Reduce the amount of arithmetic in the BlurEffect shader a bit.

We did additions and subtractions with zero, which is sort of a waste
on scalar architectures. Helps ever so slightly on the demo app on my NVidia
card (3–4%).

diff --git a/blur_effect.cpp b/blur_effect.cpp
index 9397ca423d429853a9a21c470217447ec688f2a3..d43a7caec77bcf34ce5944f5f332016bc2d740f0 100644 (file)
--- a/blur_effect.cpp
+++ b/blur_effect.cpp
@@ -108,7 +108,9 @@ SingleBlurPassEffect::SingleBlurPassEffect(BlurEffect *parent)
 
 string SingleBlurPassEffect::output_fragment_shader()
 {
 
 string SingleBlurPassEffect::output_fragment_shader()
 {

-       return read_file("blur_effect.frag");
+       char buf[256];
+       sprintf(buf, "#define DIRECTION_VERTICAL %d\n", (direction == VERTICAL));
+       return buf + read_file("blur_effect.frag");

 }
 
 void SingleBlurPassEffect::set_gl_state(GLuint glsl_program_num, const string &prefix, unsigned *sampler_num)
 }
 
 void SingleBlurPassEffect::set_gl_state(GLuint glsl_program_num, const string &prefix, unsigned *sampler_num)


@@ -159,13 +161,11 @@ void SingleBlurPassEffect::set_gl_state(GLuint glsl_program_num, const string &p
        //
        // We pack the parameters into a float4: The relative sample coordinates
        // in (x,y), and the weight in z. w is unused.
        //
        // We pack the parameters into a float4: The relative sample coordinates
        // in (x,y), and the weight in z. w is unused.

-       float samples[4 * (NUM_TAPS / 2 + 1)];
+       float samples[2 * (NUM_TAPS / 2 + 1)];

 
        // Center sample.
 
        // Center sample.

-       samples[4 * 0 + 0] = 0.0f;
-       samples[4 * 0 + 1] = 0.0f;
-       samples[4 * 0 + 2] = weight[0];
-       samples[4 * 0 + 3] = 0.0f;
+       samples[2 * 0 + 0] = 0.0f;
+       samples[2 * 0 + 1] = weight[0];

 
        // All other samples.
        for (unsigned i = 1; i < NUM_TAPS / 2 + 1; ++i) {
 
        // All other samples.
        for (unsigned i = 1; i < NUM_TAPS / 2 + 1; ++i) {


@@ -176,23 +176,18 @@ void SingleBlurPassEffect::set_gl_state(GLuint glsl_program_num, const string &p
                float offset, total_weight;
                combine_two_samples(w1, w2, &offset, &total_weight, NULL);
 
                float offset, total_weight;
                combine_two_samples(w1, w2, &offset, &total_weight, NULL);
 

-               float x = 0.0f, y = 0.0f;
-

                if (direction == HORIZONTAL) {
                if (direction == HORIZONTAL) {

-                       x = (base_pos + offset) / (float)width;
+                       samples[2 * i + 0] = (base_pos + offset) / (float)width;

                } else if (direction == VERTICAL) {
                } else if (direction == VERTICAL) {

-                       y = (base_pos + offset) / (float)height;
+                       samples[2 * i + 0] = (base_pos + offset) / (float)height;

                } else {
                        assert(false);
                }
 
                } else {
                        assert(false);
                }
 

-               samples[4 * i + 0] = x;
-               samples[4 * i + 1] = y;
-               samples[4 * i + 2] = total_weight;
-               samples[4 * i + 3] = 0.0f;
+               samples[2 * i + 1] = total_weight;

        }
 
        }
 

-       set_uniform_vec4_array(glsl_program_num, prefix, "samples", samples, NUM_TAPS / 2 + 1);
+       set_uniform_vec2_array(glsl_program_num, prefix, "samples", samples, NUM_TAPS / 2 + 1);

 }
 
 void SingleBlurPassEffect::clear_gl_state()
 }
 
 void SingleBlurPassEffect::clear_gl_state()

diff --git a/blur_effect.frag b/blur_effect.frag
index c6e4cf5465e7cbf62178edd8e231d8c962b8c1d2..8853854229038c067b0ab8f677008a24d67e64b7 100644 (file)
--- a/blur_effect.frag
+++ b/blur_effect.frag
@@ -1,14 +1,24 @@
-// A simple unidirectional blur.
+// A simple un.directional blur.
+// DIRECTION_VERTICAL will be #defined to 1 if we are doing a vertical blur,
+// 0 otherwise.

 
 #define NUM_TAPS 16
 
 
 #define NUM_TAPS 16
 

-uniform vec4 PREFIX(samples)[NUM_TAPS + 1];
+uniform vec2 PREFIX(samples)[NUM_TAPS + 1];

 
 vec4 FUNCNAME(vec2 tc) {
 
 vec4 FUNCNAME(vec2 tc) {

-       vec4 sum = vec4(PREFIX(samples)[0].z) * INPUT(tc);
+       vec4 sum = vec4(PREFIX(samples)[0].y) * INPUT(tc);

        for (int i = 1; i < NUM_TAPS + 1; ++i) {
        for (int i = 1; i < NUM_TAPS + 1; ++i) {

-               vec4 sample = PREFIX(samples)[i];
-               sum += vec4(sample.z) * (INPUT(tc - sample.xy) + INPUT(tc + sample.xy));
+               vec2 sample = PREFIX(samples)[i];
+               vec2 sample1_tc = tc, sample2_tc = tc;
+#if DIRECTION_VERTICAL
+               sample1_tc.y -= sample.x;
+               sample2_tc.y += sample.x;
+#else
+               sample1_tc.x -= sample.x;
+               sample2_tc.x += sample.x;
+#endif
+               sum += vec4(sample.y) * (INPUT(sample1_tc) + INPUT(sample2_tc));

        }
        return sum;
 }
        }
        return sum;
 }

diff --git a/effect_util.cpp b/effect_util.cpp
index 35a94a7c3e999b0ebc2795f92b40e0d10d0009cb..d485d4691cc3d16a04ce1643a485f94504664cd4 100644 (file)
--- a/effect_util.cpp
+++ b/effect_util.cpp
@@ -69,6 +69,17 @@ void set_uniform_vec4(GLuint glsl_program_num, const string &prefix, const strin
        check_error();
 }
 
        check_error();
 }
 

+void set_uniform_vec2_array(GLuint glsl_program_num, const string &prefix, const string &key, const float *values, size_t num_values)
+{
+       GLint location = get_uniform_location(glsl_program_num, prefix, key);
+       if (location == -1) {
+               return;
+       }
+       check_error();
+       glUniform2fv(location, num_values, values);
+       check_error();
+}
+

 void set_uniform_vec4_array(GLuint glsl_program_num, const string &prefix, const string &key, const float *values, size_t num_values)
 {
        GLint location = get_uniform_location(glsl_program_num, prefix, key);
 void set_uniform_vec4_array(GLuint glsl_program_num, const string &prefix, const string &key, const float *values, size_t num_values)
 {
        GLint location = get_uniform_location(glsl_program_num, prefix, key);

diff --git a/effect_util.h b/effect_util.h
index 65bd315855c5f004b2ec73bc6dc73f3b3c96aaec..5420ce516518a1cb7e9fef2f6aa8bf81c3ea28f5 100644 (file)
--- a/effect_util.h
+++ b/effect_util.h
@@ -26,6 +26,7 @@ void set_uniform_float(GLuint glsl_program_num, const std::string &prefix, const
 void set_uniform_vec2(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const float *values);
 void set_uniform_vec3(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const float *values);
 void set_uniform_vec4(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const float *values);
 void set_uniform_vec2(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const float *values);
 void set_uniform_vec3(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const float *values);
 void set_uniform_vec4(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const float *values);

+void set_uniform_vec2_array(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const float *values, size_t num_values);

 void set_uniform_vec4_array(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const float *values, size_t num_values);
 void set_uniform_mat3(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const Eigen::Matrix3d &matrix);
 
 void set_uniform_vec4_array(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const float *values, size_t num_values);
 void set_uniform_mat3(GLuint glsl_program_num, const std::string &prefix, const std::string &key, const Eigen::Matrix3d &matrix);