X-Git-Url: https://git.sesse.net/?p=fjl;a=blobdiff_plain;f=idct.c;h=ad1cc364642582b1711fb57122a7df186214f72c;hp=b187a90c07996b4af5e8a6cd0ba5d954688b4293;hb=13b48c4e2de5ceb4922ee47db32dd1a18edfffe4;hpb=4b87e80c1ee4dd6a5d5c13e7b4321a956f53378f

diff --git a/idct.c b/idct.c
index b187a90..ad1cc36 100644
--- a/idct.c
+++ b/idct.c
@@ -1,9 +1,27 @@
 #include <math.h>
+#include <string.h>
+#include <stdlib.h>
 
 #include "idct.h"
 
-void idct_reference(const int16_t* input, const uint32_t* quant_table, uint8_t* output)
+void* idct_reference_alloc(const uint32_t* quant_table)
 {
+	uint32_t* qt_copy = (uint32_t*)malloc(DCTSIZE2 * sizeof(uint32_t));
+	// FIXME: check for NULL return
+
+	memcpy(qt_copy, quant_table, DCTSIZE2 * sizeof(uint32_t));
+
+	return qt_copy;
+}
+
+void idct_reference_free(void* userdata)
+{
+	free(userdata);
+}
+
+void idct_reference(const int16_t* input, const void* userdata, uint8_t* output)
+{
+	const uint32_t* quant_table = (const uint32_t*)userdata;
 	double temp[DCTSIZE2];
 
 	for (unsigned y = 0; y < 8; ++y) {
@@ -56,6 +74,27 @@ static const double scalefac[] = {
 	1.0, 1.2727585805728336, 1.847759065022573, 3.6245097854115502
 };
 
+// Premultiply the scale factors and the overall 1/8 factor into the quantization
+// table entries (and convert to double).
+void* idct_float_alloc(const uint32_t* quant_table)
+{
+	double* qt_copy = (double*)malloc(DCTSIZE2 * sizeof(double));
+
+	for (unsigned y = 0; y < DCTSIZE; ++y) {
+		for (unsigned x = 0; x < DCTSIZE; ++x) {
+			qt_copy[y * DCTSIZE + x] = 0.125 * quant_table[y * DCTSIZE + x] *
+				scalefac[x] * scalefac[y];
+		}
+	}
+
+	return qt_copy;
+}
+
+void idct_float_free(void* userdata)
+{
+	free(userdata);
+}
+
 // 1D 8-point DCT.
 static inline void idct1d_float(double y0, double y1, double y2, double y3, double y4, double y5, double y6, double y7, double *x)
 {
@@ -67,14 +106,14 @@ static inline void idct1d_float(double y0, double y1, double y2, double y3, doub
 	static const double a5 = 0.5 * (a4 - a2);
 
 	// phase 1
-	const double p1_0 = y0 * scalefac[0];
-	const double p1_1 = y4 * scalefac[4];
-	const double p1_2 = y2 * scalefac[2];
-	const double p1_3 = y6 * scalefac[6];
-	const double p1_4 = y5 * scalefac[5];
-	const double p1_5 = y1 * scalefac[1];
-	const double p1_6 = y7 * scalefac[7];
-	const double p1_7 = y3 * scalefac[3];
+	const double p1_0 = y0;
+	const double p1_1 = y4;
+	const double p1_2 = y2;
+	const double p1_3 = y6;
+	const double p1_4 = y5;
+	const double p1_5 = y1;
+	const double p1_6 = y7;
+	const double p1_7 = y3;
 
 	// phase 2
 	const double p2_0 = p1_0;
@@ -137,20 +176,21 @@ static inline void idct1d_float(double y0, double y1, double y2, double y3, doub
 	x[7] = p6_0 - p6_7;
 }
 
-void idct_float(const int16_t* input, const uint32_t* quant_table, uint8_t* output)
+void idct_float(const int16_t* input, const void* userdata, uint8_t* output)
 {
+	const double* quant_table = (const double*)userdata;
 	double temp[DCTSIZE2];
 
 	// IDCT columns.
 	for (unsigned x = 0; x < DCTSIZE; ++x) {
-		idct1d_float(input[DCTSIZE * 0 + x] * (int32_t)quant_table[DCTSIZE * 0 + x],
-		             input[DCTSIZE * 1 + x] * (int32_t)quant_table[DCTSIZE * 1 + x],
-		             input[DCTSIZE * 2 + x] * (int32_t)quant_table[DCTSIZE * 2 + x],
-		             input[DCTSIZE * 3 + x] * (int32_t)quant_table[DCTSIZE * 3 + x],
-		             input[DCTSIZE * 4 + x] * (int32_t)quant_table[DCTSIZE * 4 + x],
-		             input[DCTSIZE * 5 + x] * (int32_t)quant_table[DCTSIZE * 5 + x],
-		             input[DCTSIZE * 6 + x] * (int32_t)quant_table[DCTSIZE * 6 + x],
-		             input[DCTSIZE * 7 + x] * (int32_t)quant_table[DCTSIZE * 7 + x],
+		idct1d_float(input[DCTSIZE * 0 + x] * quant_table[DCTSIZE * 0 + x],
+		             input[DCTSIZE * 1 + x] * quant_table[DCTSIZE * 1 + x],
+		             input[DCTSIZE * 2 + x] * quant_table[DCTSIZE * 2 + x],
+		             input[DCTSIZE * 3 + x] * quant_table[DCTSIZE * 3 + x],
+		             input[DCTSIZE * 4 + x] * quant_table[DCTSIZE * 4 + x],
+		             input[DCTSIZE * 5 + x] * quant_table[DCTSIZE * 5 + x],
+		             input[DCTSIZE * 6 + x] * quant_table[DCTSIZE * 6 + x],
+		             input[DCTSIZE * 7 + x] * quant_table[DCTSIZE * 7 + x],
 		             temp + x * DCTSIZE);
 	}
 	
@@ -167,7 +207,7 @@ void idct_float(const int16_t* input, const uint32_t* quant_table, uint8_t* outp
 		             temp[DCTSIZE * 7 + y],
 		             temp2);
 		for (unsigned x = 0; x < DCTSIZE; ++x) {
-			double val = (1.0/8.0) * temp2[x];
+			const double val = temp2[x];
 			if (val < 0.0) {
 				output[y * DCTSIZE + x] = 0;
 			} else if (val >= 255.0) {