added 64bpp macro detiler

src/video_core/amdgpu/resource.h

@@ -141,6 +141,8 @@ constexpr std::string_view NameOf(TilingMode type) {
         return "Texture_MicroTiled";
     case TilingMode::Texture_MacroTiled:
         return "Texture_MacroTiled";
+    case TilingMode::Texture_Volume:
+        return "Texture_Volume";
     default:
         return "Unknown";
     }

src/video_core/host_shaders/CMakeLists.txt

@@ -8,6 +8,7 @@ set(SHADER_FILES
     detile_m32x2.comp
     detile_m32x4.comp
     detile_macro32x1.comp
+    detile_macro32x2.comp
     fs_tri.vert
     post_process.frag
 )

src/video_core/host_shaders/detile_m32x1.comp

@@ -16,7 +16,6 @@ layout(push_constant) uniform image_info {
     uint num_levels;
     uint pitch;
     uint height;
-    uint depth;
     uint sizes[14];
 } info;
 

src/video_core/host_shaders/detile_m32x2.comp

@@ -16,7 +16,6 @@ layout(push_constant) uniform image_info {
     uint num_levels;
     uint pitch;
     uint height;
-    uint depth;
     uint sizes[14];
 } info;
 

src/video_core/host_shaders/detile_m32x4.comp

@@ -16,7 +16,6 @@ layout(push_constant) uniform image_info {
     uint num_levels;
     uint pitch;
     uint height;
-    uint depth;
     uint sizes[14];
 } info;
 

src/video_core/host_shaders/detile_m8x1.comp

@@ -19,7 +19,6 @@ layout(push_constant) uniform image_info {
     uint num_levels;
     uint pitch;
     uint height;
-    uint depth;
     uint sizes[14];
 } info;
 

src/video_core/host_shaders/detile_m8x2.comp

@@ -18,7 +18,6 @@ layout(push_constant) uniform image_info {
     uint num_levels;
     uint pitch;
     uint height;
-    uint depth;
     uint sizes[14];
 } info;
 

src/video_core/host_shaders/detile_macro32x1.comp

@@ -16,8 +16,8 @@ layout(push_constant) uniform image_info {
     uint num_levels;
     uint pitch;
     uint height;
-    uint depth;
-    uint sizes[14];
+    uint c0;
+    uint c1;
 } info;
 
 // Each LUT is 64 bytes, so should fit into K$ given tiled slices locality
@@ -78,14 +78,13 @@ void main() {
     uint row = bitfieldExtract(y, 0, 3);
     uint lut = bitfieldExtract(z, 0, 2);
     uint idx = lut_32bpp[lut][col + row * MICRO_TILE_DIM];
-    uint elem_offs = idx * BPP;
 
-    uint slice_offs = (z / 4) * MICRO_TILE_DIM * MICRO_TILE_DIM * MICRO_TILE_SZ; 
+    uint slice_offs = (z >> 2u) * info.c1 * MICRO_TILE_SZ; 
     uint tile_row = y / MICRO_TILE_DIM;
     uint tile_column = x / MICRO_TILE_DIM;
-    uint tile_offs = ((tile_row * MICRO_TILE_DIM) + tile_column) * MICRO_TILE_SZ;
-    uint offs = ((slice_offs + tile_offs) * MICRO_TILE_DIM + elem_offs) / MICRO_TILE_DIM;
+    uint tile_offs = ((tile_row * info.c0) + tile_column) * MICRO_TILE_SZ;
+    uint offs = slice_offs + tile_offs + (idx * BPP / 8);
 
-    uint p0 = in_data[offs / (BPP / 8)];
+    uint p0 = in_data[offs >> 2u];
     out_data[gl_GlobalInvocationID.x] = p0;
 }

src/video_core/host_shaders/detile_macro32x2.comp

@@ -0,0 +1,91 @@
+// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#version 450
+
+layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
+
+layout(std430, binding = 0) buffer input_buf {
+    uint in_data[];
+};
+layout(std430, binding = 1) buffer output_buf {
+    uint out_data[];
+};
+
+layout(push_constant) uniform image_info {
+    uint num_levels;
+    uint pitch;
+    uint height;
+    uint c0;
+    uint c1;
+} info;
+
+const uint lut_64bpp[][64] = {
+    {
+        0x00, 0x01, 0x08, 0x09, 0x40, 0x41, 0x48, 0x49,
+        0x02, 0x03, 0x0a, 0x0b, 0x42, 0x43, 0x4a, 0x4b,
+        0x10, 0x11, 0x18, 0x19, 0x50, 0x51, 0x58, 0x59,
+        0x12, 0x13, 0x1a, 0x1b, 0x52, 0x53, 0x5a, 0x5b,
+        0x80, 0x81, 0x88, 0x89, 0xc0, 0xc1, 0xc8, 0xc9,
+        0x82, 0x83, 0x8a, 0x8b, 0xc2, 0xc3, 0xca, 0xcb,
+        0x90, 0x91, 0x98, 0x99, 0xd0, 0xd1, 0xd8, 0xd9,
+        0x92, 0x93, 0x9a, 0x9b, 0xd2, 0xd3, 0xda, 0xdb,
+    },
+    {
+        0x04, 0x05, 0x0c, 0x0d, 0x44, 0x45, 0x4c, 0x4d,
+        0x06, 0x07, 0x0e, 0x0f, 0x46, 0x47, 0x4e, 0x4f,
+        0x14, 0x15, 0x1c, 0x1d, 0x54, 0x55, 0x5c, 0x5d,
+        0x16, 0x17, 0x1e, 0x1f, 0x56, 0x57, 0x5e, 0x5f,
+        0x84, 0x85, 0x8c, 0x8d, 0xc4, 0xc5, 0xcc, 0xcd,
+        0x86, 0x87, 0x8e, 0x8f, 0xc6, 0xc7, 0xce, 0xcf,
+        0x94, 0x95, 0x9c, 0x9d, 0xd4, 0xd5, 0xdc, 0xdd,
+        0x96, 0x97, 0x9e, 0x9f, 0xd6, 0xd7, 0xde, 0xdf,
+    },
+    {
+        0x20, 0x21, 0x28, 0x29, 0x60, 0x61, 0x68, 0x69,
+        0x22, 0x23, 0x2a, 0x2b, 0x62, 0x63, 0x6a, 0x6b,
+        0x30, 0x31, 0x38, 0x39, 0x70, 0x71, 0x78, 0x79,
+        0x32, 0x33, 0x3a, 0x3b, 0x72, 0x73, 0x7a, 0x7b,
+        0xa0, 0xa1, 0xa8, 0xa9, 0xe0, 0xe1, 0xe8, 0xe9,
+        0xa2, 0xa3, 0xaa, 0xab, 0xe2, 0xe3, 0xea, 0xeb,
+        0xb0, 0xb1, 0xb8, 0xb9, 0xf0, 0xf1, 0xf8, 0xf9,
+        0xb2, 0xb3, 0xba, 0xbb, 0xf2, 0xf3, 0xfa, 0xfb,
+    },
+    {
+        0x24, 0x25, 0x2c, 0x2d, 0x64, 0x65, 0x6c, 0x6d,
+        0x26, 0x27, 0x2e, 0x2f, 0x66, 0x67, 0x6e, 0x6f,
+        0x34, 0x35, 0x3c, 0x3d, 0x74, 0x75, 0x7c, 0x7d,
+        0x36, 0x37, 0x3e, 0x3f, 0x76, 0x77, 0x7e, 0x7f,
+        0xa4, 0xa5, 0xac, 0xad, 0xe4, 0xe5, 0xec, 0xed,
+        0xa6, 0xa7, 0xae, 0xaf, 0xe6, 0xe7, 0xee, 0xef,
+        0xb4, 0xb5, 0xbc, 0xbd, 0xf4, 0xf5, 0xfc, 0xfd,
+        0xb6, 0xb7, 0xbe, 0xbf, 0xf6, 0xf7, 0xfe, 0xff,
+    },
+};
+
+#define MICRO_TILE_DIM      (8)
+#define MICRO_TILE_SZ       (2048)
+#define TEXELS_PER_ELEMENT  (1)
+#define BPP                 (64)
+
+void main() {
+    uint x =  gl_GlobalInvocationID.x % info.pitch;
+    uint y = (gl_GlobalInvocationID.x / info.pitch) % info.height;
+    uint z =  gl_GlobalInvocationID.x / (info.pitch * info.height);
+
+    uint col = bitfieldExtract(x, 0, 3);
+    uint row = bitfieldExtract(y, 0, 3);
+    uint lut = bitfieldExtract(z, 0, 2);
+    uint idx = lut_64bpp[lut][col + row * MICRO_TILE_DIM];
+
+    uint slice_offs = (z >> 2u) * info.c1 * MICRO_TILE_SZ; 
+    uint tile_row = y / MICRO_TILE_DIM;
+    uint tile_column = x / MICRO_TILE_DIM;
+    uint tile_offs = ((tile_row * info.c0) + tile_column) * MICRO_TILE_SZ;
+    uint offs = slice_offs + tile_offs + (idx * BPP / 8);
+
+    uint p0 = in_data[(offs >> 2) + 0];
+    uint p1 = in_data[(offs >> 2) + 1];    
+    out_data[2 * gl_GlobalInvocationID.x + 0] = p0;
+    out_data[2 * gl_GlobalInvocationID.x + 1] = p1;    
+}

src/video_core/texture_cache/tile_manager.cpp

@@ -13,6 +13,7 @@
 #include "video_core/host_shaders/detile_m8x1_comp.h"
 #include "video_core/host_shaders/detile_m8x2_comp.h"
 #include "video_core/host_shaders/detile_macro32x1_comp.h"
+#include "video_core/host_shaders/detile_macro32x2_comp.h"
 
 #include <boost/container/static_vector.hpp>
 #include <magic_enum/magic_enum.hpp>
@@ -20,158 +21,7 @@
 
 namespace VideoCore {
 
-class TileManager32 {
-public:
-    u32 m_macro_tile_height = 0;
-    u32 m_bank_height = 0;
-    u32 m_num_banks = 0;
-    u32 m_num_pipes = 0;
-    u32 m_padded_width = 0;
-    u32 m_padded_height = 0;
-    u32 m_pipe_bits = 0;
-    u32 m_bank_bits = 0;
-
-    void Init(u32 width, u32 height, bool is_neo) {
-        m_macro_tile_height = (is_neo ? 128 : 64);
-        m_bank_height = is_neo ? 2 : 1;
-        m_num_banks = is_neo ? 8 : 16;
-        m_num_pipes = is_neo ? 16 : 8;
-        m_padded_width = width;
-        if (height == 1080) {
-            m_padded_height = is_neo ? 1152 : 1088;
-        }
-        if (height == 720) {
-            m_padded_height = 768;
-        }
-        m_pipe_bits = is_neo ? 4 : 3;
-        m_bank_bits = is_neo ? 3 : 4;
-    }
-
-    static u32 getElementIdx(u32 x, u32 y) {
-        u32 elem = 0;
-        elem |= ((x >> 0u) & 0x1u) << 0u;
-        elem |= ((x >> 1u) & 0x1u) << 1u;
-        elem |= ((y >> 0u) & 0x1u) << 2u;
-        elem |= ((x >> 2u) & 0x1u) << 3u;
-        elem |= ((y >> 1u) & 0x1u) << 4u;
-        elem |= ((y >> 2u) & 0x1u) << 5u;
-
-        return elem;
-    }
-
-    static u32 getPipeIdx(u32 x, u32 y, bool is_neo) {
-        u32 pipe = 0;
-
-        if (!is_neo) {
-            pipe |= (((x >> 3u) ^ (y >> 3u) ^ (x >> 4u)) & 0x1u) << 0u;
-            pipe |= (((x >> 4u) ^ (y >> 4u)) & 0x1u) << 1u;
-            pipe |= (((x >> 5u) ^ (y >> 5u)) & 0x1u) << 2u;
-        } else {
-            pipe |= (((x >> 3u) ^ (y >> 3u) ^ (x >> 4u)) & 0x1u) << 0u;
-            pipe |= (((x >> 4u) ^ (y >> 4u)) & 0x1u) << 1u;
-            pipe |= (((x >> 5u) ^ (y >> 5u)) & 0x1u) << 2u;
-            pipe |= (((x >> 6u) ^ (y >> 5u)) & 0x1u) << 3u;
-        }
-
-        return pipe;
-    }
-
-    static u32 getBankIdx(u32 x, u32 y, u32 bank_width, u32 bank_height, u32 num_banks,
-                          u32 num_pipes) {
-        const u32 x_shift_offset = std::bit_width(bank_width * num_pipes) - 1;
-        const u32 y_shift_offset = std::bit_width(bank_height) - 1;
-        const u32 xs = x >> x_shift_offset;
-        const u32 ys = y >> y_shift_offset;
-        u32 bank = 0;
-        switch (num_banks) {
-        case 8:
-            bank |= (((xs >> 3u) ^ (ys >> 5u)) & 0x1u) << 0u;
-            bank |= (((xs >> 4u) ^ (ys >> 4u) ^ (ys >> 5u)) & 0x1u) << 1u;
-            bank |= (((xs >> 5u) ^ (ys >> 3u)) & 0x1u) << 2u;
-            break;
-        case 16:
-            bank |= (((xs >> 3u) ^ (ys >> 6u)) & 0x1u) << 0u;
-            bank |= (((xs >> 4u) ^ (ys >> 5u) ^ (ys >> 6u)) & 0x1u) << 1u;
-            bank |= (((xs >> 5u) ^ (ys >> 4u)) & 0x1u) << 2u;
-            bank |= (((xs >> 6u) ^ (ys >> 3u)) & 0x1u) << 3u;
-            break;
-        default:;
-        }
-
-        return bank;
-    }
-
-    u64 getTiledOffs(u32 x, u32 y, bool is_neo) const {
-        u64 element_index = getElementIdx(x, y);
-
-        u32 xh = x;
-        u32 yh = y;
-        u64 pipe = getPipeIdx(xh, yh, is_neo);
-        u64 bank = getBankIdx(xh, yh, 1, m_bank_height, m_num_banks, m_num_pipes);
-        u32 tile_bytes = (8 * 8 * 32 + 7) / 8;
-        u64 element_offset = (element_index * 32);
-        u64 tile_split_slice = 0;
-
-        if (tile_bytes > 512) {
-            tile_split_slice = element_offset / (static_cast<u64>(512) * 8);
-            element_offset %= (static_cast<u64>(512) * 8);
-            tile_bytes = 512;
-        }
-
-        u64 macro_tile_bytes =
-            (128 / 8) * (m_macro_tile_height / 8) * tile_bytes / (m_num_pipes * m_num_banks);
-        u64 macro_tiles_per_row = m_padded_width / 128;
-        u64 macro_tile_row_index = y / m_macro_tile_height;
-        u64 macro_tile_column_index = x / 128;
-        u64 macro_tile_index =
-            (macro_tile_row_index * macro_tiles_per_row) + macro_tile_column_index;
-        u64 macro_tile_offset = macro_tile_index * macro_tile_bytes;
-        u64 macro_tiles_per_slice = macro_tiles_per_row * (m_padded_height / m_macro_tile_height);
-        u64 slice_bytes = macro_tiles_per_slice * macro_tile_bytes;
-        u64 slice_offset = tile_split_slice * slice_bytes;
-        u64 tile_row_index = (y / 8) % m_bank_height;
-        u64 tile_index = tile_row_index;
-        u64 tile_offset = tile_index * tile_bytes;
-
-        u64 tile_split_slice_rotation = ((m_num_banks / 2) + 1) * tile_split_slice;
-        bank ^= tile_split_slice_rotation;
-        bank &= (m_num_banks - 1);
-
-        u64 total_offset = (slice_offset + macro_tile_offset + tile_offset) * 8 + element_offset;
-        u64 bit_offset = total_offset & 0x7u;
-        total_offset /= 8;
-
-        u64 pipe_interleave_offset = total_offset & 0xffu;
-        u64 offset = total_offset >> 8u;
-        u64 byte_offset = pipe_interleave_offset | (pipe << (8u)) | (bank << (8u + m_pipe_bits)) |
-                          (offset << (8u + m_pipe_bits + m_bank_bits));
-
-        return ((byte_offset << 3u) | bit_offset) / 8;
-    }
-};
-
-void ConvertTileToLinear(u8* dst, const u8* src, u32 width, u32 height, bool is_neo) {
-    TileManager32 t;
-    t.Init(width, height, is_neo);
-
-    for (u32 y = 0; y < height; y++) {
-        u32 x = 0;
-        u64 linear_offset = y * width * 4;
-
-        for (; x + 1 < width; x += 2) {
-            auto tiled_offset = t.getTiledOffs(x, y, is_neo);
-
-            std::memcpy(dst + linear_offset, src + tiled_offset, sizeof(u64));
-            linear_offset += 8;
-        }
-        if (x < width) {
-            auto tiled_offset = t.getTiledOffs(x, y, is_neo);
-            std::memcpy(dst + linear_offset, src + tiled_offset, sizeof(u32));
-        }
-    }
-}
-
-vk::Format DemoteImageFormatForDetiling(vk::Format format) {
+static vk::Format DemoteImageFormatForDetiling(vk::Format format) {
     switch (format) {
     case vk::Format::eR8Uint:
     case vk::Format::eR8Unorm:
@@ -253,12 +103,9 @@ const DetilerContext* TileManager::GetDetiler(const Image& image) const {
     case AmdGpu::TilingMode::Texture_Volume:
         switch (format) {
         case vk::Format::eR32Uint:
-            if (image.info.size.width % 8 == 0 && image.info.size.height % 8 == 0) {
-                return &detilers[DetilerType::Macro32x1];
-            } else {
-                // Need to test non-tile aligned images
-                return nullptr;
-            }
+            return &detilers[DetilerType::Macro32x1];
+        case vk::Format::eR32G32Uint:
+            return &detilers[DetilerType::Macro32x2];
         default:
             return nullptr;
         }
@@ -272,16 +119,16 @@ struct DetilerParams {
     u32 num_levels;
     u32 pitch0;
     u32 height;
-    u32 depth;
     u32 sizes[14];
 };
 
 TileManager::TileManager(const Vulkan::Instance& instance, Vulkan::Scheduler& scheduler)
     : instance{instance}, scheduler{scheduler} {
     static const std::array detiler_shaders{
-        HostShaders::DETILE_M8X1_COMP,  HostShaders::DETILE_M8X2_COMP,
-        HostShaders::DETILE_M32X1_COMP, HostShaders::DETILE_M32X2_COMP,
-        HostShaders::DETILE_M32X4_COMP, HostShaders::DETILE_MACRO32X1_COMP,
+        HostShaders::DETILE_M8X1_COMP,      HostShaders::DETILE_M8X2_COMP,
+        HostShaders::DETILE_M32X1_COMP,     HostShaders::DETILE_M32X2_COMP,
+        HostShaders::DETILE_M32X4_COMP,     HostShaders::DETILE_MACRO32X1_COMP,
+        HostShaders::DETILE_MACRO32X2_COMP,
     };
 
     boost::container::static_vector<vk::DescriptorSetLayoutBinding, 2> bindings{
@@ -465,16 +312,24 @@ std::pair<vk::Buffer, u32> TileManager::TryDetile(vk::Buffer in_buffer, u32 in_o
                                 set_writes);
 
     DetilerParams params;
+    params.num_levels = image.info.resources.levels;
     params.pitch0 = image.info.pitch >> (image.info.props.is_block ? 2u : 0u);
     params.height = image.info.size.height;
-    params.depth = image.info.size.depth;
-    params.num_levels = image.info.resources.levels;
+    if (image.info.tiling_mode == AmdGpu::TilingMode::Texture_Volume) {
+        ASSERT(image.info.resources.levels == 1);
+        ASSERT(image.info.num_bits >= 32);
+        const auto tiles_per_row = image.info.pitch / 8u;
+        const auto tiles_per_slice = tiles_per_row * ((image.info.size.height + 7u) / 8u);
+        params.sizes[0] = tiles_per_row;
+        params.sizes[1] = tiles_per_slice;
+    } else {
 
-    ASSERT(image.info.resources.levels <= 14);
-    std::memset(&params.sizes, 0, sizeof(params.sizes));
-    for (int m = 0; m < image.info.resources.levels; ++m) {
-        params.sizes[m] = image.info.mips_layout[m].size * image.info.resources.layers +
-                          (m > 0 ? params.sizes[m - 1] : 0);
+        ASSERT(image.info.resources.levels <= 14);
+        std::memset(&params.sizes, 0, sizeof(params.sizes));
+        for (int m = 0; m < image.info.resources.levels; ++m) {
+            params.sizes[m] = image.info.mips_layout[m].size * image.info.resources.layers +
+                              (m > 0 ? params.sizes[m - 1] : 0);
+        }
     }
 
     cmdbuf.pushConstants(*detiler->pl_layout, vk::ShaderStageFlagBits::eCompute, 0u, sizeof(params),

src/video_core/texture_cache/tile_manager.h

@@ -25,6 +25,7 @@ enum DetilerType : u32 {
     Micro32x4,
 
     Macro32x1,
+    Macro32x2,
 
     Max
 };