X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=fft_pass_effect.cpp;h=b0006087cd2779282af2ec3750d5ba7761d084f5;hp=e3999b30682a1514b922ef64b825fe9096ab6065;hb=eff011224abc5dc81f801f3ea44572287a55bcac;hpb=60e4852ff1b04c525a9e3f1c98a1017db28b27bd

diff --git a/fft_pass_effect.cpp b/fft_pass_effect.cpp
index e3999b3..b000608 100644
--- a/fft_pass_effect.cpp
+++ b/fft_pass_effect.cpp
@@ -1,4 +1,4 @@
-#include <GL/glew.h>
+#include <epoxy/gl.h>
 #include <math.h>
 
 #include "effect_chain.h"
@@ -25,6 +25,8 @@ FFTPassEffect::FFTPassEffect()
 	register_int("direction", (int *)&direction);
 	register_int("pass_number", &pass_number);
 	register_int("inverse", &inverse);
+	register_uniform_float("num_repeats", &uniform_num_repeats);
+	register_uniform_sampler2d("support_tex", &uniform_support_tex);
 	glGenTextures(1, &tex);
 }
 
@@ -83,11 +85,11 @@ void FFTPassEffect::set_gl_state(GLuint glsl_program_num, const string &prefix,
 		generate_support_texture();
 	}
 
-	set_uniform_int(glsl_program_num, prefix, "support_tex", *sampler_num);
+	uniform_support_tex = *sampler_num;
 	++*sampler_num;
 
 	assert(input_size % fft_size == 0);
-	set_uniform_float(glsl_program_num, prefix, "num_repeats", input_size / fft_size);
+	uniform_num_repeats = input_size / fft_size;
 }
 
 void FFTPassEffect::generate_support_texture()
@@ -119,8 +121,8 @@ void FFTPassEffect::generate_support_texture()
 	// bit, so the stride is 8, and so on.
 
 	assert((fft_size & (fft_size - 1)) == 0);  // Must be power of two.
-	fp16_int_t *tmp = new fp16_int_t[fft_size * 4];
 	int subfft_size = 1 << pass_number;
+	fp16_int_t *tmp = new fp16_int_t[subfft_size * 4];
 	double mulfac;
 	if (inverse) {
 		mulfac = 2.0 * M_PI;
@@ -165,10 +167,10 @@ void FFTPassEffect::generate_support_texture()
 			support_texture_index = subfft_size - support_texture_index - 1;
 			sign = -1.0;
 		}
-		tmp[support_texture_index * 4 + 0] = fp64_to_fp16(sign * (src1 - i * stride) / double(input_size));
-		tmp[support_texture_index * 4 + 1] = fp64_to_fp16(sign * (src2 - i * stride) / double(input_size));
-		tmp[support_texture_index * 4 + 2] = fp64_to_fp16(twiddle_real);
-		tmp[support_texture_index * 4 + 3] = fp64_to_fp16(twiddle_imag);
+		tmp[support_texture_index * 4 + 0] = fp32_to_fp16(sign * (src1 - i * stride) / double(input_size));
+		tmp[support_texture_index * 4 + 1] = fp32_to_fp16(sign * (src2 - i * stride) / double(input_size));
+		tmp[support_texture_index * 4 + 2] = fp32_to_fp16(twiddle_real);
+		tmp[support_texture_index * 4 + 3] = fp32_to_fp16(twiddle_imag);
 	}
 
 	// Supposedly FFTs are very sensitive to inaccuracies in the twiddle factors,