GPU/OpenGL: Only use one upload path and clamp to max SSBO size

This commit is contained in:
Connor McLaughlin 2021-03-18 12:44:45 +10:00
parent 703715724e
commit 391114ccae
2 changed files with 30 additions and 22 deletions

View File

@ -319,22 +319,29 @@ void GPU_HW_OpenGL::SetCapabilities(HostDisplay* host_display)
#ifdef __APPLE__
// Partial texture buffer uploads appear to be broken in macOS's OpenGL driver.
m_supports_texture_buffer = false;
m_use_texture_buffer_for_vram_writes = false;
#else
m_supports_texture_buffer = (GLAD_GL_VERSION_3_1 || GLAD_GL_ES_VERSION_3_2);
m_use_texture_buffer_for_vram_writes = (GLAD_GL_VERSION_3_1 || GLAD_GL_ES_VERSION_3_2);
#endif
if (m_supports_texture_buffer)
m_texture_stream_buffer_size = VRAM_UPDATE_TEXTURE_BUFFER_SIZE;
if (m_use_texture_buffer_for_vram_writes)
{
glGetIntegerv(GL_MAX_TEXTURE_BUFFER_SIZE, reinterpret_cast<GLint*>(&m_max_texture_buffer_size));
Log_InfoPrintf("Max texel buffer size: %u", m_max_texture_buffer_size);
if (m_max_texture_buffer_size < VRAM_WIDTH * VRAM_HEIGHT)
GLint max_texel_buffer_size;
glGetIntegerv(GL_MAX_TEXTURE_BUFFER_SIZE, reinterpret_cast<GLint*>(&max_texel_buffer_size));
Log_InfoPrintf("Max texel buffer size: %u", max_texel_buffer_size);
if (max_texel_buffer_size < VRAM_WIDTH * VRAM_HEIGHT)
{
Log_WarningPrintf("Maximum texture buffer size is less than VRAM size, not using texel buffers.");
m_supports_texture_buffer = false;
m_use_texture_buffer_for_vram_writes = false;
}
else
{
m_texture_stream_buffer_size =
std::min<u32>(VRAM_UPDATE_TEXTURE_BUFFER_SIZE, static_cast<u32>(max_texel_buffer_size) * sizeof(u16));
}
}
if (!m_supports_texture_buffer || m_max_texture_buffer_size < VRAM_WIDTH * VRAM_HEIGHT)
if (!m_use_texture_buffer_for_vram_writes)
{
// Try SSBOs.
GLint max_fragment_storage_blocks = 0;
@ -352,11 +359,13 @@ void GPU_HW_OpenGL::SetCapabilities(HostDisplay* host_display)
if (m_use_ssbo_for_vram_writes)
{
Log_InfoPrintf("Using shader storage buffers for VRAM writes.");
m_texture_stream_buffer_size =
static_cast<u32>(std::min<u64>(VRAM_UPDATE_TEXTURE_BUFFER_SIZE, static_cast<u64>(max_ssbo_size)));
}
else
{
Log_WarningPrintf(
"Texture buffers are not supported, VRAM writes will be slower and multisampling will be unavailable.");
Log_WarningPrintf("Texture buffers and SSBOs are not supported, VRAM writes will be slower and multisampling "
"will be unavailable.");
m_max_multisamples = 1;
m_supports_per_sample_shading = false;
}
@ -475,15 +484,14 @@ bool GPU_HW_OpenGL::CreateUniformBuffer()
bool GPU_HW_OpenGL::CreateTextureBuffer()
{
// We use the pixel unpack buffer here because we share it with CPU-decoded VRAM writes.
const GLenum target =
(m_use_ssbo_for_vram_writes ? GL_SHADER_STORAGE_BUFFER :
(m_supports_texture_buffer ? GL_TEXTURE_BUFFER : GL_PIXEL_UNPACK_BUFFER));
m_texture_stream_buffer = GL::StreamBuffer::Create(target, VRAM_UPDATE_TEXTURE_BUFFER_SIZE);
(m_use_texture_buffer_for_vram_writes ? GL_TEXTURE_BUFFER : GL_PIXEL_UNPACK_BUFFER));
m_texture_stream_buffer = GL::StreamBuffer::Create(target, m_texture_stream_buffer_size);
if (!m_texture_stream_buffer)
return false;
if (m_max_texture_buffer_size > 0)
if (m_use_texture_buffer_for_vram_writes)
{
glGenTextures(1, &m_texture_buffer_r16ui_texture);
glBindTexture(GL_TEXTURE_BUFFER, m_texture_buffer_r16ui_texture);
@ -673,7 +681,7 @@ bool GPU_HW_OpenGL::CompilePrograms()
m_vram_update_depth_program = std::move(*prog);
UPDATE_PROGRESS();
if (m_supports_texture_buffer || m_use_ssbo_for_vram_writes)
if (m_use_texture_buffer_for_vram_writes || m_use_ssbo_for_vram_writes)
{
prog = shader_cache.GetProgram(shadergen.GenerateScreenQuadVertexShader(), {},
shadergen.GenerateVRAMWriteFragmentShader(m_use_ssbo_for_vram_writes),
@ -1060,7 +1068,7 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
}
const u32 num_pixels = width * height;
if (num_pixels < m_max_texture_buffer_size || m_use_ssbo_for_vram_writes)
if (m_use_texture_buffer_for_vram_writes || m_use_ssbo_for_vram_writes)
{
const auto map_result = m_texture_stream_buffer->Map(sizeof(u16), num_pixels * sizeof(u16));
std::memcpy(map_result.pointer, data, num_pixels * sizeof(u16));
@ -1092,10 +1100,10 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
}
else
{
if ((x + width) > VRAM_WIDTH || (y + height) > VRAM_HEIGHT)
if ((x + width) > VRAM_WIDTH || (y + height) > VRAM_HEIGHT || check_mask)
{
// CPU round trip if oversized for now.
Log_WarningPrintf("Oversized VRAM update (%u-%u, %u-%u), CPU round trip", x, x + width, y, y + height);
Log_WarningPrintf("Oversized/masked VRAM update (%u-%u, %u-%u), CPU round trip", x, x + width, y, y + height);
ReadVRAM(0, 0, VRAM_WIDTH, VRAM_HEIGHT);
GPU::UpdateVRAM(x, y, width, height, data, set_mask, check_mask);
UpdateVRAM(0, 0, VRAM_WIDTH, VRAM_HEIGHT, m_vram_shadow.data(), false, false);
@ -1109,6 +1117,7 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
// reverse copy the rows so it matches opengl's lower-left origin
const u32 source_stride = width * sizeof(u16);
const u8* source_ptr = static_cast<const u8*>(data) + (source_stride * (height - 1));
const u16 mask_or = set_mask ? 0x8000 : 0x0000;
u32* dest_ptr = static_cast<u32*>(map_result.pointer);
for (u32 row = 0; row < height; row++)
{
@ -1119,8 +1128,7 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
u16 src_col;
std::memcpy(&src_col, source_row_ptr, sizeof(src_col));
source_row_ptr += sizeof(src_col);
*(dest_ptr++) = RGBA5551ToRGBA8888(src_col);
*(dest_ptr++) = RGBA5551ToRGBA8888(src_col | mask_or);
}
source_ptr -= source_stride;

View File

@ -104,9 +104,9 @@ private:
GL::Program m_vram_update_depth_program;
u32 m_uniform_buffer_alignment = 1;
u32 m_max_texture_buffer_size = 0;
u32 m_texture_stream_buffer_size = 0;
bool m_supports_texture_buffer = false;
bool m_use_texture_buffer_for_vram_writes = false;
bool m_use_ssbo_for_vram_writes = false;
GLenum m_current_depth_test = 0;