Merge pull request #353 from Subv/compressed_formats
GPU: Add support for the DXT23 and DXT45 compressed texture formats.
This commit is contained in:
commit
d019bb16f6
|
@ -49,9 +49,11 @@ struct FormatTuple {
|
|||
};
|
||||
|
||||
static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_format_tuples = {{
|
||||
{GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, false, 1}, // ABGR8
|
||||
{GL_RGB, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV, false, 1}, // B5G6R5
|
||||
{GL_COMPRESSED_RGB_S3TC_DXT1_EXT, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, true, 16}, // DXT1
|
||||
{GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, false, 1}, // ABGR8
|
||||
{GL_RGB, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV, false, 1}, // B5G6R5
|
||||
{GL_COMPRESSED_RGB_S3TC_DXT1_EXT, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, true, 16}, // DXT1
|
||||
{GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true, 16}, // DXT23
|
||||
{GL_COMPRESSED_RGBA_S3TC_DXT5_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true, 16}, // DXT45
|
||||
}};
|
||||
|
||||
static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType component_type) {
|
||||
|
@ -81,23 +83,6 @@ static u16 GetResolutionScaleFactor() {
|
|||
: Settings::values.resolution_factor);
|
||||
}
|
||||
|
||||
template <bool morton_to_gl, PixelFormat format>
|
||||
static void MortonCopyTile(u32 stride, u8* tile_buffer, u8* gl_buffer) {
|
||||
constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / 8;
|
||||
constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format);
|
||||
for (u32 y = 0; y < 8; ++y) {
|
||||
for (u32 x = 0; x < 8; ++x) {
|
||||
u8* tile_ptr = tile_buffer + VideoCore::MortonInterleave(x, y) * bytes_per_pixel;
|
||||
u8* gl_ptr = gl_buffer + ((7 - y) * stride + x) * gl_bytes_per_pixel;
|
||||
if (morton_to_gl) {
|
||||
std::memcpy(gl_ptr, tile_ptr, bytes_per_pixel);
|
||||
} else {
|
||||
std::memcpy(tile_ptr, gl_ptr, bytes_per_pixel);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
template <bool morton_to_gl, PixelFormat format>
|
||||
void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, VAddr base, VAddr start,
|
||||
VAddr end) {
|
||||
|
@ -121,9 +106,9 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, VAddr b
|
|||
static constexpr std::array<void (*)(u32, u32, u32, u8*, VAddr, VAddr, VAddr),
|
||||
SurfaceParams::MaxPixelFormat>
|
||||
morton_to_gl_fns = {
|
||||
MortonCopy<true, PixelFormat::ABGR8>,
|
||||
MortonCopy<true, PixelFormat::B5G6R5>,
|
||||
MortonCopy<true, PixelFormat::DXT1>,
|
||||
MortonCopy<true, PixelFormat::ABGR8>, MortonCopy<true, PixelFormat::B5G6R5>,
|
||||
MortonCopy<true, PixelFormat::DXT1>, MortonCopy<true, PixelFormat::DXT23>,
|
||||
MortonCopy<true, PixelFormat::DXT45>,
|
||||
};
|
||||
|
||||
static constexpr std::array<void (*)(u32, u32, u32, u8*, VAddr, VAddr, VAddr),
|
||||
|
@ -131,7 +116,9 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, VAddr, VAddr, VAddr),
|
|||
gl_to_morton_fns = {
|
||||
MortonCopy<false, PixelFormat::ABGR8>,
|
||||
MortonCopy<false, PixelFormat::B5G6R5>,
|
||||
// TODO(Subv): Swizzling the DXT1 format is not yet supported
|
||||
// TODO(Subv): Swizzling the DXT1/DXT23/DXT45 formats is not yet supported
|
||||
nullptr,
|
||||
nullptr,
|
||||
nullptr,
|
||||
};
|
||||
|
||||
|
|
|
@ -55,6 +55,8 @@ struct SurfaceParams {
|
|||
ABGR8 = 0,
|
||||
B5G6R5 = 1,
|
||||
DXT1 = 2,
|
||||
DXT23 = 3,
|
||||
DXT45 = 4,
|
||||
|
||||
Max,
|
||||
Invalid = 255,
|
||||
|
@ -84,9 +86,11 @@ struct SurfaceParams {
|
|||
return 0;
|
||||
|
||||
constexpr std::array<unsigned int, MaxPixelFormat> bpp_table = {
|
||||
32, // ABGR8
|
||||
16, // B5G6R5
|
||||
64, // DXT1
|
||||
32, // ABGR8
|
||||
16, // B5G6R5
|
||||
64, // DXT1
|
||||
128, // DXT23
|
||||
128, // DXT45
|
||||
};
|
||||
|
||||
ASSERT(static_cast<size_t>(format) < bpp_table.size());
|
||||
|
@ -125,6 +129,10 @@ struct SurfaceParams {
|
|||
return PixelFormat::B5G6R5;
|
||||
case Tegra::Texture::TextureFormat::DXT1:
|
||||
return PixelFormat::DXT1;
|
||||
case Tegra::Texture::TextureFormat::DXT23:
|
||||
return PixelFormat::DXT23;
|
||||
case Tegra::Texture::TextureFormat::DXT45:
|
||||
return PixelFormat::DXT45;
|
||||
default:
|
||||
NGLOG_CRITICAL(HW_GPU, "Unimplemented format={}", static_cast<u32>(format));
|
||||
UNREACHABLE();
|
||||
|
@ -140,6 +148,10 @@ struct SurfaceParams {
|
|||
return Tegra::Texture::TextureFormat::B5G6R5;
|
||||
case PixelFormat::DXT1:
|
||||
return Tegra::Texture::TextureFormat::DXT1;
|
||||
case PixelFormat::DXT23:
|
||||
return Tegra::Texture::TextureFormat::DXT23;
|
||||
case PixelFormat::DXT45:
|
||||
return Tegra::Texture::TextureFormat::DXT45;
|
||||
default:
|
||||
UNREACHABLE();
|
||||
}
|
||||
|
|
|
@ -48,6 +48,10 @@ u32 BytesPerPixel(TextureFormat format) {
|
|||
case TextureFormat::DXT1:
|
||||
// In this case a 'pixel' actually refers to a 4x4 tile.
|
||||
return 8;
|
||||
case TextureFormat::DXT23:
|
||||
case TextureFormat::DXT45:
|
||||
// In this case a 'pixel' actually refers to a 4x4 tile.
|
||||
return 16;
|
||||
case TextureFormat::A8R8G8B8:
|
||||
return 4;
|
||||
case TextureFormat::B5G6R5:
|
||||
|
@ -67,7 +71,9 @@ std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width,
|
|||
|
||||
switch (format) {
|
||||
case TextureFormat::DXT1:
|
||||
// In the DXT1 format, each 4x4 tile is swizzled instead of just individual pixel values.
|
||||
case TextureFormat::DXT23:
|
||||
case TextureFormat::DXT45:
|
||||
// In the DXT formats, each 4x4 tile is swizzled instead of just individual pixel values.
|
||||
CopySwizzledData(width / 4, height / 4, bytes_per_pixel, bytes_per_pixel, data,
|
||||
unswizzled_data.data(), true, block_height);
|
||||
break;
|
||||
|
@ -91,6 +97,8 @@ std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat
|
|||
// TODO(Subv): Implement.
|
||||
switch (format) {
|
||||
case TextureFormat::DXT1:
|
||||
case TextureFormat::DXT23:
|
||||
case TextureFormat::DXT45:
|
||||
case TextureFormat::A8R8G8B8:
|
||||
case TextureFormat::B5G6R5:
|
||||
// TODO(Subv): For the time being just forward the same data without any decoding.
|
||||
|
|
Reference in New Issue