From 2a6e04988f4c6ef97f541d84596fd120dd4d101e Mon Sep 17 00:00:00 2001
From: Connor McLaughlin <stenzek@gmail.com>
Date: Thu, 2 Apr 2020 00:17:25 +1000
Subject: [PATCH] GPU/HW: Simplify 24-bit scanout and interlacing shader
Has the added bonus of being faster (no downscale copy) and fixes the
edge case where the image is cropped in VRAM.
---
src/core/gpu.cpp | 4 --
src/core/gpu_hw_d3d11.cpp | 45 ++++++-------------
src/core/gpu_hw_opengl.cpp | 69 ++++++++---------------------
src/core/gpu_hw_opengl_es.cpp | 83 ++++++++++++-----------------------
src/core/gpu_hw_shadergen.cpp | 59 +++++++------------------
5 files changed, 75 insertions(+), 185 deletions(-)
diff --git a/src/core/gpu.cpp b/src/core/gpu.cpp
index bb0567864..fba1d8d11 100644
--- a/src/core/gpu.cpp
+++ b/src/core/gpu.cpp
@@ -440,10 +440,6 @@ void GPU::UpdateCRTCDisplayParameters()
cs.display_vram_left = std::min<u16>(
m_crtc_state.regs.X + ((horizontal_display_start_tick - cs.horizontal_display_start) / cs.dot_clock_divider),
VRAM_WIDTH - 1);
-
- // for 24-bit scanout we must stay aligned
- if (m_GPUSTAT.display_area_color_depth_24 && ((cs.display_vram_left - cs.regs.X) & 1u))
- cs.display_vram_left--;
}
if (cs.horizontal_display_end <= horizontal_display_end_tick)
diff --git a/src/core/gpu_hw_d3d11.cpp b/src/core/gpu_hw_d3d11.cpp
index 10250b646..da9a9c2ef 100644
--- a/src/core/gpu_hw_d3d11.cpp
+++ b/src/core/gpu_hw_d3d11.cpp
@@ -552,43 +552,24 @@ void GPU_HW_D3D11::UpdateDisplay()
}
else
{
- const u32 field_offset = BoolToUInt8(interlaced && m_GPUSTAT.interlaced_field);
+ m_context->OMSetRenderTargets(1, m_display_texture.GetD3DRTVArray(), nullptr);
+ m_context->PSSetShaderResources(0, 1, m_vram_texture.GetD3DSRVArray());
+ const u32 reinterpret_field_offset =
+ (m_crtc_state.regs.Y + BoolToUInt8(interlaced && m_GPUSTAT.interlaced_field)) & 1u;
+ const u32 reinterpret_start_x = m_crtc_state.regs.X * m_resolution_scale;
+ const u32 reinterpret_width = scaled_display_width + (m_crtc_state.display_vram_left - m_crtc_state.regs.X);
+ const u32 uniforms[4] = {reinterpret_field_offset, reinterpret_start_x};
ID3D11PixelShader* display_pixel_shader =
m_display_pixel_shaders[BoolToUInt8(m_GPUSTAT.display_area_color_depth_24)][BoolToUInt8(interlaced)].Get();
- // Because of how the reinterpret shader works, we need to use the downscaled version.
- if (m_GPUSTAT.display_area_color_depth_24 && m_resolution_scale > 1)
- {
- const u32 copy_width = std::min<u32>((display_width * 3) / 2, VRAM_WIDTH - vram_offset_x);
- const u32 scaled_copy_width = copy_width * m_resolution_scale;
- BlitTexture(m_vram_encoding_texture.GetD3DRTV(), vram_offset_x, vram_offset_y, copy_width, display_height,
- m_vram_texture.GetD3DSRV(), scaled_vram_offset_x, scaled_vram_offset_y, scaled_copy_width,
- scaled_display_height, m_vram_texture.GetWidth(), m_vram_texture.GetHeight(), false);
+ SetViewportAndScissor(reinterpret_start_x, m_crtc_state.display_vram_top, reinterpret_width,
+ scaled_display_height);
+ DrawUtilityShader(display_pixel_shader, uniforms, sizeof(uniforms));
- m_context->OMSetRenderTargets(1, m_display_texture.GetD3DRTVArray(), nullptr);
- m_context->PSSetShaderResources(0, 1, m_vram_encoding_texture.GetD3DSRVArray());
-
- const u32 uniforms[4] = {vram_offset_x, vram_offset_y, field_offset};
- SetViewportAndScissor(0, field_offset, display_width, display_height);
- DrawUtilityShader(display_pixel_shader, uniforms, sizeof(uniforms));
-
- m_host_display->SetDisplayTexture(m_display_texture.GetD3DSRV(), m_display_texture.GetWidth(),
- m_display_texture.GetHeight(), 0, 0, display_width, display_height);
- }
- else
- {
- m_context->OMSetRenderTargets(1, m_display_texture.GetD3DRTVArray(), nullptr);
- m_context->PSSetShaderResources(0, 1, m_vram_texture.GetD3DSRVArray());
-
- const u32 uniforms[4] = {scaled_vram_offset_x, scaled_vram_offset_y, field_offset};
- SetViewportAndScissor(0, field_offset, scaled_display_width, scaled_display_height);
- DrawUtilityShader(display_pixel_shader, uniforms, sizeof(uniforms));
-
- m_host_display->SetDisplayTexture(m_display_texture.GetD3DSRV(), m_display_texture.GetWidth(),
- m_display_texture.GetHeight(), 0, 0, scaled_display_width,
- scaled_display_height);
- }
+ m_host_display->SetDisplayTexture(m_display_texture.GetD3DSRV(), m_display_texture.GetWidth(),
+ m_display_texture.GetHeight(), scaled_vram_offset_x, scaled_vram_offset_y,
+ scaled_display_width, scaled_display_height);
RestoreGraphicsAPIState();
}
diff --git a/src/core/gpu_hw_opengl.cpp b/src/core/gpu_hw_opengl.cpp
index 534c16a08..159e88087 100644
--- a/src/core/gpu_hw_opengl.cpp
+++ b/src/core/gpu_hw_opengl.cpp
@@ -507,63 +507,32 @@ void GPU_HW_OpenGL::UpdateDisplay()
}
else
{
- const u32 flipped_vram_offset_y = VRAM_HEIGHT - vram_offset_y - display_height;
- const u32 scaled_flipped_vram_offset_y =
- m_vram_texture.GetHeight() - scaled_vram_offset_y - scaled_display_height;
- const u32 field_offset = BoolToUInt8(interlaced && m_GPUSTAT.interlaced_field);
-
glDisable(GL_BLEND);
glDisable(GL_SCISSOR_TEST);
- const GL::Program& prog =
- m_display_programs[BoolToUInt8(m_GPUSTAT.display_area_color_depth_24)][BoolToUInt8(interlaced)];
- prog.Bind();
+ m_display_programs[BoolToUInt8(m_GPUSTAT.display_area_color_depth_24)][BoolToUInt8(interlaced)].Bind();
+ m_display_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
+ m_vram_texture.Bind();
- // Because of how the reinterpret shader works, we need to use the downscaled version.
- if (m_GPUSTAT.display_area_color_depth_24 && m_resolution_scale > 1)
- {
- const u32 copy_width = std::min<u32>((display_width * 3) / 2, VRAM_WIDTH - vram_offset_x);
- const u32 scaled_copy_width = copy_width * m_resolution_scale;
- m_vram_encoding_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
- m_vram_texture.BindFramebuffer(GL_READ_FRAMEBUFFER);
- glBlitFramebuffer(scaled_vram_offset_x, scaled_flipped_vram_offset_y, scaled_vram_offset_x + scaled_copy_width,
- scaled_flipped_vram_offset_y + scaled_display_height, vram_offset_x, flipped_vram_offset_y,
- vram_offset_x + copy_width, flipped_vram_offset_y + display_height, GL_COLOR_BUFFER_BIT,
- GL_NEAREST);
+ const u32 flipped_vram_offset_y = VRAM_HEIGHT - vram_offset_y - display_height;
+ const u32 scaled_flipped_vram_offset_y =
+ m_vram_texture.GetHeight() - scaled_vram_offset_y - scaled_display_height;
- m_display_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
- m_vram_encoding_texture.Bind();
+ const u32 reinterpret_field_offset =
+ (m_crtc_state.regs.Y + BoolToUInt8(interlaced && m_GPUSTAT.interlaced_field)) & 1u;
+ const u32 reinterpret_start_x = m_crtc_state.regs.X * m_resolution_scale;
+ const u32 reinterpret_width = scaled_display_width + (m_crtc_state.display_vram_left - m_crtc_state.regs.X);
+ const u32 uniforms[4] = {reinterpret_field_offset, reinterpret_start_x};
+ UploadUniformBlock(uniforms, sizeof(uniforms));
+ m_batch_ubo_dirty = true;
- glViewport(0, field_offset, display_width, display_height);
+ glViewport(reinterpret_start_x, scaled_flipped_vram_offset_y, reinterpret_width, scaled_display_height);
+ glDrawArrays(GL_TRIANGLES, 0, 3);
- const u32 uniforms[4] = {vram_offset_x, flipped_vram_offset_y, field_offset};
- UploadUniformBlock(uniforms, sizeof(uniforms));
- m_batch_ubo_dirty = true;
-
- glDrawArrays(GL_TRIANGLES, 0, 3);
-
- m_host_display->SetDisplayTexture(reinterpret_cast<void*>(static_cast<uintptr_t>(m_display_texture.GetGLId())),
- m_display_texture.GetWidth(), m_display_texture.GetHeight(), 0,
- display_height, display_width, -static_cast<s32>(display_height));
- }
- else
- {
- m_display_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
- m_vram_texture.Bind();
-
- glViewport(0, field_offset, scaled_display_width, scaled_display_height);
-
- const u32 uniforms[4] = {scaled_vram_offset_x, scaled_flipped_vram_offset_y, field_offset};
- UploadUniformBlock(uniforms, sizeof(uniforms));
- m_batch_ubo_dirty = true;
-
- glDrawArrays(GL_TRIANGLES, 0, 3);
-
- m_host_display->SetDisplayTexture(reinterpret_cast<void*>(static_cast<uintptr_t>(m_display_texture.GetGLId())),
- m_display_texture.GetWidth(), m_display_texture.GetHeight(), 0,
- scaled_display_height, scaled_display_width,
- -static_cast<s32>(scaled_display_height));
- }
+ m_host_display->SetDisplayTexture(reinterpret_cast<void*>(static_cast<uintptr_t>(m_display_texture.GetGLId())),
+ m_display_texture.GetWidth(), m_display_texture.GetHeight(),
+ scaled_vram_offset_x, m_vram_texture.GetHeight() - scaled_vram_offset_y,
+ scaled_display_width, -static_cast<s32>(scaled_display_height));
// restore state
m_vram_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
diff --git a/src/core/gpu_hw_opengl_es.cpp b/src/core/gpu_hw_opengl_es.cpp
index 900f2ab88..c1a7c6ded 100644
--- a/src/core/gpu_hw_opengl_es.cpp
+++ b/src/core/gpu_hw_opengl_es.cpp
@@ -371,75 +371,46 @@ void GPU_HW_OpenGL_ES::UpdateDisplay()
}
else
{
- const u32 flipped_vram_offset_y = VRAM_HEIGHT - vram_offset_y - display_height;
- const u32 scaled_flipped_vram_offset_y =
- m_vram_texture.GetHeight() - scaled_vram_offset_y - scaled_display_height;
- const u32 field_offset = BoolToUInt8(interlaced && m_GPUSTAT.interlaced_field);
-
glDisable(GL_BLEND);
glDisable(GL_SCISSOR_TEST);
- const GL::Program& prog =
+ GL::Program& prog =
m_display_programs[BoolToUInt8(m_GPUSTAT.display_area_color_depth_24)][BoolToUInt8(interlaced)];
prog.Bind();
+ m_display_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
+ m_vram_texture.Bind();
- // Because of how the reinterpret shader works, we need to use the downscaled version.
- if (m_GPUSTAT.display_area_color_depth_24 && m_resolution_scale > 1)
- {
- const u32 copy_width = std::min<u32>((display_width * 3) / 2, VRAM_WIDTH - vram_offset_x);
- const u32 scaled_copy_width = copy_width * m_resolution_scale;
- m_vram_encoding_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
- m_vram_texture.BindFramebuffer(GL_READ_FRAMEBUFFER);
- glBlitFramebuffer(scaled_vram_offset_x, scaled_flipped_vram_offset_y, scaled_vram_offset_x + scaled_copy_width,
- scaled_flipped_vram_offset_y + scaled_display_height, vram_offset_x, flipped_vram_offset_y,
- vram_offset_x + copy_width, flipped_vram_offset_y + display_height, GL_COLOR_BUFFER_BIT,
- GL_NEAREST);
+ const u32 flipped_vram_offset_y = VRAM_HEIGHT - vram_offset_y - display_height;
+ const u32 scaled_flipped_vram_offset_y =
+ m_vram_texture.GetHeight() - scaled_vram_offset_y - scaled_display_height;
- m_display_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
- m_vram_encoding_texture.Bind();
+ const u32 reinterpret_field_offset =
+ (m_crtc_state.regs.Y + BoolToUInt8(interlaced && m_GPUSTAT.interlaced_field)) & 1u;
+ const u32 reinterpret_start_x = m_crtc_state.regs.X * m_resolution_scale;
+ const u32 reinterpret_width = scaled_display_width + (m_crtc_state.display_vram_left - m_crtc_state.regs.X);
- glViewport(0, field_offset, display_width, display_height);
+ prog.Uniform2i(0, reinterpret_field_offset, reinterpret_start_x);
+ m_batch_ubo_dirty = true;
- prog.Uniform3i(0, static_cast<s32>(vram_offset_x), static_cast<s32>(flipped_vram_offset_y),
- static_cast<s32>(field_offset));
- m_batch_ubo_dirty = true;
+ glViewport(reinterpret_start_x, scaled_flipped_vram_offset_y, reinterpret_width, scaled_display_height);
+ glDrawArrays(GL_TRIANGLES, 0, 3);
- glDrawArrays(GL_TRIANGLES, 0, 3);
-
- m_host_display->SetDisplayTexture(reinterpret_cast<void*>(static_cast<uintptr_t>(m_display_texture.GetGLId())),
- m_display_texture.GetWidth(), m_display_texture.GetHeight(), 0,
- display_height, display_width, -static_cast<s32>(display_height));
- }
- else
- {
- m_display_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
- m_vram_texture.Bind();
-
- glViewport(0, field_offset, scaled_display_width, scaled_display_height);
-
- prog.Uniform3i(0, static_cast<s32>(scaled_vram_offset_x), static_cast<s32>(scaled_flipped_vram_offset_y),
- static_cast<s32>(field_offset));
- m_batch_ubo_dirty = true;
-
- glDrawArrays(GL_TRIANGLES, 0, 3);
-
- m_host_display->SetDisplayTexture(reinterpret_cast<void*>(static_cast<uintptr_t>(m_display_texture.GetGLId())),
- m_display_texture.GetWidth(), m_display_texture.GetHeight(), 0,
- scaled_display_height, scaled_display_width,
- -static_cast<s32>(scaled_display_height));
- }
-
- // restore state
- m_vram_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
- glViewport(0, 0, m_vram_texture.GetWidth(), m_vram_texture.GetHeight());
- glEnable(GL_SCISSOR_TEST);
+ m_host_display->SetDisplayTexture(reinterpret_cast<void*>(static_cast<uintptr_t>(m_display_texture.GetGLId())),
+ m_display_texture.GetWidth(), m_display_texture.GetHeight(),
+ scaled_vram_offset_x, m_vram_texture.GetHeight() - scaled_vram_offset_y,
+ scaled_display_width, -static_cast<s32>(scaled_display_height));
}
- m_host_display->SetDisplayParameters(m_crtc_state.display_width, m_crtc_state.display_height,
- m_crtc_state.display_origin_left, m_crtc_state.display_origin_top,
- m_crtc_state.display_vram_width, m_crtc_state.display_vram_height,
- m_crtc_state.display_aspect_ratio);
+ // restore state
+ m_vram_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
+ glViewport(0, 0, m_vram_texture.GetWidth(), m_vram_texture.GetHeight());
+ glEnable(GL_SCISSOR_TEST);
}
+
+ m_host_display->SetDisplayParameters(m_crtc_state.display_width, m_crtc_state.display_height,
+ m_crtc_state.display_origin_left, m_crtc_state.display_origin_top,
+ m_crtc_state.display_vram_width, m_crtc_state.display_vram_height,
+ m_crtc_state.display_aspect_ratio);
}
void GPU_HW_OpenGL_ES::ReadVRAM(u32 x, u32 y, u32 width, u32 height)
diff --git a/src/core/gpu_hw_shadergen.cpp b/src/core/gpu_hw_shadergen.cpp
index 3bf86d29c..7189ba8fd 100644
--- a/src/core/gpu_hw_shadergen.cpp
+++ b/src/core/gpu_hw_shadergen.cpp
@@ -750,7 +750,7 @@ std::string GPU_HW_ShaderGen::GenerateDisplayFragmentShader(bool depth_24bit, bo
DefineMacro(ss, "INTERLACED", interlaced);
WriteCommonFunctions(ss);
- DeclareUniformBuffer(ss, {"int3 u_base_coords"});
+ DeclareUniformBuffer(ss, {"int u_field_offset", "int u_vram_start_x"});
DeclareTexture(ss, "samp0", 0);
DeclareFragmentEntryPoint(ss, 0, 1, {}, true, false);
@@ -759,55 +759,28 @@ std::string GPU_HW_ShaderGen::GenerateDisplayFragmentShader(bool depth_24bit, bo
int2 icoords = int2(v_pos.xy);
#if INTERLACED
- if (((icoords.y - u_base_coords.z) & 1) != 0)
+ if (((icoords.y / RESOLUTION_SCALE) & 1) != u_field_offset)
discard;
#endif
#if DEPTH_24BIT
- // compute offset in dwords from the start of the 24-bit values
- int2 base = int2(u_base_coords.x, u_base_coords.y + icoords.y);
- int xoff = int(icoords.x);
- int dword_index = (xoff / 2) + (xoff / 4);
+ // relative to start of scanout
+ int relative_x = (icoords.x - u_vram_start_x) / RESOLUTION_SCALE;
+ icoords.x = u_vram_start_x + ((relative_x * 3) / 2) * RESOLUTION_SCALE;
- // sample two adjacent dwords, or four 16-bit values as the 24-bit value will lie somewhere between these
- uint s0 = RGBA8ToRGBA5551(LOAD_TEXTURE(samp0, int2(base.x + dword_index * 2 + 0, base.y), 0));
- uint s1 = RGBA8ToRGBA5551(LOAD_TEXTURE(samp0, int2(base.x + dword_index * 2 + 1, base.y), 0));
- uint s2 = RGBA8ToRGBA5551(LOAD_TEXTURE(samp0, int2(base.x + (dword_index + 1) * 2 + 0, base.y), 0));
- uint s3 = RGBA8ToRGBA5551(LOAD_TEXTURE(samp0, int2(base.x + (dword_index + 1) * 2 + 1, base.y), 0));
-
- // select the bit for this pixel depending on its offset in the 4-pixel block
- uint r, g, b;
- int block_offset = xoff & 3;
- if (block_offset == 0)
- {
- r = s0 & 0xFFu;
- g = s0 >> 8;
- b = s1 & 0xFFu;
- }
- else if (block_offset == 1)
- {
- r = s1 >> 8;
- g = s2 & 0xFFu;
- b = s2 >> 8;
- }
- else if (block_offset == 2)
- {
- r = s1 & 0xFFu;
- g = s1 >> 8;
- b = s2 & 0xFFu;
- }
- else
- {
- r = s2 >> 8;
- g = s3 & 0xFFu;
- b = s3 >> 8;
- }
-
- // and normalize
- o_col0 = float4(float(r) / 255.0, float(g) / 255.0, float(b) / 255.0, 1.0);
+ // load adjacent 16-bit texels
+ uint s0 = RGBA8ToRGBA5551(LOAD_TEXTURE(samp0, icoords, 0));
+ uint s1 = RGBA8ToRGBA5551(LOAD_TEXTURE(samp0, icoords + int2(RESOLUTION_SCALE, 0), 0));
+
+ // select which part of the combined 16-bit texels we are currently shading
+ uint s1s0 = ((s1 << 16) | s0) >> ((relative_x & 1) * 8);
+
+ // extract components and normalize
+ o_col0 = float4(float(s1s0 & 0xFFu) / 255.0, float((s1s0 >> 8u) & 0xFFu) / 255.0,
+ float((s1s0 >> 16u) & 0xFFu) / 255.0, 1.0);
#else
// load and return
- o_col0 = LOAD_TEXTURE(samp0, u_base_coords.xy + icoords, 0);
+ o_col0 = LOAD_TEXTURE(samp0, icoords, 0);
#endif
}
)";