16;
-
-const int BLOCK_INFO_ERROR = 0;
-const int BLOCK_INFO_VOID_EXTENT_HDR = 1;
-const int BLOCK_INFO_VOID_EXTENT_LDR = 2;
-const int BLOCK_INFO_NORMAL = 3;
+const uint BYTES_PER_BLOCK_LOG2 = 4;
 
 const int JUST_BITS = 0;
 const int QUINT = 1;
 const int TRIT = 2;
 
+// ASTC Encodings data, sorted in ascending order based on their BitLength value
+// (see GetBitLength() function)
+EncodingData encoding_values[22] = EncodingData[](
+    EncodingData(JUST_BITS, 0, 0, 0), EncodingData(JUST_BITS, 1, 0, 0), EncodingData(TRIT, 0, 0, 0),
+    EncodingData(JUST_BITS, 2, 0, 0), EncodingData(QUINT, 0, 0, 0), EncodingData(TRIT, 1, 0, 0),
+    EncodingData(JUST_BITS, 3, 0, 0), EncodingData(QUINT, 1, 0, 0), EncodingData(TRIT, 2, 0, 0),
+    EncodingData(JUST_BITS, 4, 0, 0), EncodingData(QUINT, 2, 0, 0), EncodingData(TRIT, 3, 0, 0),
+    EncodingData(JUST_BITS, 5, 0, 0), EncodingData(QUINT, 3, 0, 0), EncodingData(TRIT, 4, 0, 0),
+    EncodingData(JUST_BITS, 6, 0, 0), EncodingData(QUINT, 4, 0, 0), EncodingData(TRIT, 5, 0, 0),
+    EncodingData(JUST_BITS, 7, 0, 0), EncodingData(QUINT, 5, 0, 0), EncodingData(TRIT, 6, 0, 0),
+    EncodingData(JUST_BITS, 8, 0, 0)
+);
+
 // The following constants are expanded variants of the Replicate()
 // function calls corresponding to the following arguments:
 // value: index into the generated table
@@ -135,44 +119,37 @@ const uint REPLICATE_7_BIT_TO_8_TABLE[128] =
 // Input ASTC texture globals
 uint current_index = 0;
 int bitsread = 0;
-uint total_bitsread = 0;
-uint local_buff[16];
+int total_bitsread = 0;
+uvec4 local_buff;
 
 // Color data globals
-uint color_endpoint_data[16];
+uvec4 color_endpoint_data;
 int color_bitsread = 0;
-uint total_color_bitsread = 0;
-int color_index = 0;
 
 // Four values, two endpoints, four maximum paritions
 uint color_values[32];
 int colvals_index = 0;
 
 // Weight data globals
-uint texel_weight_data[16];
+uvec4 texel_weight_data;
 int texel_bitsread = 0;
-uint total_texel_bitsread = 0;
-int texel_index = 0;
 
 bool texel_flag = false;
 
 // Global "vectors" to be pushed into when decoding
-EncodingData result_vector[100];
+EncodingData result_vector[144];
 int result_index = 0;
 
-EncodingData texel_vector[100];
+EncodingData texel_vector[144];
 int texel_vector_index = 0;
 
 uint unquantized_texel_weights[2][144];
 
 uint SwizzleOffset(uvec2 pos) {
-    pos = pos & SWIZZLE_MASK;
-    return swizzle_table[pos.y * 64 + pos.x];
-}
-
-uint ReadTexel(uint offset) {
-    // extract the 8-bit value from the 32-bit packed data.
-    return bitfieldExtract(astc_data[offset / 4], int((offset * 8) & 24), 8);
+    uint x = pos.x;
+    uint y = pos.y;
+    return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
+                          (y % 2) * 16 + (x % 16);
 }
 
 // Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
@@ -278,14 +255,10 @@ uint Hash52(uint p) {
     return p;
 }
 
-uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bool small_block) {
-    if (partition_count == 1) {
-        return 0;
-    }
+uint Select2DPartition(uint seed, uint x, uint y, uint partition_count, bool small_block) {
     if (small_block) {
         x <<= 1;
         y <<= 1;
-        z <<= 1;
     }
 
     seed += (partition_count - 1) * 1024;
@@ -299,10 +272,6 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
     uint seed6 = uint((rnum >> 20) & 0xF);
     uint seed7 = uint((rnum >> 24) & 0xF);
     uint seed8 = uint((rnum >> 28) & 0xF);
-    uint seed9 = uint((rnum >> 18) & 0xF);
-    uint seed10 = uint((rnum >> 22) & 0xF);
-    uint seed11 = uint((rnum >> 26) & 0xF);
-    uint seed12 = uint(((rnum >> 30) | (rnum << 2)) & 0xF);
 
     seed1 = (seed1 * seed1);
     seed2 = (seed2 * seed2);
@@ -312,12 +281,8 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
     seed6 = (seed6 * seed6);
     seed7 = (seed7 * seed7);
     seed8 = (seed8 * seed8);
-    seed9 = (seed9 * seed9);
-    seed10 = (seed10 * seed10);
-    seed11 = (seed11 * seed11);
-    seed12 = (seed12 * seed12);
 
-    int sh1, sh2, sh3;
+    uint sh1, sh2;
     if ((seed & 1) > 0) {
         sh1 = (seed & 2) > 0 ? 4 : 5;
         sh2 = (partition_count == 3) ? 6 : 5;
@@ -325,25 +290,19 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
         sh1 = (partition_count == 3) ? 6 : 5;
         sh2 = (seed & 2) > 0 ? 4 : 5;
     }
-    sh3 = (seed & 0x10) > 0 ? sh1 : sh2;
+    seed1 >>= sh1;
+    seed2 >>= sh2;
+    seed3 >>= sh1;
+    seed4 >>= sh2;
+    seed5 >>= sh1;
+    seed6 >>= sh2;
+    seed7 >>= sh1;
+    seed8 >>= sh2;
 
-    seed1 = (seed1 >> sh1);
-    seed2 = (seed2 >> sh2);
-    seed3 = (seed3 >> sh1);
-    seed4 = (seed4 >> sh2);
-    seed5 = (seed5 >> sh1);
-    seed6 = (seed6 >> sh2);
-    seed7 = (seed7 >> sh1);
-    seed8 = (seed8 >> sh2);
-    seed9 = (seed9 >> sh3);
-    seed10 = (seed10 >> sh3);
-    seed11 = (seed11 >> sh3);
-    seed12 = (seed12 >> sh3);
-
-    uint a = seed1 * x + seed2 * y + seed11 * z + (rnum >> 14);
-    uint b = seed3 * x + seed4 * y + seed12 * z + (rnum >> 10);
-    uint c = seed5 * x + seed6 * y + seed9 * z + (rnum >> 6);
-    uint d = seed7 * x + seed8 * y + seed10 * z + (rnum >> 2);
+    uint a = seed1 * x + seed2 * y + (rnum >> 14);
+    uint b = seed3 * x + seed4 * y + (rnum >> 10);
+    uint c = seed5 * x + seed6 * y + (rnum >> 6);
+    uint d = seed7 * x + seed8 * y + (rnum >> 2);
 
     a &= 0x3F;
     b &= 0x3F;
@@ -368,58 +327,37 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
     }
 }
 
-uint Select2DPartition(uint seed, uint x, uint y, uint partition_count, bool small_block) {
-    return SelectPartition(seed, x, y, 0, partition_count, small_block);
-}
-
-uint ReadBit() {
-    if (current_index >= local_buff.length()) {
+uint ExtractBits(uvec4 payload, int offset, int bits) {
+    if (bits <= 0) {
         return 0;
     }
-    uint bit = bitfieldExtract(local_buff[current_index], bitsread, 1);
-    ++bitsread;
-    ++total_bitsread;
-    if (bitsread == 8) {
-        ++current_index;
-        bitsread = 0;
+    int last_offset = offset + bits - 1;
+    int shifted_offset = offset >> 5;
+    if ((last_offset >> 5) == shifted_offset) {
+        return bitfieldExtract(payload[shifted_offset], offset & 31, bits);
     }
-    return bit;
+    int first_bits = 32 - (offset & 31);
+    int result_first = int(bitfieldExtract(payload[shifted_offset], offset & 31, first_bits));
+    int result_second = int(bitfieldExtract(payload[shifted_offset + 1], 0, bits - first_bits));
+    return result_first | (result_second << first_bits);
 }
 
 uint StreamBits(uint num_bits) {
-    uint ret = 0;
-    for (uint i = 0; i < num_bits; i++) {
-        ret |= ((ReadBit() & 1) << i);
-    }
+    int int_bits = int(num_bits);
+    uint ret = ExtractBits(local_buff, total_bitsread, int_bits);
+    total_bitsread += int_bits;
     return ret;
 }
 
-uint ReadColorBit() {
-    uint bit = 0;
-    if (texel_flag) {
-        bit = bitfieldExtract(texel_weight_data[texel_index], texel_bitsread, 1);
-        ++texel_bitsread;
-        ++total_texel_bitsread;
-        if (texel_bitsread == 8) {
-            ++texel_index;
-            texel_bitsread = 0;
-        }
-    } else {
-        bit = bitfieldExtract(color_endpoint_data[color_index], color_bitsread, 1);
-        ++color_bitsread;
-        ++total_color_bitsread;
-        if (color_bitsread == 8) {
-            ++color_index;
-            color_bitsread = 0;
-        }
-    }
-    return bit;
-}
-
 uint StreamColorBits(uint num_bits) {
     uint ret = 0;
-    for (uint i = 0; i < num_bits; i++) {
-        ret |= ((ReadColorBit() & 1) << i);
+    int int_bits = int(num_bits);
+    if (texel_flag) {
+        ret = ExtractBits(texel_weight_data, texel_bitsread, int_bits);
+        texel_bitsread += int_bits;
+    } else {
+        ret = ExtractBits(color_endpoint_data, color_bitsread, int_bits);
+        color_bitsread += int_bits;
     }
     return ret;
 }
@@ -596,22 +534,16 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
     for (uint i = 0; i < num_partitions; i++) {
         num_values += ((modes[i] >> 2) + 1) << 1;
     }
-    int range = 256;
-    while (--range > 0) {
-        EncodingData val = encoding_values[range];
+    // Find the largest encoding that's within color_data_bits
+    // TODO(ameerj): profile with binary search
+    int range = 0;
+    while (++range < encoding_values.length()) {
         uint bit_length = GetBitLength(num_values, range);
-        if (bit_length <= color_data_bits) {
-            while (--range > 0) {
-                EncodingData newval = encoding_values[range];
-                if (newval.encoding != val.encoding && newval.num_bits != val.num_bits) {
-                    break;
-                }
-            }
-            ++range;
+        if (bit_length > color_data_bits) {
             break;
         }
     }
-    DecodeIntegerSequence(range, num_values);
+    DecodeIntegerSequence(range - 1, num_values);
     uint out_index = 0;
     for (int itr = 0; itr < result_index; ++itr) {
         if (out_index >= num_values) {
@@ -1028,7 +960,7 @@ int FindLayout(uint mode) {
     return 5;
 }
 
-TexelWeightParams DecodeBlockInfo(uint block_index) {
+TexelWeightParams DecodeBlockInfo() {
     TexelWeightParams params = TexelWeightParams(uvec2(0), 0, false, false, false, false);
     uint mode = StreamBits(11);
     if ((mode & 0x1ff) == 0x1fc) {
@@ -1110,10 +1042,10 @@ TexelWeightParams DecodeBlockInfo(uint block_index) {
     }
     weight_index -= 2;
     if ((mode_layout != 9) && ((mode & 0x200) != 0)) {
-        const int max_weights[6] = int[6](9, 11, 15, 19, 23, 31);
+        const int max_weights[6] = int[6](7, 8, 9, 10, 11, 12);
         params.max_weight = max_weights[weight_index];
     } else {
-        const int max_weights[6] = int[6](1, 2, 3, 4, 5, 7);
+        const int max_weights[6] = int[6](1, 2, 3, 4, 5, 6);
         params.max_weight = max_weights[weight_index];
     }
     return params;
@@ -1144,8 +1076,8 @@ void FillVoidExtentLDR(ivec3 coord) {
     }
 }
 
-void DecompressBlock(ivec3 coord, uint block_index) {
-    TexelWeightParams params = DecodeBlockInfo(block_index);
+void DecompressBlock(ivec3 coord) {
+    TexelWeightParams params = DecodeBlockInfo();
     if (params.error_state) {
         FillError(coord);
         return;
@@ -1212,7 +1144,7 @@ void DecompressBlock(ivec3 coord, uint block_index) {
     // Read color data...
     uint color_data_bits = remaining_bits;
     while (remaining_bits > 0) {
-        int nb = int(min(remaining_bits, 8U));
+        int nb = int(min(remaining_bits, 32U));
         uint b = StreamBits(nb);
         color_endpoint_data[ced_pointer] = uint(bitfieldExtract(b, 0, nb));
         ++ced_pointer;
@@ -1254,25 +1186,20 @@ void DecompressBlock(ivec3 coord, uint block_index) {
         ComputeEndpoints(endpoints[i][0], endpoints[i][1], color_endpoint_mode[i]);
     }
 
-    for (uint i = 0; i < 16; i++) {
-        texel_weight_data[i] = local_buff[i];
-    }
-    for (uint i = 0; i < 8; i++) {
-#define REVERSE_BYTE(b) ((b * 0x0802U & 0x22110U) | (b * 0x8020U & 0x88440U)) * 0x10101U >> 16
-        uint a = REVERSE_BYTE(texel_weight_data[i]);
-        uint b = REVERSE_BYTE(texel_weight_data[15 - i]);
-#undef REVERSE_BYTE
-        texel_weight_data[i] = uint(bitfieldExtract(b, 0, 8));
-        texel_weight_data[15 - i] = uint(bitfieldExtract(a, 0, 8));
-    }
+    texel_weight_data = local_buff;
+    texel_weight_data = bitfieldReverse(texel_weight_data).wzyx;
     uint clear_byte_start =
         (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) >> 3) + 1;
-    texel_weight_data[clear_byte_start - 1] =
-        texel_weight_data[clear_byte_start - 1] &
+
+    uint byte_insert = ExtractBits(texel_weight_data, int(clear_byte_start - 1) * 8, 8) &
         uint(
             ((1 << (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) % 8)) - 1));
-    for (uint i = 0; i < 16 - clear_byte_start; i++) {
-        texel_weight_data[clear_byte_start + i] = 0U;
+    uint vec_index = (clear_byte_start - 1) >> 2;
+    texel_weight_data[vec_index] =
+        bitfieldInsert(texel_weight_data[vec_index], byte_insert, int((clear_byte_start - 1) % 4) * 8, 8);
+    for (uint i = clear_byte_start; i < 16; ++i) {
+        uint idx = i >> 2;
+        texel_weight_data[idx] = bitfieldInsert(texel_weight_data[idx], 0, int(i % 4) * 8, 8);
     }
     texel_flag = true; // use texel "vector" and bit stream in integer decoding
     DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane));
@@ -1281,8 +1208,11 @@ void DecompressBlock(ivec3 coord, uint block_index) {
 
     for (uint j = 0; j < block_dims.y; j++) {
         for (uint i = 0; i < block_dims.x; i++) {
-            uint local_partition = Select2DPartition(partition_index, i, j, num_partitions,
+            uint local_partition = 0;
+            if (num_partitions > 1) {
+                local_partition = Select2DPartition(partition_index, i, j, num_partitions,
                                                      (block_dims.y * block_dims.x) < 32);
+            }
             vec4 p;
             uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]);
             uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]);
@@ -1303,7 +1233,7 @@ void DecompressBlock(ivec3 coord, uint block_index) {
 
 void main() {
     uvec3 pos = gl_GlobalInvocationID;
-    pos.x <<= bytes_per_block_log2;
+    pos.x <<= BYTES_PER_BLOCK_LOG2;
 
     // Read as soon as possible due to its latency
     const uint swizzle = SwizzleOffset(pos.xy);
@@ -1321,13 +1251,8 @@ void main() {
     if (any(greaterThanEqual(coord, imageSize(dest_image)))) {
         return;
     }
-    uint block_index =
-        pos.z * gl_WorkGroupSize.x * gl_WorkGroupSize.y + pos.y * gl_WorkGroupSize.x + pos.x;
-
     current_index = 0;
     bitsread = 0;
-    for (int i = 0; i < 16; i++) {
-        local_buff[i] = ReadTexel(offset + i);
-    }
-    DecompressBlock(coord, block_index);
+    local_buff = astc_data[offset / 16];
+    DecompressBlock(coord);
 }

src/video_core/renderer_opengl/util_shaders.cpp

@@ -60,19 +60,14 @@ UtilShaders::UtilShaders(ProgramManager& program_manager_)
       copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)) {
     const auto swizzle_table = Tegra::Texture::MakeSwizzleTable();
     swizzle_table_buffer.Create();
-    astc_buffer.Create();
     glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0);
-    glNamedBufferStorage(astc_buffer.handle, sizeof(ASTC_ENCODINGS_VALUES), &ASTC_ENCODINGS_VALUES,
-                         0);
 }
 
 UtilShaders::~UtilShaders() = default;
 
 void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
                              std::span<const VideoCommon::SwizzleParameters> swizzles) {
-    static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0;
-    static constexpr GLuint BINDING_INPUT_BUFFER = 1;
-    static constexpr GLuint BINDING_ENC_BUFFER = 2;
+    static constexpr GLuint BINDING_INPUT_BUFFER = 0;
     static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
 
     const Extent2D tile_size{
@@ -80,34 +75,32 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
         .height = VideoCore::Surface::DefaultBlockHeight(image.info.format),
     };
     program_manager.BindComputeProgram(astc_decoder_program.handle);
-    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
-    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_ENC_BUFFER, astc_buffer.handle);
-
     glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
     glUniform2ui(1, tile_size.width, tile_size.height);
+
     // Ensure buffer data is valid before dispatching
     glFlush();
     for (const SwizzleParameters& swizzle : swizzles) {
         const size_t input_offset = swizzle.buffer_offset + map.offset;
-        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
-        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
+        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 8U);
+        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U);
 
         const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
         ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
         ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
+        ASSERT(params.bytes_per_block_log2 == 4);
 
-        glUniform1ui(2, params.bytes_per_block_log2);
-        glUniform1ui(3, params.layer_stride);
-        glUniform1ui(4, params.block_size);
-        glUniform1ui(5, params.x_shift);
-        glUniform1ui(6, params.block_height);
-        glUniform1ui(7, params.block_height_mask);
+        glUniform1ui(2, params.layer_stride);
+        glUniform1ui(3, params.block_size);
+        glUniform1ui(4, params.x_shift);
+        glUniform1ui(5, params.block_height);
+        glUniform1ui(6, params.block_height_mask);
 
-        glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), swizzle.level, GL_TRUE, 0,
-                           GL_WRITE_ONLY, GL_RGBA8);
         // ASTC texture data
         glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset,
                           image.guest_size_bytes - swizzle.buffer_offset);
+        glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), swizzle.level, GL_TRUE, 0,
+                           GL_WRITE_ONLY, GL_RGBA8);
 
         glDispatchCompute(num_dispatches_x, num_dispatches_y, image.info.resources.layers);
     }

src/video_core/renderer_opengl/util_shaders.h

@@ -62,7 +62,6 @@ private:
     ProgramManager& program_manager;
 
     OGLBuffer swizzle_table_buffer;
-    OGLBuffer astc_buffer;
 
     OGLProgram astc_decoder_program;
     OGLProgram block_linear_unswizzle_2d_program;

src/video_core/renderer_vulkan/pipeline_statistics.cpp

@@ -0,0 +1,100 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <string_view>
+
+#include <fmt/format.h>
+
+#include "common/common_types.h"
+#include "common/logging/log.h"
+#include "video_core/renderer_vulkan/pipeline_statistics.h"
+#include "video_core/vulkan_common/vulkan_device.h"
+#include "video_core/vulkan_common/vulkan_wrapper.h"
+
+namespace Vulkan {
+
+using namespace std::string_view_literals;
+
+static u64 GetUint64(const VkPipelineExecutableStatisticKHR& statistic) {
+    switch (statistic.format) {
+    case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_INT64_KHR:
+        return static_cast<u64>(statistic.value.i64);
+    case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_UINT64_KHR:
+        return statistic.value.u64;
+    case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_FLOAT64_KHR:
+        return static_cast<u64>(statistic.value.f64);
+    default:
+        return 0;
+    }
+}
+
+PipelineStatistics::PipelineStatistics(const Device& device_) : device{device_} {}
+
+void PipelineStatistics::Collect(VkPipeline pipeline) {
+    const auto& dev{device.GetLogical()};
+    const std::vector properties{dev.GetPipelineExecutablePropertiesKHR(pipeline)};
+    const u32 num_executables{static_cast<u32>(properties.size())};
+    for (u32 executable = 0; executable < num_executables; ++executable) {
+        const auto statistics{dev.GetPipelineExecutableStatisticsKHR(pipeline, executable)};
+        if (statistics.empty()) {
+            continue;
+        }
+        Stats stage_stats;
+        for (const auto& statistic : statistics) {
+            const char* const name{statistic.name};
+            if (name == "Binary Size"sv || name == "Code size"sv || name == "Instruction Count"sv) {
+                stage_stats.code_size = GetUint64(statistic);
+            } else if (name == "Register Count"sv) {
+                stage_stats.register_count = GetUint64(statistic);
+            } else if (name == "SGPRs"sv || name == "numUsedSgprs"sv) {
+                stage_stats.sgpr_count = GetUint64(statistic);
+            } else if (name == "VGPRs"sv || name == "numUsedVgprs"sv) {
+                stage_stats.vgpr_count = GetUint64(statistic);
+            } else if (name == "Branches"sv) {
+                stage_stats.branches_count = GetUint64(statistic);
+            } else if (name == "Basic Block Count"sv) {
+                stage_stats.basic_block_count = GetUint64(statistic);
+            }
+        }
+        std::lock_guard lock{mutex};
+        collected_stats.push_back(stage_stats);
+    }
+}
+
+void PipelineStatistics::Report() const {
+    double num{};
+    Stats total;
+    {
+        std::lock_guard lock{mutex};
+        for (const Stats& stats : collected_stats) {
+            total.code_size += stats.code_size;
+            total.register_count += stats.register_count;
+            total.sgpr_count += stats.sgpr_count;
+            total.vgpr_count += stats.vgpr_count;
+            total.branches_count += stats.branches_count;
+            total.basic_block_count += stats.basic_block_count;
+        }
+        num = static_cast<double>(collected_stats.size());
+    }
+    std::string report;
+    const auto add = [&](const char* fmt, u64 value) {
+        if (value > 0) {
+            report += fmt::format(fmt::runtime(fmt), static_cast<double>(value) / num);
+        }
+    };
+    add("Code size:      {:9.03f}\n", total.code_size);
+    add("Register count: {:9.03f}\n", total.register_count);
+    add("SGPRs:          {:9.03f}\n", total.sgpr_count);
+    add("VGPRs:          {:9.03f}\n", total.vgpr_count);
+    add("Branches count: {:9.03f}\n", total.branches_count);
+    add("Basic blocks:   {:9.03f}\n", total.basic_block_count);
+
+    LOG_INFO(Render_Vulkan,
+             "\nAverage pipeline statistics\n"
+             "==========================================\n"
+             "{}\n",
+             report);
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/pipeline_statistics.h

@@ -0,0 +1,40 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <mutex>
+#include <vector>
+
+#include "common/common_types.h"
+#include "video_core/vulkan_common/vulkan_wrapper.h"
+
+namespace Vulkan {
+
+class Device;
+
+class PipelineStatistics {
+public:
+    explicit PipelineStatistics(const Device& device_);
+
+    void Collect(VkPipeline pipeline);
+
+    void Report() const;
+
+private:
+    struct Stats {
+        u64 code_size{};
+        u64 register_count{};
+        u64 sgpr_count{};
+        u64 vgpr_count{};
+        u64 branches_count{};
+        u64 basic_block_count{};
+    };
+
+    const Device& device;
+    mutable std::mutex mutex;
+    std::vector<Stats> collected_stats;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -30,16 +30,12 @@
 namespace Vulkan {
 
 using Tegra::Texture::SWIZZLE_TABLE;
-using Tegra::Texture::ASTC::ASTC_ENCODINGS_VALUES;
-using namespace Tegra::Texture::ASTC;
 
 namespace {
 
 constexpr u32 ASTC_BINDING_INPUT_BUFFER = 0;
-constexpr u32 ASTC_BINDING_ENC_BUFFER = 1;
-constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 2;
-constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 3;
-constexpr size_t ASTC_NUM_BINDINGS = 4;
+constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 1;
+constexpr size_t ASTC_NUM_BINDINGS = 2;
 
 template <size_t size>
 inline constexpr VkPushConstantRange COMPUTE_PUSH_CONSTANT_RANGE{
@@ -75,7 +71,7 @@ constexpr DescriptorBankInfo INPUT_OUTPUT_BANK_INFO{
     .score = 2,
 };
 
-constexpr std::array<VkDescriptorSetLayoutBinding, 4> ASTC_DESCRIPTOR_SET_BINDINGS{{
+constexpr std::array<VkDescriptorSetLayoutBinding, ASTC_NUM_BINDINGS> ASTC_DESCRIPTOR_SET_BINDINGS{{
     {
         .binding = ASTC_BINDING_INPUT_BUFFER,
         .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
@@ -83,20 +79,6 @@ constexpr std::array<VkDescriptorSetLayoutBinding, 4> ASTC_DESCRIPTOR_SET_BINDIN
         .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
         .pImmutableSamplers = nullptr,
     },
-    {
-        .binding = ASTC_BINDING_ENC_BUFFER,
-        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-        .descriptorCount = 1,
-        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-        .pImmutableSamplers = nullptr,
-    },
-    {
-        .binding = ASTC_BINDING_SWIZZLE_BUFFER,
-        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-        .descriptorCount = 1,
-        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-        .pImmutableSamplers = nullptr,
-    },
     {
         .binding = ASTC_BINDING_OUTPUT_IMAGE,
         .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
@@ -108,12 +90,12 @@ constexpr std::array<VkDescriptorSetLayoutBinding, 4> ASTC_DESCRIPTOR_SET_BINDIN
 
 constexpr DescriptorBankInfo ASTC_BANK_INFO{
     .uniform_buffers = 0,
-    .storage_buffers = 3,
+    .storage_buffers = 1,
     .texture_buffers = 0,
     .image_buffers = 0,
     .textures = 0,
     .images = 1,
-    .score = 4,
+    .score = 2,
 };
 
 constexpr VkDescriptorUpdateTemplateEntryKHR INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE{
@@ -135,22 +117,6 @@ constexpr std::array<VkDescriptorUpdateTemplateEntryKHR, ASTC_NUM_BINDINGS>
             .offset = ASTC_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
             .stride = sizeof(DescriptorUpdateEntry),
         },
-        {
-            .dstBinding = ASTC_BINDING_ENC_BUFFER,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-            .offset = ASTC_BINDING_ENC_BUFFER * sizeof(DescriptorUpdateEntry),
-            .stride = sizeof(DescriptorUpdateEntry),
-        },
-        {
-            .dstBinding = ASTC_BINDING_SWIZZLE_BUFFER,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-            .offset = ASTC_BINDING_SWIZZLE_BUFFER * sizeof(DescriptorUpdateEntry),
-            .stride = sizeof(DescriptorUpdateEntry),
-        },
         {
             .dstBinding = ASTC_BINDING_OUTPUT_IMAGE,
             .dstArrayElement = 0,
@@ -163,7 +129,6 @@ constexpr std::array<VkDescriptorUpdateTemplateEntryKHR, ASTC_NUM_BINDINGS>
 
 struct AstcPushConstants {
     std::array<u32, 2> blocks_dims;
-    u32 bytes_per_block_log2;
     u32 layer_stride;
     u32 block_size;
     u32 x_shift;
@@ -354,46 +319,6 @@ ASTCDecoderPass::ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,
 
 ASTCDecoderPass::~ASTCDecoderPass() = default;
 
-void ASTCDecoderPass::MakeDataBuffer() {
-    constexpr size_t TOTAL_BUFFER_SIZE = sizeof(ASTC_ENCODINGS_VALUES) + sizeof(SWIZZLE_TABLE);
-    data_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
-        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .size = TOTAL_BUFFER_SIZE,
-        .usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
-        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
-        .queueFamilyIndexCount = 0,
-        .pQueueFamilyIndices = nullptr,
-    });
-    data_buffer_commit = memory_allocator.Commit(data_buffer, MemoryUsage::Upload);
-
-    const auto staging_ref = staging_buffer_pool.Request(TOTAL_BUFFER_SIZE, MemoryUsage::Upload);
-    std::memcpy(staging_ref.mapped_span.data(), &ASTC_ENCODINGS_VALUES,
-                sizeof(ASTC_ENCODINGS_VALUES));
-    // Tack on the swizzle table at the end of the buffer
-    std::memcpy(staging_ref.mapped_span.data() + sizeof(ASTC_ENCODINGS_VALUES), &SWIZZLE_TABLE,
-                sizeof(SWIZZLE_TABLE));
-
-    scheduler.Record([src = staging_ref.buffer, offset = staging_ref.offset, dst = *data_buffer,
-                      TOTAL_BUFFER_SIZE](vk::CommandBuffer cmdbuf) {
-        static constexpr VkMemoryBarrier write_barrier{
-            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
-        };
-        const VkBufferCopy copy{
-            .srcOffset = offset,
-            .dstOffset = 0,
-            .size = TOTAL_BUFFER_SIZE,
-        };
-        cmdbuf.CopyBuffer(src, dst, copy);
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
-                               0, write_barrier);
-    });
-}
-
 void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
                                std::span<const VideoCommon::SwizzleParameters> swizzles) {
     using namespace VideoCommon::Accelerated;
@@ -402,9 +327,6 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
         VideoCore::Surface::DefaultBlockHeight(image.info.format),
     };
     scheduler.RequestOutsideRenderPassOperationContext();
-    if (!data_buffer) {
-        MakeDataBuffer();
-    }
     const VkPipeline vk_pipeline = *pipeline;
     const VkImageAspectFlags aspect_mask = image.AspectMask();
     const VkImage vk_image = image.Handle();
@@ -436,16 +358,13 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
         });
     for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
         const size_t input_offset = swizzle.buffer_offset + map.offset;
-        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
-        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
+        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 8U);
+        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U);
         const u32 num_dispatches_z = image.info.resources.layers;
 
         update_descriptor_queue.Acquire();
         update_descriptor_queue.AddBuffer(map.buffer, input_offset,
                                           image.guest_size_bytes - swizzle.buffer_offset);
-        update_descriptor_queue.AddBuffer(*data_buffer, 0, sizeof(ASTC_ENCODINGS_VALUES));
-        update_descriptor_queue.AddBuffer(*data_buffer, sizeof(ASTC_ENCODINGS_VALUES),
-                                          sizeof(SWIZZLE_TABLE));
         update_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
         const void* const descriptor_data{update_descriptor_queue.UpdateData()};
 
@@ -453,11 +372,11 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
         const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
         ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
         ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
+        ASSERT(params.bytes_per_block_log2 == 4);
         scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, block_dims,
                           params, descriptor_data](vk::CommandBuffer cmdbuf) {
             const AstcPushConstants uniforms{
                 .blocks_dims = block_dims,
-                .bytes_per_block_log2 = params.bytes_per_block_log2,
                 .layer_stride = params.layer_stride,
                 .block_size = params.block_size,
                 .x_shift = params.x_shift,

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -96,15 +96,10 @@ public:
                   std::span<const VideoCommon::SwizzleParameters> swizzles);
 
 private:
-    void MakeDataBuffer();
-
     VKScheduler& scheduler;
     StagingBufferPool& staging_buffer_pool;
     VKUpdateDescriptorQueue& update_descriptor_queue;
     MemoryAllocator& memory_allocator;
-
-    vk::Buffer data_buffer;
-    MemoryCommit data_buffer_commit;
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -8,6 +8,7 @@
 #include <boost/container/small_vector.hpp>
 
 #include "video_core/renderer_vulkan/pipeline_helper.h"
+#include "video_core/renderer_vulkan/pipeline_statistics.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
 #include "video_core/renderer_vulkan/vk_compute_pipeline.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
@@ -26,6 +27,7 @@ using Tegra::Texture::TexturePair;
 ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descriptor_pool,
                                  VKUpdateDescriptorQueue& update_descriptor_queue_,
                                  Common::ThreadWorker* thread_worker,
+                                 PipelineStatistics* pipeline_statistics,
                                  VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_,
                                  vk::ShaderModule spv_module_)
     : device{device_}, update_descriptor_queue{update_descriptor_queue_}, info{info_},
@@ -36,7 +38,7 @@ ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descript
     std::copy_n(info.constant_buffer_used_sizes.begin(), uniform_buffer_sizes.size(),
                 uniform_buffer_sizes.begin());
 
-    auto func{[this, &descriptor_pool, shader_notify] {
+    auto func{[this, &descriptor_pool, shader_notify, pipeline_statistics] {
         DescriptorLayoutBuilder builder{device};
         builder.Add(info, VK_SHADER_STAGE_COMPUTE_BIT);
 
@@ -50,10 +52,14 @@ ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descript
             .pNext = nullptr,
             .requiredSubgroupSize = GuestWarpSize,
         };
+        VkPipelineCreateFlags flags{};
+        if (device.IsKhrPipelineEexecutablePropertiesEnabled()) {
+            flags |= VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR;
+        }
         pipeline = device.GetLogical().CreateComputePipeline({
             .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
             .pNext = nullptr,
-            .flags = 0,
+            .flags = flags,
             .stage{
                 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
                 .pNext = device.IsExtSubgroupSizeControlSupported() ? &subgroup_size_ci : nullptr,
@@ -67,6 +73,9 @@ ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descript
             .basePipelineHandle = 0,
             .basePipelineIndex = 0,
         });
+        if (pipeline_statistics) {
+            pipeline_statistics->Collect(*pipeline);
+        }
         std::lock_guard lock{build_mutex};
         is_built = true;
         build_condvar.notify_one();

src/video_core/renderer_vulkan/vk_compute_pipeline.h

@@ -25,6 +25,7 @@ class ShaderNotify;
 namespace Vulkan {
 
 class Device;
+class PipelineStatistics;
 class VKScheduler;
 
 class ComputePipeline {
@@ -32,6 +33,7 @@ public:
     explicit ComputePipeline(const Device& device, DescriptorPool& descriptor_pool,
                              VKUpdateDescriptorQueue& update_descriptor_queue,
                              Common::ThreadWorker* thread_worker,
+                             PipelineStatistics* pipeline_statistics,
                              VideoCore::ShaderNotify* shader_notify, const Shader::Info& info,
                              vk::ShaderModule spv_module);
 

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -11,6 +11,7 @@
 #include "common/bit_field.h"
 #include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/renderer_vulkan/pipeline_helper.h"
+#include "video_core/renderer_vulkan/pipeline_statistics.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
 #include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
 #include "video_core/renderer_vulkan/vk_render_pass_cache.h"
@@ -217,8 +218,8 @@ GraphicsPipeline::GraphicsPipeline(
     VKScheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_,
     VideoCore::ShaderNotify* shader_notify, const Device& device_, DescriptorPool& descriptor_pool,
     VKUpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* worker_thread,
-    RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key_,
-    std::array<vk::ShaderModule, NUM_STAGES> stages,
+    PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache,
+    const GraphicsPipelineCacheKey& key_, std::array<vk::ShaderModule, NUM_STAGES> stages,
     const std::array<const Shader::Info*, NUM_STAGES>& infos)
     : key{key_}, maxwell3d{maxwell3d_}, gpu_memory{gpu_memory_}, device{device_},
       texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, scheduler{scheduler_},
@@ -235,7 +236,7 @@ GraphicsPipeline::GraphicsPipeline(
         enabled_uniform_buffer_masks[stage] = info->constant_buffer_mask;
         std::ranges::copy(info->constant_buffer_used_sizes, uniform_buffer_sizes[stage].begin());
     }
-    auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool] {
+    auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool, pipeline_statistics] {
         DescriptorLayoutBuilder builder{MakeBuilder(device, stage_infos)};
         uses_push_descriptor = builder.CanUsePushDescriptor();
         descriptor_set_layout = builder.CreateDescriptorSetLayout(uses_push_descriptor);
@@ -250,6 +251,9 @@ GraphicsPipeline::GraphicsPipeline(
         const VkRenderPass render_pass{render_pass_cache.Get(MakeRenderPassKey(key.state))};
         Validate();
         MakePipeline(render_pass);
+        if (pipeline_statistics) {
+            pipeline_statistics->Collect(*pipeline);
+        }
 
         std::lock_guard lock{build_mutex};
         is_built = true;
@@ -782,10 +786,14 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
         }
         */
     }
+    VkPipelineCreateFlags flags{};
+    if (device.IsKhrPipelineEexecutablePropertiesEnabled()) {
+        flags |= VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR;
+    }
     pipeline = device.GetLogical().CreateGraphicsPipeline({
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
         .pNext = nullptr,
-        .flags = 0,
+        .flags = flags,
         .stageCount = static_cast<u32>(shader_stages.size()),
         .pStages = shader_stages.data(),
         .pVertexInputState = &vertex_input_ci,

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -60,6 +60,7 @@ struct hash<Vulkan::GraphicsPipelineCacheKey> {
 namespace Vulkan {
 
 class Device;
+class PipelineStatistics;
 class RenderPassCache;
 class VKScheduler;
 class VKUpdateDescriptorQueue;
@@ -73,8 +74,9 @@ public:
         VKScheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache,
         VideoCore::ShaderNotify* shader_notify, const Device& device,
         DescriptorPool& descriptor_pool, VKUpdateDescriptorQueue& update_descriptor_queue,
-        Common::ThreadWorker* worker_thread, RenderPassCache& render_pass_cache,
-        const GraphicsPipelineCacheKey& key, std::array<vk::ShaderModule, NUM_STAGES> stages,
+        Common::ThreadWorker* worker_thread, PipelineStatistics* pipeline_statistics,
+        RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key,
+        std::array<vk::ShaderModule, NUM_STAGES> stages,
         const std::array<const Shader::Info*, NUM_STAGES>& infos);
 
     GraphicsPipeline& operator=(GraphicsPipeline&&) noexcept = delete;

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -29,6 +29,7 @@
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
 #include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/renderer_vulkan/pipeline_helper.h"
+#include "video_core/renderer_vulkan/pipeline_statistics.h"
 #include "video_core/renderer_vulkan/vk_compute_pipeline.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
@@ -389,15 +390,19 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
         size_t total{};
         size_t built{};
         bool has_loaded{};
+        std::unique_ptr<PipelineStatistics> statistics;
     } state;
 
+    if (device.IsKhrPipelineEexecutablePropertiesEnabled()) {
+        state.statistics = std::make_unique<PipelineStatistics>(device);
+    }
     const auto load_compute{[&](std::ifstream& file, FileEnvironment env) {
         ComputePipelineCacheKey key;
         file.read(reinterpret_cast<char*>(&key), sizeof(key));
 
         workers.QueueWork([this, key, env = std::move(env), &state, &callback]() mutable {
             ShaderPools pools;
-            auto pipeline{CreateComputePipeline(pools, key, env, false)};
+            auto pipeline{CreateComputePipeline(pools, key, env, state.statistics.get(), false)};
             std::lock_guard lock{state.mutex};
             if (pipeline) {
                 compute_cache.emplace(key, std::move(pipeline));
@@ -425,7 +430,8 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
             for (auto& env : envs) {
                 env_ptrs.push_back(&env);
             }
-            auto pipeline{CreateGraphicsPipeline(pools, key, MakeSpan(env_ptrs), false)};
+            auto pipeline{CreateGraphicsPipeline(pools, key, MakeSpan(env_ptrs),
+                                                 state.statistics.get(), false)};
 
             std::lock_guard lock{state.mutex};
             graphics_cache.emplace(key, std::move(pipeline));
@@ -445,6 +451,10 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
     lock.unlock();
 
     workers.WaitForRequests();
+
+    if (state.statistics) {
+        state.statistics->Report();
+    }
 }
 
 GraphicsPipeline* PipelineCache::CurrentGraphicsPipelineSlowPath() {
@@ -486,7 +496,8 @@ GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const
 
 std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
     ShaderPools& pools, const GraphicsPipelineCacheKey& key,
-    std::span<Shader::Environment* const> envs, bool build_in_parallel) try {
+    std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
+    bool build_in_parallel) try {
     LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
     size_t env_index{0};
     std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs;
@@ -540,7 +551,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
     Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
     return std::make_unique<GraphicsPipeline>(
         maxwell3d, gpu_memory, scheduler, buffer_cache, texture_cache, &shader_notify, device,
-        descriptor_pool, update_descriptor_queue, thread_worker, render_pass_cache, key,
+        descriptor_pool, update_descriptor_queue, thread_worker, statistics, render_pass_cache, key,
         std::move(modules), infos);
 
 } catch (const Shader::Exception& exception) {
@@ -553,7 +564,8 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() {
     GetGraphicsEnvironments(environments, graphics_key.unique_hashes);
 
     main_pools.ReleaseContents();
-    auto pipeline{CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(), true)};
+    auto pipeline{
+        CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(), nullptr, true)};
     if (!pipeline || pipeline_cache_filename.empty()) {
         return pipeline;
     }
@@ -578,7 +590,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
     env.SetCachedSize(shader->size_bytes);
 
     main_pools.ReleaseContents();
-    auto pipeline{CreateComputePipeline(main_pools, key, env, true)};
+    auto pipeline{CreateComputePipeline(main_pools, key, env, nullptr, true)};
     if (!pipeline || pipeline_cache_filename.empty()) {
         return pipeline;
     }
@@ -591,7 +603,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
 
 std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
     ShaderPools& pools, const ComputePipelineCacheKey& key, Shader::Environment& env,
-    bool build_in_parallel) try {
+    PipelineStatistics* statistics, bool build_in_parallel) try {
     LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
 
     Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
@@ -605,8 +617,8 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
     }
     Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
     return std::make_unique<ComputePipeline>(device, descriptor_pool, update_descriptor_queue,
-                                             thread_worker, &shader_notify, program.info,
-                                             std::move(spv_module));
+                                             thread_worker, statistics, &shader_notify,
+                                             program.info, std::move(spv_module));
 
 } catch (const Shader::Exception& exception) {
     LOG_ERROR(Render_Vulkan, "{}", exception.what());

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -80,8 +80,9 @@ struct hash<Vulkan::ComputePipelineCacheKey> {
 namespace Vulkan {
 
 class ComputePipeline;
-class Device;
 class DescriptorPool;
+class Device;
+class PipelineStatistics;
 class RasterizerVulkan;
 class RenderPassCache;
 class VKScheduler;
@@ -128,7 +129,8 @@ private:
 
     std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(
         ShaderPools& pools, const GraphicsPipelineCacheKey& key,
-        std::span<Shader::Environment* const> envs, bool build_in_parallel);
+        std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
+        bool build_in_parallel);
 
     std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineCacheKey& key,
                                                            const ShaderInfo* shader);
@@ -136,6 +138,7 @@ private:
     std::unique_ptr<ComputePipeline> CreateComputePipeline(ShaderPools& pools,
                                                            const ComputePipelineCacheKey& key,
                                                            Shader::Environment& env,
+                                                           PipelineStatistics* statistics,
                                                            bool build_in_parallel);
 
     const Device& device;

src/video_core/textures/astc.cpp

@@ -151,6 +151,76 @@ private:
     const IntType& m_Bits;
 };
 
+enum class IntegerEncoding { JustBits, Quint, Trit };
+
+struct IntegerEncodedValue {
+    constexpr IntegerEncodedValue() = default;
+
+    constexpr IntegerEncodedValue(IntegerEncoding encoding_, u32 num_bits_)
+        : encoding{encoding_}, num_bits{num_bits_} {}
+
+    constexpr bool MatchesEncoding(const IntegerEncodedValue& other) const {
+        return encoding == other.encoding && num_bits == other.num_bits;
+    }
+
+    // Returns the number of bits required to encode num_vals values.
+    u32 GetBitLength(u32 num_vals) const {
+        u32 total_bits = num_bits * num_vals;
+        if (encoding == IntegerEncoding::Trit) {
+            total_bits += (num_vals * 8 + 4) / 5;
+        } else if (encoding == IntegerEncoding::Quint) {
+            total_bits += (num_vals * 7 + 2) / 3;
+        }
+        return total_bits;
+    }
+
+    IntegerEncoding encoding{};
+    u32 num_bits = 0;
+    u32 bit_value = 0;
+    union {
+        u32 quint_value = 0;
+        u32 trit_value;
+    };
+};
+
+// Returns a new instance of this struct that corresponds to the
+// can take no more than mav_value values
+static constexpr IntegerEncodedValue CreateEncoding(u32 mav_value) {
+    while (mav_value > 0) {
+        u32 check = mav_value + 1;
+
+        // Is mav_value a power of two?
+        if (!(check & (check - 1))) {
+            return IntegerEncodedValue(IntegerEncoding::JustBits, std::popcount(mav_value));
+        }
+
+        // Is mav_value of the type 3*2^n - 1?
+        if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) {
+            return IntegerEncodedValue(IntegerEncoding::Trit, std::popcount(check / 3 - 1));
+        }
+
+        // Is mav_value of the type 5*2^n - 1?
+        if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) {
+            return IntegerEncodedValue(IntegerEncoding::Quint, std::popcount(check / 5 - 1));
+        }
+
+        // Apparently it can't be represented with a bounded integer sequence...
+        // just iterate.
+        mav_value--;
+    }
+    return IntegerEncodedValue(IntegerEncoding::JustBits, 0);
+}
+
+static constexpr std::array<IntegerEncodedValue, 256> MakeEncodedValues() {
+    std::array<IntegerEncodedValue, 256> encodings{};
+    for (std::size_t i = 0; i < encodings.size(); ++i) {
+        encodings[i] = CreateEncoding(static_cast<u32>(i));
+    }
+    return encodings;
+}
+
+static constexpr std::array<IntegerEncodedValue, 256> ASTC_ENCODINGS_VALUES = MakeEncodedValues();
+
 namespace Tegra::Texture::ASTC {
 using IntegerEncodedVector = boost::container::static_vector<
     IntegerEncodedValue, 256,
@@ -521,35 +591,41 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) {
     return params;
 }
 
-static void FillVoidExtentLDR(InputBitStream& strm, std::span<u32> outBuf, u32 blockWidth,
-                              u32 blockHeight) {
-    // Don't actually care about the void extent, just read the bits...
-    for (s32 i = 0; i < 4; ++i) {
-        strm.ReadBits<13>();
+// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
+// is the same as [(num_bits - 1):0] and repeats all the way down.
+template <typename IntType>
+static constexpr IntType Replicate(IntType val, u32 num_bits, u32 to_bit) {
+    if (num_bits == 0 || to_bit == 0) {
+        return 0;
     }
-
-    // Decode the RGBA components and renormalize them to the range [0, 255]
-    u16 r = static_cast<u16>(strm.ReadBits<16>());
-    u16 g = static_cast<u16>(strm.ReadBits<16>());
-    u16 b = static_cast<u16>(strm.ReadBits<16>());
-    u16 a = static_cast<u16>(strm.ReadBits<16>());
-
-    u32 rgba = (r >> 8) | (g & 0xFF00) | (static_cast<u32>(b) & 0xFF00) << 8 |
-               (static_cast<u32>(a) & 0xFF00) << 16;
-
-    for (u32 j = 0; j < blockHeight; j++) {
-        for (u32 i = 0; i < blockWidth; i++) {
-            outBuf[j * blockWidth + i] = rgba;
+    const IntType v = val & static_cast<IntType>((1 << num_bits) - 1);
+    IntType res = v;
+    u32 reslen = num_bits;
+    while (reslen < to_bit) {
+        u32 comp = 0;
+        if (num_bits > to_bit - reslen) {
+            u32 newshift = to_bit - reslen;
+            comp = num_bits - newshift;
+            num_bits = newshift;
         }
+        res = static_cast<IntType>(res << num_bits);
+        res = static_cast<IntType>(res | (v >> comp));
+        reslen += num_bits;
     }
+    return res;
 }
 
-static void FillError(std::span<u32> outBuf, u32 blockWidth, u32 blockHeight) {
-    for (u32 j = 0; j < blockHeight; j++) {
-        for (u32 i = 0; i < blockWidth; i++) {
-            outBuf[j * blockWidth + i] = 0xFFFF00FF;
-        }
+static constexpr std::size_t NumReplicateEntries(u32 num_bits) {
+    return std::size_t(1) << num_bits;
+}
+
+template <typename IntType, u32 num_bits, u32 to_bit>
+static constexpr auto MakeReplicateTable() {
+    std::array<IntType, NumReplicateEntries(num_bits)> table{};
+    for (IntType value = 0; value < static_cast<IntType>(std::size(table)); ++value) {
+        table[value] = Replicate(value, num_bits, to_bit);
     }
+    return table;
 }
 
 static constexpr auto REPLICATE_BYTE_TO_16_TABLE = MakeReplicateTable<u32, 8, 16>();
@@ -572,6 +648,9 @@ static constexpr auto REPLICATE_2_BIT_TO_8_TABLE = MakeReplicateTable<u32, 2, 8>
 static constexpr auto REPLICATE_3_BIT_TO_8_TABLE = MakeReplicateTable<u32, 3, 8>();
 static constexpr auto REPLICATE_4_BIT_TO_8_TABLE = MakeReplicateTable<u32, 4, 8>();
 static constexpr auto REPLICATE_5_BIT_TO_8_TABLE = MakeReplicateTable<u32, 5, 8>();
+static constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
+static constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
+static constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
 /// Use a precompiled table with the most common usages, if it's not in the expected range, fallback
 /// to the runtime implementation
 static constexpr u32 FastReplicateTo8(u32 value, u32 num_bits) {
@@ -1316,6 +1395,37 @@ static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const u32*& colorValues,
 #undef READ_INT_VALUES
 }
 
+static void FillVoidExtentLDR(InputBitStream& strm, std::span<u32> outBuf, u32 blockWidth,
+                              u32 blockHeight) {
+    // Don't actually care about the void extent, just read the bits...
+    for (s32 i = 0; i < 4; ++i) {
+        strm.ReadBits<13>();
+    }
+
+    // Decode the RGBA components and renormalize them to the range [0, 255]
+    u16 r = static_cast<u16>(strm.ReadBits<16>());
+    u16 g = static_cast<u16>(strm.ReadBits<16>());
+    u16 b = static_cast<u16>(strm.ReadBits<16>());
+    u16 a = static_cast<u16>(strm.ReadBits<16>());
+
+    u32 rgba = (r >> 8) | (g & 0xFF00) | (static_cast<u32>(b) & 0xFF00) << 8 |
+               (static_cast<u32>(a) & 0xFF00) << 16;
+
+    for (u32 j = 0; j < blockHeight; j++) {
+        for (u32 i = 0; i < blockWidth; i++) {
+            outBuf[j * blockWidth + i] = rgba;
+        }
+    }
+}
+
+static void FillError(std::span<u32> outBuf, u32 blockWidth, u32 blockHeight) {
+    for (u32 j = 0; j < blockHeight; j++) {
+        for (u32 i = 0; i < blockWidth; i++) {
+            outBuf[j * blockWidth + i] = 0xFFFF00FF;
+        }
+    }
+}
+
 static void DecompressBlock(std::span<const u8, 16> inBuf, const u32 blockWidth,
                             const u32 blockHeight, std::span<u32, 12 * 12> outBuf) {
     InputBitStream strm(inBuf);

src/video_core/textures/astc.h

@@ -9,117 +9,6 @@
 
 namespace Tegra::Texture::ASTC {
 
-enum class IntegerEncoding { JustBits, Quint, Trit };
-
-struct IntegerEncodedValue {
-    constexpr IntegerEncodedValue() = default;
-
-    constexpr IntegerEncodedValue(IntegerEncoding encoding_, u32 num_bits_)
-        : encoding{encoding_}, num_bits{num_bits_} {}
-
-    constexpr bool MatchesEncoding(const IntegerEncodedValue& other) const {
-        return encoding == other.encoding && num_bits == other.num_bits;
-    }
-
-    // Returns the number of bits required to encode num_vals values.
-    u32 GetBitLength(u32 num_vals) const {
-        u32 total_bits = num_bits * num_vals;
-        if (encoding == IntegerEncoding::Trit) {
-            total_bits += (num_vals * 8 + 4) / 5;
-        } else if (encoding == IntegerEncoding::Quint) {
-            total_bits += (num_vals * 7 + 2) / 3;
-        }
-        return total_bits;
-    }
-
-    IntegerEncoding encoding{};
-    u32 num_bits = 0;
-    u32 bit_value = 0;
-    union {
-        u32 quint_value = 0;
-        u32 trit_value;
-    };
-};
-
-// Returns a new instance of this struct that corresponds to the
-// can take no more than mav_value values
-constexpr IntegerEncodedValue CreateEncoding(u32 mav_value) {
-    while (mav_value > 0) {
-        u32 check = mav_value + 1;
-
-        // Is mav_value a power of two?
-        if (!(check & (check - 1))) {
-            return IntegerEncodedValue(IntegerEncoding::JustBits, std::popcount(mav_value));
-        }
-
-        // Is mav_value of the type 3*2^n - 1?
-        if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) {
-            return IntegerEncodedValue(IntegerEncoding::Trit, std::popcount(check / 3 - 1));
-        }
-
-        // Is mav_value of the type 5*2^n - 1?
-        if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) {
-            return IntegerEncodedValue(IntegerEncoding::Quint, std::popcount(check / 5 - 1));
-        }
-
-        // Apparently it can't be represented with a bounded integer sequence...
-        // just iterate.
-        mav_value--;
-    }
-    return IntegerEncodedValue(IntegerEncoding::JustBits, 0);
-}
-
-constexpr std::array<IntegerEncodedValue, 256> MakeEncodedValues() {
-    std::array<IntegerEncodedValue, 256> encodings{};
-    for (std::size_t i = 0; i < encodings.size(); ++i) {
-        encodings[i] = CreateEncoding(static_cast<u32>(i));
-    }
-    return encodings;
-}
-
-constexpr std::array<IntegerEncodedValue, 256> ASTC_ENCODINGS_VALUES = MakeEncodedValues();
-
-// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
-// is the same as [(num_bits - 1):0] and repeats all the way down.
-template <typename IntType>
-constexpr IntType Replicate(IntType val, u32 num_bits, u32 to_bit) {
-    if (num_bits == 0 || to_bit == 0) {
-        return 0;
-    }
-    const IntType v = val & static_cast<IntType>((1 << num_bits) - 1);
-    IntType res = v;
-    u32 reslen = num_bits;
-    while (reslen < to_bit) {
-        u32 comp = 0;
-        if (num_bits > to_bit - reslen) {
-            u32 newshift = to_bit - reslen;
-            comp = num_bits - newshift;
-            num_bits = newshift;
-        }
-        res = static_cast<IntType>(res << num_bits);
-        res = static_cast<IntType>(res | (v >> comp));
-        reslen += num_bits;
-    }
-    return res;
-}
-
-constexpr std::size_t NumReplicateEntries(u32 num_bits) {
-    return std::size_t(1) << num_bits;
-}
-
-template <typename IntType, u32 num_bits, u32 to_bit>
-constexpr auto MakeReplicateTable() {
-    std::array<IntType, NumReplicateEntries(num_bits)> table{};
-    for (IntType value = 0; value < static_cast<IntType>(std::size(table)); ++value) {
-        table[value] = Replicate(value, num_bits, to_bit);
-    }
-    return table;
-}
-
-constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
-constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
-constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
-
 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);
 

src/video_core/vulkan_common/vulkan_device.cpp

@@ -526,6 +526,17 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
         SetNext(next, workgroup_layout);
     }
 
+    VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR executable_properties;
+    if (khr_pipeline_executable_properties) {
+        LOG_INFO(Render_Vulkan, "Enabling shader feedback, expect slower shader build times");
+        executable_properties = {
+            .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PIPELINE_EXECUTABLE_PROPERTIES_FEATURES_KHR,
+            .pNext = nullptr,
+            .pipelineExecutableInfo = VK_TRUE,
+        };
+        SetNext(next, executable_properties);
+    }
+
     if (!ext_depth_range_unrestricted) {
         LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted");
     }
@@ -824,6 +835,7 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) {
 
     bool has_khr_shader_float16_int8{};
     bool has_khr_workgroup_memory_explicit_layout{};
+    bool has_khr_pipeline_executable_properties{};
     bool has_ext_subgroup_size_control{};
     bool has_ext_transform_feedback{};
     bool has_ext_custom_border_color{};
@@ -878,6 +890,10 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) {
             test(nv_device_diagnostics_config, VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME,
                  true);
         }
+        if (Settings::values.renderer_shader_feedback) {
+            test(has_khr_pipeline_executable_properties,
+                 VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME, false);
+        }
     }
     VkPhysicalDeviceFeatures2KHR features{};
     features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2_KHR;
@@ -1033,6 +1049,19 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) {
             khr_workgroup_memory_explicit_layout = true;
         }
     }
+    if (has_khr_pipeline_executable_properties) {
+        VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR executable_properties;
+        executable_properties.sType =
+            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PIPELINE_EXECUTABLE_PROPERTIES_FEATURES_KHR;
+        executable_properties.pNext = nullptr;
+        features.pNext = &executable_properties;
+        physical.GetFeatures2KHR(features);
+
+        if (executable_properties.pipelineExecutableInfo) {
+            extensions.push_back(VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME);
+            khr_pipeline_executable_properties = true;
+        }
+    }
     if (khr_push_descriptor) {
         VkPhysicalDevicePushDescriptorPropertiesKHR push_descriptor;
         push_descriptor.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR;

src/video_core/vulkan_common/vulkan_device.h

@@ -214,6 +214,11 @@ public:
         return khr_push_descriptor;
     }
 
+    /// Returns true if VK_KHR_pipeline_executable_properties is enabled.
+    bool IsKhrPipelineEexecutablePropertiesEnabled() const {
+        return khr_pipeline_executable_properties;
+    }
+
     /// Returns true if the device supports VK_KHR_workgroup_memory_explicit_layout.
     bool IsKhrWorkgroupMemoryExplicitLayoutSupported() const {
         return khr_workgroup_memory_explicit_layout;
@@ -378,6 +383,7 @@ private:
     bool khr_spirv_1_4{};                       ///< Support for VK_KHR_spirv_1_4.
     bool khr_workgroup_memory_explicit_layout{}; ///< Support for explicit workgroup layouts.
     bool khr_push_descriptor{};                  ///< Support for VK_KHR_push_descritor.
+    bool khr_pipeline_executable_properties{};   ///< Support for executable properties.
     bool ext_index_type_uint8{};                 ///< Support for VK_EXT_index_type_uint8.
     bool ext_sampler_filter_minmax{};            ///< Support for VK_EXT_sampler_filter_minmax.
     bool ext_depth_range_unrestricted{};         ///< Support for VK_EXT_depth_range_unrestricted.

src/video_core/vulkan_common/vulkan_wrapper.cpp

@@ -181,6 +181,8 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkGetMemoryWin32HandleKHR);
 #endif
     X(vkGetQueryPoolResults);
+    X(vkGetPipelineExecutablePropertiesKHR);
+    X(vkGetPipelineExecutableStatisticsKHR);
     X(vkGetSemaphoreCounterValueKHR);
     X(vkMapMemory);
     X(vkQueueSubmit);
@@ -809,6 +811,42 @@ VkMemoryRequirements Device::GetImageMemoryRequirements(VkImage image) const noe
     return requirements;
 }
 
+std::vector<VkPipelineExecutablePropertiesKHR> Device::GetPipelineExecutablePropertiesKHR(
+    VkPipeline pipeline) const {
+    const VkPipelineInfoKHR info{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_INFO_KHR,
+        .pNext = nullptr,
+        .pipeline = pipeline,
+    };
+    u32 num{};
+    dld->vkGetPipelineExecutablePropertiesKHR(handle, &info, &num, nullptr);
+    std::vector<VkPipelineExecutablePropertiesKHR> properties(num);
+    for (auto& property : properties) {
+        property.sType = VK_STRUCTURE_TYPE_PIPELINE_EXECUTABLE_PROPERTIES_KHR;
+    }
+    Check(dld->vkGetPipelineExecutablePropertiesKHR(handle, &info, &num, properties.data()));
+    return properties;
+}
+
+std::vector<VkPipelineExecutableStatisticKHR> Device::GetPipelineExecutableStatisticsKHR(
+    VkPipeline pipeline, u32 executable_index) const {
+    const VkPipelineExecutableInfoKHR executable_info{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_EXECUTABLE_INFO_KHR,
+        .pNext = nullptr,
+        .pipeline = pipeline,
+        .executableIndex = executable_index,
+    };
+    u32 num{};
+    dld->vkGetPipelineExecutableStatisticsKHR(handle, &executable_info, &num, nullptr);
+    std::vector<VkPipelineExecutableStatisticKHR> statistics(num);
+    for (auto& statistic : statistics) {
+        statistic.sType = VK_STRUCTURE_TYPE_PIPELINE_EXECUTABLE_STATISTIC_KHR;
+    }
+    Check(dld->vkGetPipelineExecutableStatisticsKHR(handle, &executable_info, &num,
+                                                    statistics.data()));
+    return statistics;
+}
+
 void Device::UpdateDescriptorSets(Span<VkWriteDescriptorSet> writes,
                                   Span<VkCopyDescriptorSet> copies) const noexcept {
     dld->vkUpdateDescriptorSets(handle, writes.size(), writes.data(), copies.size(), copies.data());

src/video_core/vulkan_common/vulkan_wrapper.h

@@ -295,6 +295,8 @@ struct DeviceDispatch : InstanceDispatch {
 #ifdef _WIN32
     PFN_vkGetMemoryWin32HandleKHR vkGetMemoryWin32HandleKHR{};
 #endif
+    PFN_vkGetPipelineExecutablePropertiesKHR vkGetPipelineExecutablePropertiesKHR{};
+    PFN_vkGetPipelineExecutableStatisticsKHR vkGetPipelineExecutableStatisticsKHR{};
     PFN_vkGetQueryPoolResults vkGetQueryPoolResults{};
     PFN_vkGetSemaphoreCounterValueKHR vkGetSemaphoreCounterValueKHR{};
     PFN_vkMapMemory vkMapMemory{};
@@ -879,6 +881,12 @@ public:
 
     VkMemoryRequirements GetImageMemoryRequirements(VkImage image) const noexcept;
 
+    std::vector<VkPipelineExecutablePropertiesKHR> GetPipelineExecutablePropertiesKHR(
+        VkPipeline pipeline) const;
+
+    std::vector<VkPipelineExecutableStatisticKHR> GetPipelineExecutableStatisticsKHR(
+        VkPipeline pipeline, u32 executable_index) const;
+
     void UpdateDescriptorSets(Span<VkWriteDescriptorSet> writes,
                               Span<VkCopyDescriptorSet> copies) const noexcept;
 

src/yuzu/configuration/config.cpp

@@ -825,6 +825,7 @@ void Config::ReadRendererValues() {
     if (global) {
         ReadBasicSetting(Settings::values.fps_cap);
         ReadBasicSetting(Settings::values.renderer_debug);
+        ReadBasicSetting(Settings::values.renderer_shader_feedback);
         ReadBasicSetting(Settings::values.enable_nsight_aftermath);
         ReadBasicSetting(Settings::values.disable_shader_loop_safety_checks);
     }
@@ -1363,6 +1364,7 @@ void Config::SaveRendererValues() {
     if (global) {
         WriteBasicSetting(Settings::values.fps_cap);
         WriteBasicSetting(Settings::values.renderer_debug);
+        WriteBasicSetting(Settings::values.renderer_shader_feedback);
         WriteBasicSetting(Settings::values.enable_nsight_aftermath);
         WriteBasicSetting(Settings::values.disable_shader_loop_safety_checks);
     }

src/yuzu/configuration/configure_debug.cpp

@@ -43,6 +43,8 @@ void ConfigureDebug::SetConfiguration() {
     ui->use_auto_stub->setChecked(Settings::values.use_auto_stub.GetValue());
     ui->enable_graphics_debugging->setEnabled(runtime_lock);
     ui->enable_graphics_debugging->setChecked(Settings::values.renderer_debug.GetValue());
+    ui->enable_shader_feedback->setEnabled(runtime_lock);
+    ui->enable_shader_feedback->setChecked(Settings::values.renderer_shader_feedback.GetValue());
     ui->enable_cpu_debugging->setEnabled(runtime_lock);
     ui->enable_cpu_debugging->setChecked(Settings::values.cpu_debug_mode.GetValue());
     ui->enable_nsight_aftermath->setEnabled(runtime_lock);
@@ -65,6 +67,7 @@ void ConfigureDebug::ApplyConfiguration() {
     Settings::values.use_debug_asserts = ui->use_debug_asserts->isChecked();
     Settings::values.use_auto_stub = ui->use_auto_stub->isChecked();
     Settings::values.renderer_debug = ui->enable_graphics_debugging->isChecked();
+    Settings::values.renderer_shader_feedback = ui->enable_shader_feedback->isChecked();
     Settings::values.cpu_debug_mode = ui->enable_cpu_debugging->isChecked();
     Settings::values.enable_nsight_aftermath = ui->enable_nsight_aftermath->isChecked();
     Settings::values.disable_shader_loop_safety_checks =

src/yuzu/configuration/configure_debug.ui

@@ -111,8 +111,8 @@
      <property name="title">
       <string>Graphics</string>
      </property>
-     <layout class="QVBoxLayout" name="verticalLayout_6">
-      <item>
+     <layout class="QGridLayout" name="gridLayout_3">
+      <item row="0" column="0">
        <widget class="QCheckBox" name="enable_graphics_debugging">
         <property name="enabled">
          <bool>true</bool>
@@ -125,7 +125,7 @@
         </property>
        </widget>
       </item>
-      <item>
+      <item row="2" column="0">
        <widget class="QCheckBox" name="enable_nsight_aftermath">
         <property name="toolTip">
          <string>When checked, it enables Nsight Aftermath crash dumps</string>
@@ -135,7 +135,7 @@
         </property>
        </widget>
       </item>
-      <item>
+      <item row="0" column="1">
        <widget class="QCheckBox" name="disable_macro_jit">
         <property name="enabled">
          <bool>true</bool>
@@ -148,7 +148,17 @@
         </property>
        </widget>
       </item>
-      <item>
+      <item row="1" column="0">
+       <widget class="QCheckBox" name="enable_shader_feedback">
+        <property name="toolTip">
+         <string>When checked, yuzu will log statistics about the compiled pipeline cache</string>
+        </property>
+        <property name="text">
+         <string>Enable Shader Feedback</string>
+        </property>
+       </widget>
+      </item>
+      <item row="1" column="1">
        <widget class="QCheckBox" name="disable_loop_safety_checks">
         <property name="toolTip">
          <string>When checked, it executes shaders without loop logic changes</string>
@@ -276,11 +286,14 @@
   <tabstop>open_log_button</tabstop>
   <tabstop>homebrew_args_edit</tabstop>
   <tabstop>enable_graphics_debugging</tabstop>
+  <tabstop>enable_shader_feedback</tabstop>
   <tabstop>enable_nsight_aftermath</tabstop>
   <tabstop>disable_macro_jit</tabstop>
   <tabstop>disable_loop_safety_checks</tabstop>
+  <tabstop>fs_access_log</tabstop>
   <tabstop>reporting_services</tabstop>
   <tabstop>quest_flag</tabstop>
+  <tabstop>enable_cpu_debugging</tabstop>
   <tabstop>use_debug_asserts</tabstop>
   <tabstop>use_auto_stub</tabstop>
  </tabstops>

src/yuzu/configuration/configure_system.ui

@@ -401,6 +401,11 @@
               <string>Traditional Chinese (正體中文)</string>
              </property>
             </item>
+            <item>
+             <property name="text">
+              <string>Brazilian Portuguese (português do Brasil)</string>
+             </property>
+            </item>
            </widget>
           </item>
           <item row="5" column="0">

src/yuzu_cmd/config.cpp

@@ -444,6 +444,7 @@ void Config::ReadValues() {
     // Renderer
     ReadSetting("Renderer", Settings::values.renderer_backend);
     ReadSetting("Renderer", Settings::values.renderer_debug);
+    ReadSetting("Renderer", Settings::values.renderer_shader_feedback);
     ReadSetting("Renderer", Settings::values.enable_nsight_aftermath);
     ReadSetting("Renderer", Settings::values.disable_shader_loop_safety_checks);
     ReadSetting("Renderer", Settings::values.vulkan_device);

src/yuzu_cmd/default_ini.h

@@ -221,6 +221,10 @@ backend =
 # 0 (default): Disabled, 1: Enabled
 debug =
 
+# Enable shader feedback.
+# 0 (default): Disabled, 1: Enabled
+renderer_shader_feedback =
+
 # Enable Nsight Aftermath crash dumps
 # 0 (default): Disabled, 1: Enabled
 nsight_aftermath =
@@ -363,7 +367,7 @@ custom_rtc =
 # Sets the systems language index
 # 0: Japanese, 1: English (default), 2: French, 3: German, 4: Italian, 5: Spanish, 6: Chinese,
 # 7: Korean, 8: Dutch, 9: Portuguese, 10: Russian, 11: Taiwanese, 12: British English, 13: Canadian French,
-# 14: Latin American Spanish, 15: Simplified Chinese, 16: Traditional Chinese
+# 14: Latin American Spanish, 15: Simplified Chinese, 16: Traditional Chinese, 17: Brazilian Portuguese
 language_index =
 
 # The system region that yuzu will use during emulation