OpenGLTexture: Fix format table, MSAA creation and upload sizing

This commit is contained in:
Stenzek committed 2026-10-13 20:48:44 +10:00
1 parent ce9ee819ed
commit 0aa7687e7b
2 files changed
+85 -31

No files matched your search

+13
View File
@@ -377,6 +377,19 @@ bool OpenGLDevice::CheckFeatures(CreateFlags create_flags)
DEV_LOG("GL_MAX_TEXTURE_SIZE: {}", max_texture_size);
glGetIntegerv(GL_MAX_SAMPLES, &max_samples);
DEV_LOG("GL_MAX_SAMPLES: {}", max_samples);
// Multisampled textures have their own limits, and are not available at all in GLES 3.0.
if (GLAD_GL_VERSION_3_2 || GLAD_GL_ARB_texture_multisample || GLAD_GL_ES_VERSION_3_1)
{
GLint max_color_samples = 1, max_depth_samples = 1;
glGetIntegerv(GL_MAX_COLOR_TEXTURE_SAMPLES, &max_color_samples);
glGetIntegerv(GL_MAX_DEPTH_TEXTURE_SAMPLES, &max_depth_samples);
max_samples = std::min(max_samples, std::min(max_color_samples, max_depth_samples));
}
else
{
max_samples = 1;
}
m_max_texture_size = std::max(1024u, static_cast<u32>(max_texture_size));
m_max_multisamples = static_cast<u16>(std::max(1u, static_cast<u32>(max_samples)));
+72 -31
View File
@@ -40,10 +40,10 @@ const std::tuple<GLenum, GLenum, GLenum>& OpenGLTexture::GetPixelFormatMapping(G
{}, // RGB5A1
{GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_5_5_5_1}, // A1BGR5
{GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // R8
{GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_SHORT}, // D16
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT}, // D24S8
{GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24S8
{GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32F
{GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, GL_FLOAT}, // D32FS8
{GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, GL_FLOAT_32_UNSIGNED_INT_24_8_REV}, // D32FS8
{GL_R16, GL_RED, GL_UNSIGNED_SHORT}, // R16
{GL_R16I, GL_RED_INTEGER, GL_SHORT}, // R16I
{GL_R16UI, GL_RED_INTEGER, GL_UNSIGNED_SHORT}, // R16U
@@ -58,7 +58,7 @@ const std::tuple<GLenum, GLenum, GLenum>& OpenGLTexture::GetPixelFormatMapping(G
{GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT}, // RGBA16
{GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT}, // RGBA16F
{GL_RGBA32F, GL_RGBA, GL_FLOAT}, // RGBA32F
{GL_RGB10_A2, GL_BGRA, GL_UNSIGNED_INT_2_10_10_10_REV}, // RGB10A2
{GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV}, // RGB10A2
{GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE}, // SRGBA8
{GL_COMPRESSED_RGBA_S3TC_DXT1_EXT, GL_COMPRESSED_RGBA_S3TC_DXT1_EXT, GL_UNSIGNED_BYTE}, // BC1
{GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_UNSIGNED_BYTE}, // BC2
@@ -71,15 +71,15 @@ const std::tuple<GLenum, GLenum, GLenum>& OpenGLTexture::GetPixelFormatMapping(G
mapping_gles = {{
{}, // Unknown
{GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE}, // RGBA8
{GL_RGBA8, GL_BGRA, GL_UNSIGNED_BYTE}, // BGRA8
{}, // BGRA8
{GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5}, // RGB565
{}, // RGB5A1
{GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_5_5_5_1}, // A1BGR5
{GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // R8
{GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_SHORT}, // D16
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT}, // D24S8
{GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24S8
{GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32F
{GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, GL_FLOAT}, // D32FS8
{GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, GL_FLOAT_32_UNSIGNED_INT_24_8_REV}, // D32FS8
{GL_R16F, GL_RED, GL_HALF_FLOAT}, // R16
{GL_R16I, GL_RED_INTEGER, GL_SHORT}, // R16I
{GL_R16UI, GL_RED_INTEGER, GL_UNSIGNED_SHORT}, // R16U
@@ -88,13 +88,13 @@ const std::tuple<GLenum, GLenum, GLenum>& OpenGLTexture::GetPixelFormatMapping(G
{GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT}, // R32U
{GL_R32F, GL_RED, GL_FLOAT}, // R32F
{GL_RG8, GL_RG, GL_UNSIGNED_BYTE}, // RG8
{GL_RG16, GL_RG, GL_HALF_FLOAT}, // RG16
{GL_RG16F, GL_RG, GL_HALF_FLOAT}, // RG16
{GL_RG16F, GL_RG, GL_HALF_FLOAT}, // RG16F
{GL_RG32F, GL_RG, GL_FLOAT}, // RG32F
{GL_RGBA16, GL_RGBA, GL_HALF_FLOAT}, // RGBA16
{GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT}, // RGBA16
{GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT}, // RGBA16F
{GL_RGBA32F, GL_RGBA, GL_FLOAT}, // RGBA32F
{GL_RGB10_A2, GL_BGRA, GL_UNSIGNED_INT_2_10_10_10_REV}, // RGB10A2
{GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV}, // RGB10A2
{GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE}, // SRGBA8
{GL_COMPRESSED_RGBA_S3TC_DXT1_EXT, GL_COMPRESSED_RGBA_S3TC_DXT1_EXT, GL_UNSIGNED_BYTE}, // BC1
{GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_UNSIGNED_BYTE}, // BC2
@@ -110,6 +110,13 @@ ALWAYS_INLINE static u32 GetUploadAlignment(u32 pitch)
return ((pitch % 4) == 0) ? 4 : (((pitch % 2) == 0) ? 2 : 1);
}
ALWAYS_INLINE static u32 GetStreamUploadPitch(GPUTextureFormat format, u32 width)
{
// GL_UNPACK_ROW_LENGTH is ignored for compressed data, so it has to be tightly packed.
const u32 pitch = GPUTexture::CalcUploadPitch(format, width);
return GPUTexture::IsCompressedFormat(format) ? pitch : Common::AlignUpPow2(pitch, TEXTURE_UPLOAD_PITCH_ALIGNMENT);
}
OpenGLTexture::OpenGLTexture(u32 width, u32 height, u32 layers, u32 levels, u32 samples, Type type,
GPUTextureFormat format, Flags flags, GLuint id)
: GPUTexture(static_cast<u16>(width), static_cast<u16>(height), static_cast<u8>(layers), static_cast<u8>(levels),
@@ -130,8 +137,10 @@ OpenGLTexture::~OpenGLTexture()
bool OpenGLTexture::UseTextureStorage(bool multisampled)
{
return (GLAD_GL_VERSION_4_2 || GLAD_GL_ARB_texture_storage) ||
(multisampled ? GLAD_GL_ES_VERSION_3_1 : GLAD_GL_ES_VERSION_3_0);
if (multisampled)
return (GLAD_GL_VERSION_4_3 || GLAD_GL_ARB_texture_storage_multisample || GLAD_GL_ES_VERSION_3_1);
return (GLAD_GL_VERSION_4_2 || GLAD_GL_ARB_texture_storage || GLAD_GL_ES_VERSION_3_0);
}
bool OpenGLTexture::UseTextureStorage() const
@@ -177,7 +186,7 @@ std::unique_ptr<OpenGLTexture> OpenGLTexture::Create(u32 width, u32 height, u32
}
glTexParameteri(target, GL_TEXTURE_BASE_LEVEL, 0);
glTexParameteri(target, GL_TEXTURE_MAX_LEVEL, levels);
glTexParameteri(target, GL_TEXTURE_MAX_LEVEL, levels - 1);
}
else
{
@@ -193,7 +202,6 @@ std::unique_ptr<OpenGLTexture> OpenGLTexture::Create(u32 width, u32 height, u32
if (!use_texture_storage || data)
{
const u32 pixel_size = GetPixelSize(format);
const u32 alignment = GetUploadAlignment(data_pitch);
if (data)
{
@@ -205,24 +213,46 @@ std::unique_ptr<OpenGLTexture> OpenGLTexture::Create(u32 width, u32 height, u32
glPixelStorei(GL_UNPACK_ALIGNMENT, alignment);
}
// Compressed data can't use the row length, repack it if the caller's pitch has padding.
DynamicHeapArray<u8> packed_data;
if (data && is_compressed && data_pitch != CalcUploadPitch(format, width))
{
const u32 packed_pitch = CalcUploadPitch(format, width);
packed_data.resize(CalcUploadSize(format, height, packed_pitch));
CopyTextureDataForUpload(width, height, format, packed_data.data(), packed_pitch, data, data_pitch);
}
const u8* data_ptr = static_cast<const u8*>(data);
u32 current_width = width;
u32 current_height = height;
for (u32 i = 0; i < levels; i++)
{
const u8* const level_data_ptr = data_ptr;
const u32 compressed_size = CalcUploadSize(format, current_height, CalcUploadPitch(format, current_width));
if (i == 0 && !packed_data.empty())
{
data_ptr = packed_data.data();
}
else if (data && i > 0)
{
glPixelStorei(GL_UNPACK_ROW_LENGTH, CalcUploadRowLengthFromPitch(format, data_pitch));
glPixelStorei(GL_UNPACK_ALIGNMENT, GetUploadAlignment(data_pitch));
}
if (use_texture_storage)
{
if (is_compressed)
{
const u32 size = CalcUploadSize(format, current_height, data_pitch);
if (layers > 1)
{
glCompressedTexSubImage3D(target, i, 0, 0, 0, current_width, current_height, layers, gl_format, size,
glCompressedTexSubImage3D(target, i, 0, 0, 0, current_width, current_height, layers, gl_format,
compressed_size,
data_ptr);
}
else
{
glCompressedTexSubImage2D(target, i, 0, 0, current_width, current_height, gl_format, size, data_ptr);
glCompressedTexSubImage2D(target, i, 0, 0, current_width, current_height, gl_format, compressed_size,
data_ptr);
}
}
else
@@ -237,15 +267,16 @@ std::unique_ptr<OpenGLTexture> OpenGLTexture::Create(u32 width, u32 height, u32
{
if (is_compressed)
{
const u32 size = CalcUploadSize(format, current_height, data_pitch);
if (layers > 1)
{
glCompressedTexImage3D(target, i, gl_internal_format, current_width, current_height, layers, 0, size,
glCompressedTexImage3D(target, i, gl_internal_format, current_width, current_height, layers, 0,
compressed_size,
data_ptr);
}
else
{
glCompressedTexImage2D(target, i, gl_internal_format, current_width, current_height, 0, size, data_ptr);
glCompressedTexImage2D(target, i, gl_internal_format, current_width, current_height, 0, compressed_size,
data_ptr);
}
}
else
@@ -263,19 +294,19 @@ std::unique_ptr<OpenGLTexture> OpenGLTexture::Create(u32 width, u32 height, u32
}
}
if (data_ptr)
data_ptr += data_pitch * current_height;
if (level_data_ptr)
data_ptr = level_data_ptr + CalcUploadSize(format, current_height, data_pitch);
current_width = (current_width > 1) ? (current_width / 2u) : current_width;
current_height = (current_height > 1) ? (current_height / 2u) : current_height;
// TODO: Incorrect assumption.
data_pitch = pixel_size * current_width;
// Subsequent levels are assumed to be tightly packed.
data_pitch = CalcUploadPitch(format, current_width);
}
if (data)
{
if (alignment != DEFAULT_UPLOAD_ALIGNMENT)
if (alignment != DEFAULT_UPLOAD_ALIGNMENT || levels > 1)
glPixelStorei(GL_UNPACK_ALIGNMENT, DEFAULT_UPLOAD_ALIGNMENT);
glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
}
@@ -310,8 +341,8 @@ bool OpenGLTexture::Update(u32 x, u32 y, u32 width, u32 height, const void* data
// Worth using the PBO? Driver probably knows better...
const GLenum target = GetGLTarget();
const auto [gl_internal_format, gl_format, gl_type] = GetPixelFormatMapping(m_format, OpenGLDevice::IsGLES());
const u32 preferred_pitch = Common::AlignUpPow2(CalcUploadPitch(width), TEXTURE_UPLOAD_PITCH_ALIGNMENT);
const u32 map_size = CalcUploadSize(height, pitch);
const u32 preferred_pitch = GetStreamUploadPitch(m_format, width);
const u32 map_size = CalcUploadSize(height, preferred_pitch);
OpenGLStreamBuffer* sb = OpenGLDevice::GetTextureStreamBuffer();
CommitClear();
@@ -326,6 +357,16 @@ bool OpenGLTexture::Update(u32 x, u32 y, u32 width, u32 height, const void* data
{
GL_INS_FMT("Not using PBO for map size {}", map_size);
// Compressed data can't use the row length, repack it if the caller's pitch has padding.
DynamicHeapArray<u8> packed_data;
if (IsCompressedFormat() && pitch != preferred_pitch)
{
packed_data.resize(map_size);
CopyTextureDataForUpload(width, height, m_format, packed_data.data(), preferred_pitch, data, pitch);
data = packed_data.data();
pitch = preferred_pitch;
}
const u32 alignment = GetUploadAlignment(pitch);
if (alignment != DEFAULT_UPLOAD_ALIGNMENT)
glPixelStorei(GL_UNPACK_ALIGNMENT, alignment);
@@ -361,7 +402,7 @@ bool OpenGLTexture::Update(u32 x, u32 y, u32 width, u32 height, const void* data
glPixelStorei(GL_UNPACK_ROW_LENGTH, CalcUploadRowLengthFromPitch(preferred_pitch));
if (IsCompressedFormat())
{
const u32 size = CalcUploadSize(height, pitch);
const u32 size = map_size;
if (IsTextureArray())
{
glCompressedTexSubImage3D(target, level, x, y, layer, width, height, 1, gl_format, size,
@@ -402,7 +443,7 @@ bool OpenGLTexture::Map(void** map, u32* map_stride, u32 x, u32 y, u32 width, u3
if ((x + width) > GetMipWidth(level) || (y + height) > GetMipHeight(level) || layer >= m_layers || level >= m_levels)
return false;
const u32 pitch = Common::AlignUpPow2(CalcUploadPitch(width), TEXTURE_UPLOAD_PITCH_ALIGNMENT);
const u32 pitch = GetStreamUploadPitch(m_format, width);
const u32 upload_size = CalcUploadSize(height, pitch);
OpenGLStreamBuffer* sb = OpenGLDevice::GetTextureStreamBuffer();
if (!sb || upload_size > sb->GetSize())
@@ -426,7 +467,7 @@ void OpenGLTexture::Unmap()
{
CommitClear();
const u32 pitch = Common::AlignUpPow2(CalcUploadPitch(m_map_width), TEXTURE_UPLOAD_PITCH_ALIGNMENT);
const u32 pitch = GetStreamUploadPitch(m_format, m_map_width);
const u32 upload_size = CalcUploadSize(m_map_height, pitch);
GPUDevice::GetStatistics().buffer_streamed += upload_size;