Merge "Optimised hwui rounded corners shader" am: d7287c96eb am: ed7bdd6549

am: 92ef0a59f8

Change-Id: I9f5e6ef057ba6049804226d65f2f2f06594efc69

2 files changed, 29 insertions, 18 deletions

diff --git a/libs/hwui/ProgramCache.cpp b/libs/hwui/ProgramCache.cpp
index d0f0949d5e78..8cc0aa7b414c 100644
--- a/libs/hwui/ProgramCache.cpp
+++ b/libs/hwui/ProgramCache.cpp
@@ -58,7 +58,9 @@ const char* gVS_Header_Uniforms_HasBitmap =
         "uniform mat4 textureTransform;\n"
         "uniform mediump vec2 textureDimension;\n";
 const char* gVS_Header_Uniforms_HasRoundRectClip =
-        "uniform mat4 roundRectInvTransform;\n";
+        "uniform mat4 roundRectInvTransform;\n"
+        "uniform mediump vec4 roundRectInnerRectLTWH;\n"
+        "uniform mediump float roundRectRadius;\n";
 const char* gVS_Header_Varyings_HasTexture =
         "varying vec2 outTexCoords;\n";
 const char* gVS_Header_Varyings_HasColors =
@@ -81,7 +83,7 @@ const char* gVS_Header_Varyings_HasGradient[6] = {
         "varying highp vec2 sweep;\n",
 };
 const char* gVS_Header_Varyings_HasRoundRectClip =
-        "varying highp vec2 roundRectPos;\n";
+        "varying mediump vec2 roundRectPos;\n";
 const char* gVS_Main =
         "\nvoid main(void) {\n";
 const char* gVS_Main_OutTexCoords =
@@ -113,7 +115,7 @@ const char* gVS_Main_VertexAlpha =
         "    alpha = vtxAlpha;\n";
 
 const char* gVS_Main_HasRoundRectClip =
-        "    roundRectPos = (roundRectInvTransform * transformedPosition).xy;\n";
+        "    roundRectPos = ((roundRectInvTransform * transformedPosition).xy / roundRectRadius) - roundRectInnerRectLTWH.xy;\n";
 const char* gVS_Footer =
         "}\n\n";
 
@@ -158,8 +160,8 @@ const char* gFS_Uniforms_ColorOp[3] = {
 };
 
 const char* gFS_Uniforms_HasRoundRectClip =
-        "uniform vec4 roundRectInnerRectLTRB;\n"
-        "uniform float roundRectRadius;\n";
+        "uniform mediump vec4 roundRectInnerRectLTWH;\n"
+        "uniform mediump float roundRectRadius;\n";
 
 const char* gFS_Uniforms_ColorSpaceConversion =
         // TODO: Should we use a 3D LUT to combine the matrix and transfer functions?
@@ -431,15 +433,18 @@ const char* gFS_Main_ApplyColorOp[3] = {
         "    fragColor = blendColors(colorBlend, fragColor);\n"
 };
 
-// Note: LTRB -> xyzw
+// Note: LTWH (left top width height) -> xyzw
+// roundRectPos is now divided by roundRectRadius in vertex shader
+// after we also subtract roundRectInnerRectLTWH.xy from roundRectPos
 const char* gFS_Main_FragColor_HasRoundRectClip =
-        "    mediump vec2 fragToLT = roundRectInnerRectLTRB.xy - roundRectPos;\n"
-        "    mediump vec2 fragFromRB = roundRectPos - roundRectInnerRectLTRB.zw;\n"
-
-        // divide + multiply by 128 to avoid falling out of range in length() function
-        "    mediump vec2 dist = max(max(fragToLT, fragFromRB), vec2(0.0, 0.0)) / 128.0;\n"
-        "    mediump float linearDist = roundRectRadius - (length(dist) * 128.0);\n"
-        "    gl_FragColor *= clamp(linearDist, 0.0, 1.0);\n";
+        "    mediump vec2 fragToLT = -roundRectPos;\n"
+        "    mediump vec2 fragFromRB = roundRectPos - roundRectInnerRectLTWH.zw;\n"
+
+        // since distance is divided by radius, it's in [0;1] so precision is not an issue
+        // this also lets us clamp(0.0, 1.0) instead of max() which is cheaper on GPUs
+        "    mediump vec2 dist = clamp(max(fragToLT, fragFromRB), 0.0, 1.0);\n"
+        "    mediump float linearDist = clamp(roundRectRadius - (length(dist) * roundRectRadius), 0.0, 1.0);\n"
+        "    gl_FragColor *= linearDist;\n";
 
 const char* gFS_Main_DebugHighlight =
         "    gl_FragColor.rgb = vec3(0.0, gl_FragColor.a, 0.0);\n";
diff --git a/libs/hwui/renderstate/RenderState.cpp b/libs/hwui/renderstate/RenderState.cpp
index 481932dd7967..2c92924cc12c 100644
--- a/libs/hwui/renderstate/RenderState.cpp
+++ b/libs/hwui/renderstate/RenderState.cpp
@@ -296,14 +296,20 @@ void RenderState::render(const Glop& glop, const Matrix4& orthoMatrix) {
         // TODO: avoid query, and cache values (or RRCS ptr) in program
         const RoundRectClipState* state = glop.roundRectClipState;
         const Rect& innerRect = state->innerRect;
-        glUniform4f(fill.program->getUniform("roundRectInnerRectLTRB"),
-                innerRect.left, innerRect.top,
-                innerRect.right, innerRect.bottom);
-        glUniformMatrix4fv(fill.program->getUniform("roundRectInvTransform"),
-                1, GL_FALSE, &state->matrix.data[0]);
 
         // add half pixel to round out integer rect space to cover pixel centers
         float roundedOutRadius = state->radius + 0.5f;
+
+        // Divide by the radius to simplify the calculations in the fragment shader
+        // roundRectPos is also passed from vertex shader relative to top/left & radius
+        glUniform4f(fill.program->getUniform("roundRectInnerRectLTWH"),
+                innerRect.left / roundedOutRadius, innerRect.top / roundedOutRadius,
+                (innerRect.right - innerRect.left) / roundedOutRadius,
+                (innerRect.bottom - innerRect.top) / roundedOutRadius);
+
+        glUniformMatrix4fv(fill.program->getUniform("roundRectInvTransform"),
+                1, GL_FALSE, &state->matrix.data[0]);
+
         glUniform1f(fill.program->getUniform("roundRectRadius"),
                 roundedOutRadius);
     }