X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=util.cpp;h=b12ee80712ba83e541f28472760167e5e8fb1d22;hp=7fbc67a07206b81df5a196c580e704a1ba369444;hb=185ced44b129739c9b5438da691e71d664d6443a;hpb=fe7a508885d19950c43e46dc60db675375f49ffb

diff --git a/util.cpp b/util.cpp
index 7fbc67a..b12ee80 100644
--- a/util.cpp
+++ b/util.cpp
@@ -158,8 +158,9 @@ string output_glsl_mat3(const string &name, const Eigen::Matrix3d &m)
 	return buf;
 }
 
-void combine_two_samples(float w1, float w2, float pos1, float pos2, unsigned size, CombineRoundingBehavior rounding_behavior,
-                         float *offset, float *total_weight, float *sum_sq_error)
+template<class DestFloat>
+void combine_two_samples(float w1, float w2, float pos1, float pos2, unsigned size,
+                         DestFloat *offset, DestFloat *total_weight, float *sum_sq_error)
 {
 	assert(movit_initialized);
 	assert(w1 * w2 >= 0.0f);  // Should not have differing signs.
@@ -170,14 +171,9 @@ void combine_two_samples(float w1, float w2, float pos1, float pos2, unsigned si
 		z = w2 / (w1 + w2);
 	}
 
-	*offset = pos1 + z * (pos2 - pos1);
-	if (rounding_behavior == COMBINE_ROUND_TO_FP16) {	
-		// Round to fp16. Note that this might take z outside the 0..1 range.
-		*offset = fp16_to_fp64(fp64_to_fp16(*offset));
-		z = (*offset - pos1) / (pos2 - pos1);
-	} else {
-		assert(rounding_behavior == COMBINE_DO_NOT_ROUND);
-	}
+	// Round to the desired precision. Note that this might take z outside the 0..1 range.
+	*offset = from_fp64<DestFloat>(pos1 + z * (pos2 - pos1));
+	z = (to_fp64(*offset) - pos1) / (pos2 - pos1);
 
 	// Round to the minimum number of bits we have measured earlier.
 	// The card will do this for us anyway, but if we know what the real z
@@ -208,6 +204,15 @@ void combine_two_samples(float w1, float w2, float pos1, float pos2, unsigned si
 	}
 }
 
+// Explicit instantiations.
+template
+void combine_two_samples<float>(float w1, float w2, float pos1, float pos2, unsigned size,
+                                float *offset, float *total_weight, float *sum_sq_error);
+
+template
+void combine_two_samples<fp16_int_t>(float w1, float w2, float pos1, float pos2, unsigned size,
+                                     fp16_int_t *offset, fp16_int_t *total_weight, float *sum_sq_error);
+
 GLuint fill_vertex_attribute(GLuint glsl_program_num, const string &attribute_name, GLint size, GLenum type, GLsizeiptr data_size, const GLvoid *data)
 {
 	int attrib = glGetAttribLocation(glsl_program_num, attribute_name.c_str());