doing blur shader without alpha blending

src/render/d3d11/generated/render_d3d11.meta.h

@@ -227,6 +227,7 @@ str8_lit_comp(
 "  float4 position        : SV_POSITION;\n"
 "  float2 texcoord        : TEX;\n"
 "  float2 sdf_sample_pos  : SDF;\n"
+"  nointerpolation float2 rect_half_size : RHS;\n"
 "  float corner_radius    : RAD;\n"
 "};\n"
 "\n"
@@ -271,6 +272,7 @@ str8_lit_comp(
 "    v2p.position = float4(vertex_position__scr.x, -vertex_position__scr.y, 0.f, 1.f);\n"
 "    v2p.texcoord = vertex_position__pct;\n"
 "    v2p.sdf_sample_pos = (2.f * cornercoords__pct - 1.f) * rect_half_size;\n"
+"    v2p.rect_half_size = rect_half_size - 2.f;\n"
 "    v2p.corner_radius = corner_radii__px[c2v.vertex_id];\n"
 "  }\n"
 "  return v2p;\n"
@@ -282,33 +284,28 @@ str8_lit_comp(
 "ps_main(Vertex2Pixel v2p) : SV_TARGET\n"
 "{\n"
 "  // rjf: blend weighted texture samples into color\n"
-"  float4 color = kernel[0].x * stage_t2d.Sample(stage_sampler, v2p.texcoord);\n"
-"  color.a = kernel[0].x;\n"
+"  float3 color = kernel[0].x * stage_t2d.Sample(stage_sampler, v2p.texcoord).rgb;\n"
 "  \n"
 "  for(uint i = 1; i < blur_count; i += 1)\n"
 "  {\n"
 "    float weight = kernel[i].x;\n"
 "    float offset = kernel[i].y;\n"
-"    float4 min_sample = stage_t2d.Sample(stage_sampler, v2p.texcoord - offset * direction);\n"
-"    float4 max_sample = stage_t2d.Sample(stage_sampler, v2p.texcoord + offset * direction);\n"
-"    min_sample.a = 1.f;\n"
-"    max_sample.a = 1.f;\n"
-"    color += min_sample * weight;\n"
-"    color += max_sample * weight;\n"
+"    color += weight * stage_t2d.Sample(stage_sampler, v2p.texcoord - offset * direction).rgb;\n"
+"    color += weight * stage_t2d.Sample(stage_sampler, v2p.texcoord + offset * direction).rgb;\n"
 "  }\n"
 "  \n"
-"  // rjf: determine SDF sample position\n"
-"  float2 rect_half_size = float2((rect.z-rect.x)/2, (rect.w-rect.y)/2);\n"
-"  float2 sdf_sample_pos = v2p.sdf_sample_pos;\n"
-"  \n"
 "  // rjf: sample for corners\n"
-"  float corner_sdf_s = rect_sdf(sdf_sample_pos, rect_half_size - 2.f, v2p.corner_radius);\n"
+"  float corner_sdf_s = rect_sdf(v2p.sdf_sample_pos, v2p.rect_half_size, v2p.corner_radius);\n"
 "  float corner_sdf_t = 1-smoothstep(0, 2, corner_sdf_s);\n"
 "  \n"
 "  // rjf: weight output color by sdf\n"
-"  color.a *= corner_sdf_t;\n"
+"  // this is doing alpha testing, leave blurring only where mostly opaque pixels are\n"
+"  if (corner_sdf_t < 0.9f)\n"
+"  {\n"
+"    discard;\n"
+"  }\n"
 "  \n"
-"  return color;\n"
+"  return float4(color, 1.f);\n"
 "}\n"
 ""
 );

src/render/d3d11/render_d3d11.cpp

@@ -232,6 +232,15 @@ r_init(CmdLine *cmdln)
     error = r_d3d11_state->device->CreateBlendState(&desc, &r_d3d11_state->main_blend_state);
   }
 
+  {
+    D3D11_BLEND_DESC desc = {0};
+    {
+      desc.RenderTarget[0].BlendEnable           = FALSE;
+      desc.RenderTarget[0].RenderTargetWriteMask = D3D11_COLOR_WRITE_ENABLE_ALL;
+    }
+    error = r_d3d11_state->device->CreateBlendState(&desc, &r_d3d11_state->no_blend_state);
+  }
+
   //- rjf: create nearest-neighbor sampler
   {
     D3D11_SAMPLER_DESC desc = zero_struct;
@@ -1190,7 +1199,7 @@ r_window_submit(OS_Handle window, R_Handle window_equip, R_PassList *passes)
 
           // rjf: setup output merger
           d_ctx->OMSetDepthStencilState(r_d3d11_state->noop_depth_stencil, 0);
-          d_ctx->OMSetBlendState(r_d3d11_state->main_blend_state, 0, 0xffffffff);
+          d_ctx->OMSetBlendState(r_d3d11_state->no_blend_state, 0, 0xffffffff);
 
           // rjf: set up viewport
           Vec2S32 resolution = wnd->last_resolution;

src/render/d3d11/render_d3d11.h

@@ -134,6 +134,7 @@ struct R_D3D11_State
   IDXGIFactory2           *dxgi_factory;
   ID3D11RasterizerState1  *main_rasterizer;
   ID3D11BlendState        *main_blend_state;
+  ID3D11BlendState        *no_blend_state;
   ID3D11SamplerState      *samplers[R_Tex2DSampleKind_COUNT];
   ID3D11DepthStencilState *noop_depth_stencil;
   ID3D11DepthStencilState *plain_depth_stencil;

src/render/d3d11/render_d3d11.mdesk

@@ -226,6 +226,7 @@ struct Vertex2Pixel
   float4 position        : SV_POSITION;
   float2 texcoord        : TEX;
   float2 sdf_sample_pos  : SDF;
+  nointerpolation float2 rect_half_size : RHS;
   float corner_radius    : RAD;
 };
 
@@ -270,6 +271,7 @@ vs_main(CPU2Vertex c2v)
     v2p.position = float4(vertex_position__scr.x, -vertex_position__scr.y, 0.f, 1.f);
     v2p.texcoord = vertex_position__pct;
     v2p.sdf_sample_pos = (2.f * cornercoords__pct - 1.f) * rect_half_size;
+    v2p.rect_half_size = rect_half_size - 2.f;
     v2p.corner_radius = corner_radii__px[c2v.vertex_id];
   }
   return v2p;
@@ -281,33 +283,28 @@ float4
 ps_main(Vertex2Pixel v2p) : SV_TARGET
 {
   // rjf: blend weighted texture samples into color
-  float4 color = kernel[0].x * stage_t2d.Sample(stage_sampler, v2p.texcoord);
-  color.a = kernel[0].x;
+  float3 color = kernel[0].x * stage_t2d.Sample(stage_sampler, v2p.texcoord).rgb;
 
   for(uint i = 1; i < blur_count; i += 1)
   {
     float weight = kernel[i].x;
     float offset = kernel[i].y;
-    float4 min_sample = stage_t2d.Sample(stage_sampler, v2p.texcoord - offset * direction);
-    float4 max_sample = stage_t2d.Sample(stage_sampler, v2p.texcoord + offset * direction);
-    min_sample.a = 1.f;
-    max_sample.a = 1.f;
-    color += min_sample * weight;
-    color += max_sample * weight;
+    color += weight * stage_t2d.Sample(stage_sampler, v2p.texcoord - offset * direction).rgb;
+    color += weight * stage_t2d.Sample(stage_sampler, v2p.texcoord + offset * direction).rgb;
   }
 
-  // rjf: determine SDF sample position
-  float2 rect_half_size = float2((rect.z-rect.x)/2, (rect.w-rect.y)/2);
-  float2 sdf_sample_pos = v2p.sdf_sample_pos;
-
   // rjf: sample for corners
-  float corner_sdf_s = rect_sdf(sdf_sample_pos, rect_half_size - 2.f, v2p.corner_radius);
+  float corner_sdf_s = rect_sdf(v2p.sdf_sample_pos, v2p.rect_half_size, v2p.corner_radius);
   float corner_sdf_t = 1-smoothstep(0, 2, corner_sdf_s);
 
   // rjf: weight output color by sdf
-  color.a *= corner_sdf_t;
+  // this is doing alpha testing, leave blurring only where mostly opaque pixels are
+  if (corner_sdf_t < 0.9f)
+  {
+    discard;
+  }
 
-  return color;
+  return float4(color, 1.f);
 }
 """