GPU/HW: Fallback path for VRAM writes for GPUs w/o texel buffers

src/core/gpu_hw_opengl.cpp

@@ -127,6 +127,10 @@ void GPU_HW_OpenGL::SetCapabilities()
 
   if (!GLAD_GL_VERSION_4_3 && !GLAD_GL_EXT_copy_image)
     Log_WarningPrintf("GL_EXT_copy_image missing, this may affect performance.");
+
+  glGetIntegerv(GL_MAX_TEXTURE_BUFFER_SIZE, reinterpret_cast<GLint*>(&m_max_texture_buffer_size));
+  if (m_max_texture_buffer_size < VRAM_WIDTH * VRAM_HEIGHT)
+    Log_WarningPrintf("Maximum texture buffer size is less than VRAM size, VRAM writes may be slower.");
 }
 
 void GPU_HW_OpenGL::CreateFramebuffer()
@@ -225,9 +229,8 @@ void GPU_HW_OpenGL::CreateUniformBuffer()
 
 void GPU_HW_OpenGL::CreateTextureBuffer()
 {
-  // const GLenum target = GL_PIXEL_UNPACK_BUFFER;
+  // We use the pixel unpack buffer here because we share it with CPU-decoded VRAM writes.
-  const GLenum target = GL_TEXTURE_BUFFER;
+  m_texture_stream_buffer = GL::StreamBuffer::Create(GL_PIXEL_UNPACK_BUFFER, VRAM_UPDATE_TEXTURE_BUFFER_SIZE);
-  m_texture_stream_buffer = GL::StreamBuffer::Create(target, VRAM_UPDATE_TEXTURE_BUFFER_SIZE);
   if (!m_texture_stream_buffer)
     Panic("Failed to create texture stream buffer");
 
@@ -573,7 +576,34 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
   GPU_HW::UpdateVRAM(x, y, width, height, data);
 
   const u32 num_pixels = width * height;
-#if 0
+  if (num_pixels < m_max_texture_buffer_size)
+  {
+    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));
+    m_texture_stream_buffer->Unmap(num_pixels * sizeof(u16));
+    m_texture_stream_buffer->Unbind();
+
+    // viewport should be set to the whole VRAM size, so we can just set the scissor
+    const u32 flipped_y = VRAM_HEIGHT - y - height;
+    const u32 scaled_width = width * m_resolution_scale;
+    const u32 scaled_height = height * m_resolution_scale;
+    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;
+    glScissor(scaled_x, scaled_flipped_y, scaled_width, scaled_height);
+
+    m_vram_write_program.Bind();
+    glBindTexture(GL_TEXTURE_BUFFER, m_texture_buffer_r16ui_texture);
+
+    const u32 uniforms[5] = {x, flipped_y, width, height, map_result.index_aligned};
+    UploadUniformBlock(uniforms, sizeof(uniforms));
+
+    glDrawArrays(GL_TRIANGLES, 0, 3);
+
+    RestoreGraphicsAPIState();
+  }
+  else
+  {
     const auto map_result = m_texture_stream_buffer->Map(sizeof(u32), num_pixels * sizeof(u32));
 
     // reverse copy the rows so it matches opengl's lower-left origin
@@ -623,35 +653,11 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
       const u32 scaled_flipped_y = m_vram_texture->GetHeight() - scaled_y - scaled_height;
       glDisable(GL_SCISSOR_TEST);
       m_vram_downsample_texture->BindFramebuffer(GL_READ_FRAMEBUFFER);
-    glBlitFramebuffer(x, flipped_y, x + width, flipped_y + height, scaled_x, scaled_flipped_y, scaled_x + scaled_width,
+      glBlitFramebuffer(x, flipped_y, x + width, flipped_y + height, scaled_x, scaled_flipped_y,
-                      scaled_flipped_y + scaled_height, GL_COLOR_BUFFER_BIT, GL_NEAREST);
+                        scaled_x + scaled_width, scaled_flipped_y + scaled_height, GL_COLOR_BUFFER_BIT, GL_NEAREST);
       glEnable(GL_SCISSOR_TEST);
     }
-#else
+  }
-  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));
-  m_texture_stream_buffer->Unmap(num_pixels * sizeof(u16));
-
-  // viewport should be set to the whole VRAM size, so we can just set the scissor
-  const u32 flipped_y = VRAM_HEIGHT - y - height;
-  const u32 scaled_width = width * m_resolution_scale;
-  const u32 scaled_height = height * m_resolution_scale;
-  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;
-  glScissor(scaled_x, scaled_flipped_y, scaled_width, scaled_height);
-
-  m_vram_write_program.Bind();
-  glBindTexture(GL_TEXTURE_BUFFER, m_texture_buffer_r16ui_texture);
-
-  const u32 uniforms[5] = {x, flipped_y, width, height, map_result.index_aligned};
-  UploadUniformBlock(uniforms, sizeof(uniforms));
-  m_batch_ubo_dirty = true;
-
-  glDrawArrays(GL_TRIANGLES, 0, 3);
-
-  SetScissorFromDrawingArea();
-#endif
 }
 
 void GPU_HW_OpenGL::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height)

src/core/gpu_hw_opengl.h

@@ -75,6 +75,7 @@ private:
   GLuint m_texture_buffer_r16ui_texture = 0;
 
   u32 m_uniform_buffer_alignment = 1;
+  u32 m_max_texture_buffer_size = 0;
 
   std::array<std::array<std::array<GL::Program, 2>, 9>, 4> m_render_programs; // [render_mode][texture_mode][dithering]
   std::array<std::array<GL::Program, 2>, 2> m_display_programs;               // [depth_24][interlaced]