GPU/OpenGL: Switch back to framebuffer blits over shader copy

Ends up being slower on Mali.
This commit is contained in:
Connor McLaughlin 2021-04-09 18:06:42 +10:00
parent 934b6e8f45
commit 327e8f3f59
2 changed files with 46 additions and 96 deletions

View File

@ -369,7 +369,10 @@ void GPU_HW_OpenGL::SetCapabilities(HostDisplay* host_display)
}
else
{
Log_WarningPrintf("Texture buffers and SSBOs are not supported, VRAM writes will be slower.");
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;
}
}
@ -725,21 +728,6 @@ bool GPU_HW_OpenGL::CompilePrograms()
m_downsample_program = std::move(*prog);
}
prog = shader_cache.GetProgram(shadergen.GenerateUVQuadVertexShader(), {}, shadergen.GenerateSampleFragmentShader(),
[this, use_binding_layout](GL::Program& prog) {
if (!IsGLES() && !use_binding_layout)
prog.BindFragData(0, "o_col0");
});
if (!prog)
return false;
if (!use_binding_layout)
{
prog->Bind();
prog->Uniform1i("samp0", 0);
}
m_blit_program = std::move(*prog);
UPDATE_PROGRESS();
#undef UPDATE_PROGRESS
@ -1155,20 +1143,21 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
m_texture_stream_buffer->Unmap(num_pixels * sizeof(u32));
m_texture_stream_buffer->Bind();
// have to write to the 1x texture first
if (m_resolution_scale > 1)
m_vram_encoding_texture.Bind();
else
m_vram_texture.Bind();
// lower-left origin flip happens here
const u32 flipped_y = VRAM_HEIGHT - y - height;
// have to write to the 1x texture first
GL::Texture& dst_texture =
(m_resolution_scale > 1 || m_multisamples > 1) ? m_vram_encoding_texture : m_vram_texture;
// update texture data
dst_texture.Bind();
glTexSubImage2D(dst_texture.GetGLTarget(), 0, x, flipped_y, width, height, GL_RGBA, GL_UNSIGNED_BYTE,
glTexSubImage2D(m_vram_texture.GetGLTarget(), 0, x, flipped_y, width, height, GL_RGBA, GL_UNSIGNED_BYTE,
reinterpret_cast<void*>(static_cast<uintptr_t>(map_result.buffer_offset)));
m_texture_stream_buffer->Unbind();
if (&dst_texture != &m_vram_texture)
if (m_resolution_scale > 1)
{
// scale to internal resolution
const u32 scaled_width = width * m_resolution_scale;
@ -1176,12 +1165,12 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
const u32 scaled_x = x * m_resolution_scale;
const u32 scaled_y = y * m_resolution_scale;
const u32 scaled_flipped_y = m_vram_texture.GetHeight() - scaled_y - scaled_height;
BlitTextureToFramebuffer(dst_texture, x, flipped_y, width, height, scaled_x, scaled_flipped_y, scaled_width,
scaled_height);
glDisable(GL_SCISSOR_TEST);
m_vram_encoding_texture.BindFramebuffer(GL_READ_FRAMEBUFFER);
glBlitFramebuffer(x, flipped_y, x + width, flipped_y + height, scaled_x, scaled_flipped_y,
scaled_x + scaled_width, scaled_flipped_y + scaled_height, GL_COLOR_BUFFER_BIT, GL_NEAREST);
glEnable(GL_SCISSOR_TEST);
}
RestoreGraphicsAPIState();
}
}
@ -1253,76 +1242,19 @@ void GPU_HW_OpenGL::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 wid
}
else
{
// glBlitFramebufer with same source/destination should be legal, but on Mali (at least Bifrost) it breaks.
// So, blit from the shadow texture, like in the other renderers.
if (src_dirty)
UpdateVRAMReadTexture();
CopyTexture(m_vram_texture, m_vram_fbo_id, m_vram_read_texture, m_vram_read_texture.GetGLFramebufferID(), src_x,
src_y, dst_x, dst_y, width, height);
}
IncludeVRAMDirtyRectangle(dst_bounds);
}
void GPU_HW_OpenGL::CopyTexture(GL::Texture& dest, GLuint dest_fbo, GL::Texture& src, GLuint src_fbo, u32 src_x,
u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height)
{
if (src.IsMultisampled())
{
// The MSAA case still needs framebuffer blits.
glBindFramebuffer(GL_DRAW_FRAMEBUFFER, dest_fbo);
glBindFramebuffer(GL_READ_FRAMEBUFFER, src_fbo);
glDisable(GL_SCISSOR_TEST);
m_vram_read_texture.BindFramebuffer(GL_READ_FRAMEBUFFER);
glBlitFramebuffer(src_x, src_y, src_x + width, src_y + height, dst_x, dst_y, dst_x + width, dst_y + height,
GL_COLOR_BUFFER_BIT, GL_NEAREST);
glEnable(GL_SCISSOR_TEST);
RestoreGraphicsAPIState();
return;
}
if (GLAD_GL_VERSION_4_3)
{
glCopyImageSubData(src.GetGLId(), src.GetGLTarget(), 0, src_x, src_y, 0, dest.GetGLId(), src.GetGLTarget(), 0,
dst_x, dst_y, 0, width, height, 1);
}
else if (GLAD_GL_EXT_copy_image)
{
glCopyImageSubDataEXT(src.GetGLId(), src.GetGLTarget(), 0, src_x, src_y, 0, dest.GetGLId(), src.GetGLTarget(), 0,
dst_x, dst_y, 0, width, height, 1);
}
else if (GLAD_GL_OES_copy_image)
{
glCopyImageSubDataOES(src.GetGLId(), src.GetGLTarget(), 0, src_x, src_y, 0, dest.GetGLId(), src.GetGLTarget(), 0,
dst_x, dst_y, 0, width, height, 1);
}
else
{
glBindFramebuffer(GL_DRAW_FRAMEBUFFER, dest_fbo);
BlitTextureToFramebuffer(src, src_x, src_y, width, height, dst_x, dst_y, width, height);
RestoreGraphicsAPIState();
}
}
void GPU_HW_OpenGL::BlitTextureToFramebuffer(GL::Texture& src, u32 src_x, u32 src_y, u32 src_width, u32 src_height,
u32 dst_x, u32 dst_y, u32 dst_width, u32 dst_height)
{
// But a copy shader is probably better on mobile drivers.
const float uniforms[4] = {
static_cast<float>(src_x) / static_cast<float>(src.GetWidth()),
static_cast<float>(src_y) / static_cast<float>(src.GetHeight()),
static_cast<float>(src_x + src_width) / static_cast<float>(src.GetWidth()),
static_cast<float>(src_y + src_height) / static_cast<float>(src.GetHeight()),
};
UploadUniformBuffer(uniforms, sizeof(uniforms));
glDisable(GL_SCISSOR_TEST);
glDisable(GL_BLEND);
SetDepthFunc(GL_ALWAYS);
glViewport(dst_x, dst_y, dst_width, dst_height);
src.Bind();
m_blit_program.Bind();
glBindVertexArray(m_attributeless_vao_id);
glDrawArrays(GL_TRIANGLES, 0, 3);
IncludeVRAMDirtyRectangle(dst_bounds);
}
void GPU_HW_OpenGL::UpdateVRAMReadTexture()
@ -1332,9 +1264,32 @@ void GPU_HW_OpenGL::UpdateVRAMReadTexture()
const u32 height = scaled_rect.GetHeight();
const u32 x = scaled_rect.left;
const u32 y = m_vram_texture.GetHeight() - scaled_rect.top - height;
const bool multisampled = m_vram_texture.IsMultisampled();
CopyTexture(m_vram_read_texture, m_vram_read_texture.GetGLFramebufferID(), m_vram_texture, m_vram_fbo_id, x, y, x, y,
width, height);
if (!multisampled && GLAD_GL_VERSION_4_3)
{
glCopyImageSubData(m_vram_texture.GetGLId(), m_vram_texture.GetGLTarget(), 0, x, y, 0,
m_vram_read_texture.GetGLId(), m_vram_texture.GetGLTarget(), 0, x, y, 0, width, height, 1);
}
else if (!multisampled && GLAD_GL_EXT_copy_image)
{
glCopyImageSubDataEXT(m_vram_texture.GetGLId(), m_vram_texture.GetGLTarget(), 0, x, y, 0,
m_vram_read_texture.GetGLId(), m_vram_texture.GetGLTarget(), 0, x, y, 0, width, height, 1);
}
else if (!multisampled && GLAD_GL_OES_copy_image)
{
glCopyImageSubDataOES(m_vram_texture.GetGLId(), m_vram_texture.GetGLTarget(), 0, x, y, 0,
m_vram_read_texture.GetGLId(), m_vram_texture.GetGLTarget(), 0, x, y, 0, width, height, 1);
}
else
{
m_vram_read_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
glBindFramebuffer(GL_READ_FRAMEBUFFER, m_vram_fbo_id);
glDisable(GL_SCISSOR_TEST);
glBlitFramebuffer(x, y, x + width, y + height, x, y, x + width, y + height, GL_COLOR_BUFFER_BIT, GL_NEAREST);
glEnable(GL_SCISSOR_TEST);
glBindFramebuffer(GL_DRAW_FRAMEBUFFER, m_vram_fbo_id);
}
GPU_HW::UpdateVRAMReadTexture();
}

View File

@ -76,10 +76,6 @@ private:
bool BlitVRAMReplacementTexture(const TextureReplacementTexture* tex, u32 dst_x, u32 dst_y, u32 width, u32 height);
void DownsampleFramebuffer(GL::Texture& source, u32 left, u32 top, u32 width, u32 height);
void DownsampleFramebufferBoxFilter(GL::Texture& source, u32 left, u32 top, u32 width, u32 height);
void CopyTexture(GL::Texture& dest, GLuint dest_fbo, GL::Texture& src, GLuint src_fbo, u32 src_x, u32 src_y,
u32 dst_x, u32 dst_y, u32 width, u32 height);
void BlitTextureToFramebuffer(GL::Texture& src, u32 src_x, u32 src_y, u32 src_width, u32 src_height, u32 dst_x, u32 dst_y,
u32 dst_width, u32 dst_height);
// downsample texture - used for readbacks at >1xIR.
GL::Texture m_vram_texture;
@ -108,7 +104,6 @@ private:
GL::Program m_vram_write_program;
GL::Program m_vram_copy_program;
GL::Program m_vram_update_depth_program;
GL::Program m_blit_program;
u32 m_uniform_buffer_alignment = 1;
u32 m_texture_stream_buffer_size = 0;