From 6d4c94c8a49e5637580c0b5ba30f97e34e5d2ff5 Mon Sep 17 00:00:00 2001
From: "Steinar H. Gunderson" <sgunderson@bigfoot.com>
Date: Sat, 4 Aug 2018 22:35:43 +0200
Subject: [PATCH] Fix a bug where the first black pass of SOR would read junk
 data.

---
 flow.cpp | 17 ++++++++++++-----
 sor.frag | 11 ++++++++---
 2 files changed, 20 insertions(+), 8 deletions(-)

diff --git a/flow.cpp b/flow.cpp
index 7658942..2ed5bb8 100644
--- a/flow.cpp
+++ b/flow.cpp
@@ -764,7 +764,7 @@ private:
 	GLuint uniform_diff_flow_tex;
 	GLuint uniform_equation_red_tex, uniform_equation_black_tex;
 	GLuint uniform_diffusivity_tex;
-	GLuint uniform_phase, uniform_zero_diff_flow;
+	GLuint uniform_phase, uniform_num_nonzero_phases;
 };
 
 SOR::SOR()
@@ -778,7 +778,7 @@ SOR::SOR()
 	uniform_equation_black_tex = glGetUniformLocation(sor_program, "equation_black_tex");
 	uniform_diffusivity_tex = glGetUniformLocation(sor_program, "diffusivity_tex");
 	uniform_phase = glGetUniformLocation(sor_program, "phase");
-	uniform_zero_diff_flow = glGetUniformLocation(sor_program, "zero_diff_flow");
+	uniform_num_nonzero_phases = glGetUniformLocation(sor_program, "num_nonzero_phases");
 }
 
 void SOR::exec(GLuint diff_flow_tex, GLuint equation_red_tex, GLuint equation_black_tex, GLuint diffusivity_tex, int level_width, int level_height, int num_iterations, bool zero_diff_flow, ScopedTimer *sor_timer)
@@ -790,7 +790,9 @@ void SOR::exec(GLuint diff_flow_tex, GLuint equation_red_tex, GLuint equation_bl
 	bind_sampler(sor_program, uniform_equation_red_tex, 2, equation_red_tex, nearest_sampler);
 	bind_sampler(sor_program, uniform_equation_black_tex, 3, equation_black_tex, nearest_sampler);
 
-	glProgramUniform1i(sor_program, uniform_zero_diff_flow, zero_diff_flow);
+	if (!zero_diff_flow) {
+		glProgramUniform1i(sor_program, uniform_num_nonzero_phases, 2);
+	}
 
 	// NOTE: We bind to the texture we are rendering from, but we never write any value
 	// that we read in the same shader pass (we call discard for red values when we compute
@@ -803,6 +805,9 @@ void SOR::exec(GLuint diff_flow_tex, GLuint equation_red_tex, GLuint equation_bl
 	for (int i = 0; i < num_iterations; ++i) {
 		{
 			ScopedTimer timer("Red pass", sor_timer);
+			if (zero_diff_flow && i == 0) {
+				glProgramUniform1i(sor_program, uniform_num_nonzero_phases, 0);
+			}
 			glProgramUniform1i(sor_program, uniform_phase, 0);
 			glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
 			glTextureBarrier();
@@ -810,11 +815,13 @@ void SOR::exec(GLuint diff_flow_tex, GLuint equation_red_tex, GLuint equation_bl
 		{
 			ScopedTimer timer("Black pass", sor_timer);
 			if (zero_diff_flow && i == 0) {
-				// Not zero anymore.
-				glProgramUniform1i(sor_program, uniform_zero_diff_flow, 0);
+				glProgramUniform1i(sor_program, uniform_num_nonzero_phases, 1);
 			}
 			glProgramUniform1i(sor_program, uniform_phase, 1);
 			glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
+			if (zero_diff_flow && i == 0) {
+				glProgramUniform1i(sor_program, uniform_num_nonzero_phases, 2);
+			}
 			if (i != num_iterations - 1) {
 				glTextureBarrier();
 			}
diff --git a/sor.frag b/sor.frag
index ef431d3..b072839 100644
--- a/sor.frag
+++ b/sor.frag
@@ -9,7 +9,7 @@ uniform sampler2D diff_flow_tex, diffusivity_tex;
 uniform usampler2D equation_red_tex, equation_black_tex;
 uniform int phase;
 
-uniform bool zero_diff_flow;
+uniform int num_nonzero_phases;
 
 // See pack_floats_shared() in equations.frag.
 vec2 unpack_floats_shared(uint c)
@@ -70,7 +70,7 @@ void main()
 
 	const float omega = 1.8;  // Marginally better than 1.6, it seems.
 
-	if (zero_diff_flow) {
+	if (num_nonzero_phases == 0) {
 		// Simplified version of the code below, assuming diff_flow == 0.0f everywhere.
 		diff_flow.x = omega * b.x * inv_A11;
 		diff_flow.y = omega * b.y * inv_A22;
@@ -87,7 +87,12 @@ void main()
 		b += smooth_r * textureOffset(diff_flow_tex, tc, ivec2( 1,  0)).xy;
 		b += smooth_d * textureOffset(diff_flow_tex, tc, ivec2( 0, -1)).xy;
 		b += smooth_u * textureOffset(diff_flow_tex, tc, ivec2( 0,  1)).xy;
-		diff_flow = texture(diff_flow_tex, tc).xy;
+
+		if (num_nonzero_phases == 1) {
+			diff_flow = vec2(0.0f);
+		} else {
+			diff_flow = texture(diff_flow_tex, tc).xy;
+		}
 
 		// From https://en.wikipedia.org/wiki/Successive_over-relaxation.
 		float sigma_u = A12 * diff_flow.y;
-- 
2.39.2