GPUShader: Add GPU_SHADER_2D_IMAGE_MULTISAMPLE_2/4/8/16

This shader is used instead of blitting back and forth to a single sample
buffer.

This means it resolves the color and depth samples and outputs a fragment
which can be depth tested and blended on top of an existing framebuffer.

We do static shader variation with manual loop unrolling for performance
reason. In my test I get 25% more perf with intel integrated gpu and 75%
performance gain with dedicated nvidia card compared to a single shader
with a uniform for sample count.

4 files changed, 89 insertions, 0 deletions

diff --git a/source/blender/gpu/CMakeLists.txt b/source/blender/gpu/CMakeLists.txt
index 575dff1e811..f733e37ff51 100644
--- a/source/blender/gpu/CMakeLists.txt
+++ b/source/blender/gpu/CMakeLists.txt
@@ -158,6 +158,7 @@ data_to_c_simple(shaders/gpu_shader_image_varying_color_frag.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_image_depth_linear_frag.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_image_depth_copy_frag.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_image_interlace_frag.glsl SRC)
+data_to_c_simple(shaders/gpu_shader_image_multisample_resolve_frag.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_3D_image_vert.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_3D_vert.glsl SRC)
 data_to_c_simple(shaders/gpu_shader_3D_normal_vert.glsl SRC)
diff --git a/source/blender/gpu/GPU_shader.h b/source/blender/gpu/GPU_shader.h
index 59bf92aa672..6d75447c8a3 100644
--- a/source/blender/gpu/GPU_shader.h
+++ b/source/blender/gpu/GPU_shader.h
@@ -119,6 +119,10 @@ typedef enum GPUBuiltinShader {
 	GPU_SHADER_2D_IMAGE_ALPHA,
 	GPU_SHADER_2D_IMAGE_RECT_COLOR,
 	GPU_SHADER_2D_IMAGE_MULTI_RECT_COLOR,
+	GPU_SHADER_2D_IMAGE_MULTISAMPLE_2,
+	GPU_SHADER_2D_IMAGE_MULTISAMPLE_4,
+	GPU_SHADER_2D_IMAGE_MULTISAMPLE_8,
+	GPU_SHADER_2D_IMAGE_MULTISAMPLE_16,
 	GPU_SHADER_2D_CHECKER,
 	GPU_SHADER_2D_DIAG_STRIPES,
 	/* for simple 3D drawing */
diff --git a/source/blender/gpu/intern/gpu_shader.c b/source/blender/gpu/intern/gpu_shader.c
index 34dee37001e..26267b95f59 100644
--- a/source/blender/gpu/intern/gpu_shader.c
+++ b/source/blender/gpu/intern/gpu_shader.c
@@ -90,6 +90,7 @@ extern char datatoc_gpu_shader_image_mask_uniform_color_frag_glsl[];
 extern char datatoc_gpu_shader_image_modulate_alpha_frag_glsl[];
 extern char datatoc_gpu_shader_image_depth_linear_frag_glsl[];
 extern char datatoc_gpu_shader_image_depth_copy_frag_glsl[];
+extern char datatoc_gpu_shader_image_multisample_resolve_frag_glsl[];
 extern char datatoc_gpu_shader_3D_vert_glsl[];
 extern char datatoc_gpu_shader_3D_normal_vert_glsl[];
 extern char datatoc_gpu_shader_3D_flat_color_vert_glsl[];
@@ -690,6 +691,10 @@ GPUShader *GPU_shader_get_builtin_shader(GPUBuiltinShader shader)
 		                                datatoc_gpu_shader_image_depth_linear_frag_glsl },
 		[GPU_SHADER_3D_IMAGE_DEPTH_COPY] = { datatoc_gpu_shader_3D_image_vert_glsl,
 		                                     datatoc_gpu_shader_image_depth_copy_frag_glsl },
+		[GPU_SHADER_2D_IMAGE_MULTISAMPLE_2] = { datatoc_gpu_shader_2D_vert_glsl, datatoc_gpu_shader_image_multisample_resolve_frag_glsl },
+		[GPU_SHADER_2D_IMAGE_MULTISAMPLE_4] = { datatoc_gpu_shader_2D_vert_glsl, datatoc_gpu_shader_image_multisample_resolve_frag_glsl },
+		[GPU_SHADER_2D_IMAGE_MULTISAMPLE_8] = { datatoc_gpu_shader_2D_vert_glsl, datatoc_gpu_shader_image_multisample_resolve_frag_glsl },
+		[GPU_SHADER_2D_IMAGE_MULTISAMPLE_16] = { datatoc_gpu_shader_2D_vert_glsl, datatoc_gpu_shader_image_multisample_resolve_frag_glsl },
 
 		[GPU_SHADER_2D_IMAGE_INTERLACE] = { datatoc_gpu_shader_2D_image_vert_glsl,
 		                                    datatoc_gpu_shader_image_interlace_frag_glsl },
@@ -830,6 +835,18 @@ GPUShader *GPU_shader_get_builtin_shader(GPUBuiltinShader shader)
 		/* just a few special cases */
 		const char *defines = NULL;
 		switch (shader) {
+			case GPU_SHADER_2D_IMAGE_MULTISAMPLE_2:
+				defines = "#define SAMPLES 2\n";
+				break;
+			case GPU_SHADER_2D_IMAGE_MULTISAMPLE_4:
+				defines = "#define SAMPLES 4\n";
+				break;
+			case GPU_SHADER_2D_IMAGE_MULTISAMPLE_8:
+				defines = "#define SAMPLES 8\n";
+				break;
+			case GPU_SHADER_2D_IMAGE_MULTISAMPLE_16:
+				defines = "#define SAMPLES 16\n";
+				break;
 			case GPU_SHADER_2D_WIDGET_BASE_INST:
 			case GPU_SHADER_2D_NODELINK_INST:
 				defines = "#define USE_INSTANCE\n";
diff --git a/source/blender/gpu/shaders/gpu_shader_image_multisample_resolve_frag.glsl b/source/blender/gpu/shaders/gpu_shader_image_multisample_resolve_frag.glsl
new file mode 100644
index 00000000000..de1fd8b6b58
--- /dev/null
+++ b/source/blender/gpu/shaders/gpu_shader_image_multisample_resolve_frag.glsl
@@ -0,0 +1,67 @@
+
+uniform sampler2DMS depthMulti;
+uniform sampler2DMS colorMulti;
+
+out vec4 fragColor;
+
+#if SAMPLES > 16
+#error "Too many samples"
+#endif
+
+void main()
+{
+	ivec2 texel = ivec2(gl_FragCoord.xy);
+
+	float depth = 1.0;
+	depth = min(depth, texelFetch(depthMulti, texel, 0).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 1).r);
+#if SAMPLES > 2
+	depth = min(depth, texelFetch(depthMulti, texel, 2).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 3).r);
+#endif
+#if SAMPLES > 4
+	depth = min(depth, texelFetch(depthMulti, texel, 4).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 5).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 6).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 7).r);
+#endif
+#if SAMPLES > 8
+	depth = min(depth, texelFetch(depthMulti, texel, 8).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 9).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 10).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 11).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 12).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 13).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 14).r);
+	depth = min(depth, texelFetch(depthMulti, texel, 15).r);
+#endif
+
+	vec4 color = vec4(0.0);
+	color += texelFetch(colorMulti, texel, 0);
+	color += texelFetch(colorMulti, texel, 1);
+#if SAMPLES > 2
+	color += texelFetch(colorMulti, texel, 2);
+	color += texelFetch(colorMulti, texel, 3);
+#endif
+#if SAMPLES > 4
+	color += texelFetch(colorMulti, texel, 4);
+	color += texelFetch(colorMulti, texel, 5);
+	color += texelFetch(colorMulti, texel, 6);
+	color += texelFetch(colorMulti, texel, 7);
+#endif
+#if SAMPLES > 8
+	color += texelFetch(colorMulti, texel, 8);
+	color += texelFetch(colorMulti, texel, 9);
+	color += texelFetch(colorMulti, texel, 10);
+	color += texelFetch(colorMulti, texel, 11);
+	color += texelFetch(colorMulti, texel, 12);
+	color += texelFetch(colorMulti, texel, 13);
+	color += texelFetch(colorMulti, texel, 14);
+	color += texelFetch(colorMulti, texel, 15);
+#endif
+
+	const float inv_samples = 1.0 / float(SAMPLES);
+
+	fragColor = color * inv_samples;
+	gl_FragDepth = depth;
+}