From 8025def8d25e8df32719ecfbf6d93069685ab182 Mon Sep 17 00:00:00 2001 From: "Unknown W. Brackets" Date: Tue, 4 Oct 2022 22:10:24 -0700 Subject: [PATCH] Vulkan: Clip to neg z in the geometry shader. This is only used when clip distance is unsupported, such as on Mali. --- GPU/Common/GeometryShaderGenerator.cpp | 152 ++++++++++++++++++++++--- GPU/Common/VertexShaderGenerator.cpp | 2 +- 2 files changed, 135 insertions(+), 19 deletions(-) diff --git a/GPU/Common/GeometryShaderGenerator.cpp b/GPU/Common/GeometryShaderGenerator.cpp index b701bb547c73..786fd2573403 100644 --- a/GPU/Common/GeometryShaderGenerator.cpp +++ b/GPU/Common/GeometryShaderGenerator.cpp @@ -48,7 +48,7 @@ bool GenerateGeometryShader(const GShaderID &id, char *buffer, const ShaderLangu ShaderWriter p(buffer, compat, ShaderStage::Geometry, gl_exts.data(), gl_exts.size()); p.C("layout(triangles) in;\n"); - p.C("layout(triangle_strip, max_vertices = 3) out;\n"); + p.C("layout(triangle_strip, max_vertices = 6) out;\n"); if (compat.shaderLanguage == GLSL_VULKAN) { WRITE(p, "\n"); @@ -76,6 +76,10 @@ bool GenerateGeometryShader(const GShaderID &id, char *buffer, const ShaderLangu // Apply culling p.C(" bool anyInside = false;\n"); + // And apply manual clipping if necessary. + if (!gstate_c.Supports(GPU_SUPPORTS_CLIP_DISTANCE)) { + p.C(" float clip0[3];\n"); + } p.C(" for (int i = 0; i < 3; i++) {\n"); // TODO: 3 or gl_in.length()? which will be faster? p.C(" vec4 outPos = gl_in[i].gl_Position;\n"); @@ -98,6 +102,13 @@ bool GenerateGeometryShader(const GShaderID &id, char *buffer, const ShaderLangu p.C(" if (projPos.z >= u_cullRangeMin.z) { anyInside = true; }\n"); p.C(" if (projPos.z <= u_cullRangeMax.z) { anyInside = true; }\n"); p.C(" }\n"); + + if (!gstate_c.Supports(GPU_SUPPORTS_CLIP_DISTANCE)) { + // This is basically the same value as gl_ClipDistance would take, z + w. + // TODO: Ignore triangles from GE_PRIM_RECTANGLES in transform mode, which should not clip to neg z. + p.F(" clip0[i] = projZ * outPos.w + outPos.w;\n"); + } + p.C(" } // for\n"); // Cull any triangle fully outside in the same direction when depth clamp enabled. @@ -106,27 +117,132 @@ bool GenerateGeometryShader(const GShaderID &id, char *buffer, const ShaderLangu p.C(" return;\n"); p.C(" }\n"); - const char *clip0 = compat.shaderLanguage == HLSL_D3D11 ? "" : "[0]"; - - p.C(" for (int i = 0; i < 3; i++) {\n"); // TODO: 3 or gl_in.length()? which will be faster? - p.C(" vec4 outPos = gl_in[i].gl_Position;\n"); - p.C(" gl_Position = outPos;\n"); - // TODO: Not rectangles... - if (gstate_c.Supports(GPU_SUPPORTS_CLIP_DISTANCE)) { + if (!gstate_c.Supports(GPU_SUPPORTS_CLIP_DISTANCE)) { + // Clipping against one half-space cuts a triangle (17/27), culls (7/27), or creates two triangles (3/27). + p.C(" int emitted = 0;\n"); + p.C(" for (int i = 0; i < 3; i++) {\n"); + // First, emit this vertex if it doesn't need clipping + p.C(" if (clip0[i] >= 0.0) {\n"); + + // But before we emit that, is this the second triangle? We'll need extra verts. + p.C(" if (emitted == 3) {\n"); + p.C(" EndPrimitive();\n"); + // In this case, it can only be +/-/+, we must emit vert 0 and then mix(1, 2) again first. + p.C(" gl_Position = gl_in[0].gl_Position;\n"); + for (size_t i = 0; i < varyings.size(); i++) { + VaryingDef &in = varyings[i]; + VaryingDef &out = outVaryings[i]; + p.F(" %s = %s[0];\n", outVaryings[i].name, varyings[i].name); + } + p.C(" EmitVertex();\n"); + p.C(" emitted++;\n"); + // Next, it can only be the interpolated between 1 and 2 (before this one, direct 2.) + p.C(" float t = clip0[1] / (clip0[1] - clip0[2]);\n"); + p.C(" gl_Position = mix(gl_in[1].gl_Position, gl_in[2].gl_Position, t);\n"); + for (size_t i = 0; i < varyings.size(); i++) { + VaryingDef &in = varyings[i]; + VaryingDef &out = outVaryings[i]; + p.F(" %s = mix(%s[1], %s[2], t);\n", outVaryings[i].name, varyings[i].name, varyings[i].name); + } + p.C(" EmitVertex();\n"); + p.C(" emitted++;\n"); + p.C(" }\n"); + + // Now emit the regular vertex itself. + p.C(" gl_Position = gl_in[i].gl_Position;\n"); + for (size_t i = 0; i < varyings.size(); i++) { + VaryingDef &in = varyings[i]; + VaryingDef &out = outVaryings[i]; + p.F(" %s = %s[i];\n", outVaryings[i].name, varyings[i].name); + } + p.C(" EmitVertex();\n"); + p.C(" emitted++;\n"); + p.C(" }\n"); + + // Next, we generate an interpolated vertex if signs differ. + p.C(" int inext = i == 2 ? 0 : i + 1;\n"); + p.C(" if (clip0[i] * clip0[inext] < 0.0) {\n"); + + // There are two cases here: +/+/- and -/+/+. + p.C(" if (emitted == 3 && clip0[i] < 0.0) {\n"); + p.C(" EndPrimitive();\n"); + // In this case, it can only be +/-/+, we must emit vert 0 and then mix(1, 2) again first. + p.C(" gl_Position = gl_in[0].gl_Position;\n"); + for (size_t i = 0; i < varyings.size(); i++) { + VaryingDef &in = varyings[i]; + VaryingDef &out = outVaryings[i]; + p.F(" %s = %s[0];\n", outVaryings[i].name, varyings[i].name); + } + p.C(" EmitVertex();\n"); + p.C(" emitted++;\n"); + // Next, it can only be the interpolated between 1 and 2 (before this one, direct 2.) + p.C(" float t = 1.0 - (clip0[2] / (clip0[2] - clip0[1]));\n"); + p.C(" gl_Position = mix(gl_in[1].gl_Position, gl_in[2].gl_Position, t);\n"); + for (size_t i = 0; i < varyings.size(); i++) { + VaryingDef &in = varyings[i]; + VaryingDef &out = outVaryings[i]; + p.F(" %s = mix(%s[1], %s[2], t);\n", outVaryings[i].name, varyings[i].name, varyings[i].name); + } + p.C(" EmitVertex();\n"); + p.C(" emitted++;\n"); + p.C(" } else if (emitted == 3) {\n"); + p.C(" EndPrimitive();\n"); + // Now we emit mix(0, 1) first, then vert 2 again. + p.C(" float t = clip0[0] / (clip0[0] - clip0[1]);\n"); + p.C(" gl_Position = mix(gl_in[0].gl_Position, gl_in[1].gl_Position, t);\n"); + for (size_t i = 0; i < varyings.size(); i++) { + VaryingDef &in = varyings[i]; + VaryingDef &out = outVaryings[i]; + p.F(" %s = mix(%s[0], %s[1], t);\n", outVaryings[i].name, varyings[i].name, varyings[i].name); + } + p.C(" EmitVertex();\n"); + p.C(" emitted++;\n"); + // Then here's vert 2. + p.C(" gl_Position = gl_in[2].gl_Position;\n"); + for (size_t i = 0; i < varyings.size(); i++) { + VaryingDef &in = varyings[i]; + VaryingDef &out = outVaryings[i]; + p.F(" %s = %s[2];\n", outVaryings[i].name, varyings[i].name); + } + p.C(" EmitVertex();\n"); + p.C(" emitted++;\n"); + p.C(" }\n"); + + // Finally, the actual interpolated vertex. + p.C(" float t = clip0[i] < 0.0 ? clip0[i] / (clip0[i] - clip0[inext]) : 1.0 - (clip0[inext] / (clip0[inext] - clip0[i]));\n"); + p.C(" gl_Position = mix(gl_in[i].gl_Position, gl_in[inext].gl_Position, t);\n"); + for (size_t i = 0; i < varyings.size(); i++) { + VaryingDef &in = varyings[i]; + VaryingDef &out = outVaryings[i]; + p.F(" %s = mix(%s[i], %s[inext], t);\n", outVaryings[i].name, varyings[i].name, varyings[i].name); + } + p.C(" EmitVertex();\n"); + p.C(" emitted++;\n"); + p.C(" }\n"); + p.C(" }\n"); + } else { + const char *clipSuffix0 = compat.shaderLanguage == HLSL_D3D11 ? "" : "[0]"; + + p.C(" for (int i = 0; i < 3; i++) {\n"); // TODO: 3 or gl_in.length()? which will be faster? + p.C(" vec4 outPos = gl_in[i].gl_Position;\n"); p.C(" vec3 projPos = outPos.xyz / outPos.w;\n"); p.C(" float projZ = (projPos.z - u_depthRange.z) * u_depthRange.w;\n"); - p.F(" gl_ClipDistance%s = projZ * outPos.w + outPos.w;\n", clip0); - } + // TODO: Ignore triangles from GE_PRIM_RECTANGLES in transform mode, which should not clip to neg z. + p.F(" gl_ClipDistance%s = projZ * outPos.w + outPos.w;\n", clipSuffix0); + p.C(" gl_Position = outPos;\n"); + if (gstate_c.Supports(GPU_SUPPORTS_CLIP_DISTANCE)) { + } - for (size_t i = 0; i < varyings.size(); i++) { - VaryingDef &in = varyings[i]; - VaryingDef &out = outVaryings[i]; - p.F(" %s = %s[i];\n", outVaryings[i].name, varyings[i].name); + for (size_t i = 0; i < varyings.size(); i++) { + VaryingDef &in = varyings[i]; + VaryingDef &out = outVaryings[i]; + p.F(" %s = %s[i];\n", outVaryings[i].name, varyings[i].name); + } + // Debug - null the red channel + //p.C(" if (i == 0) v_color0Out.x = 0.0;\n"); + p.C(" EmitVertex();\n"); + p.C(" }\n"); } - // Debug - null the red channel - //p.C(" if (i == 0) v_color0Out.x = 0.0;\n"); - p.C(" EmitVertex();\n"); - p.C(" }\n"); p.EndGSMain(); diff --git a/GPU/Common/VertexShaderGenerator.cpp b/GPU/Common/VertexShaderGenerator.cpp index d6f356c6d4dd..928aa3f6ddd9 100644 --- a/GPU/Common/VertexShaderGenerator.cpp +++ b/GPU/Common/VertexShaderGenerator.cpp @@ -1312,7 +1312,7 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag const char *cull0 = compat.shaderLanguage == HLSL_D3D11 ? ".x" : "[0]"; const char *cull1 = compat.shaderLanguage == HLSL_D3D11 ? ".y" : "[1]"; if (gstate_c.Supports(GPU_SUPPORTS_CLIP_DISTANCE)) { - // TODO: Not rectangles... + // TODO: Ignore triangles from GE_PRIM_RECTANGLES in transform mode, which should not clip to neg z. WRITE(p, " %sgl_ClipDistance%s = projZ * outPos.w + outPos.w;\n", compat.vsOutPrefix, vertexRangeClipSuffix); } if (gstate_c.Supports(GPU_SUPPORTS_CULL_DISTANCE)) {