Make quant_matrix a bit more compact.

diff --git a/encoder.shader b/encoder.shader
index da9e6c07b2c730274513087a6ffaa5239146b29e..163f8fa0110aed25ff9654156f115b54167d95e5 100644 (file)
--- a/encoder.shader
+++ b/encoder.shader
@@ -49,15 +49,16 @@ const float W[64] = {
 const float S = 4.0 * 0.5;  // whatever?
 
 // NOTE: Contains factors to counteract the scaling in the DCT implementation.
+#define QM(x, y) (sf[x] * sf[y] / (W[y*8 + x] * S))
 const float quant_matrix[64] = {
-        sf[0] * sf[0] / 64.0,         sf[1] * sf[0] / (W[ 1] * S),  sf[2] * sf[0] / (W[ 2] * S),  sf[3] * sf[0] / (W[ 3] * S),  sf[4] * sf[0] / (W[ 4] * S),  sf[5] * sf[0] / (W[ 5] * S),  sf[6] * sf[0] / (W[ 6] * S),  sf[7] * sf[0] / (W[ 7] * S),
-        sf[0] * sf[1] / (W[ 8] * S),  sf[1] * sf[1] / (W[ 9] * S),  sf[2] * sf[1] / (W[10] * S),  sf[3] * sf[1] / (W[11] * S),  sf[4] * sf[1] / (W[12] * S),  sf[5] * sf[1] / (W[13] * S),  sf[6] * sf[1] / (W[14] * S),  sf[7] * sf[1] / (W[15] * S),
-        sf[0] * sf[2] / (W[16] * S),  sf[1] * sf[2] / (W[17] * S),  sf[2] * sf[2] / (W[18] * S),  sf[3] * sf[2] / (W[19] * S),  sf[4] * sf[2] / (W[20] * S),  sf[5] * sf[2] / (W[21] * S),  sf[6] * sf[2] / (W[22] * S),  sf[7] * sf[2] / (W[23] * S),
-        sf[0] * sf[3] / (W[24] * S),  sf[1] * sf[3] / (W[25] * S),  sf[2] * sf[3] / (W[26] * S),  sf[3] * sf[3] / (W[27] * S),  sf[4] * sf[3] / (W[28] * S),  sf[5] * sf[3] / (W[29] * S),  sf[6] * sf[3] / (W[30] * S),  sf[7] * sf[3] / (W[31] * S),
-        sf[0] * sf[4] / (W[32] * S),  sf[1] * sf[4] / (W[33] * S),  sf[2] * sf[4] / (W[34] * S),  sf[3] * sf[4] / (W[35] * S),  sf[4] * sf[4] / (W[36] * S),  sf[5] * sf[4] / (W[37] * S),  sf[6] * sf[4] / (W[38] * S),  sf[7] * sf[4] / (W[39] * S),
-        sf[0] * sf[5] / (W[40] * S),  sf[1] * sf[5] / (W[41] * S),  sf[2] * sf[5] / (W[42] * S),  sf[3] * sf[5] / (W[43] * S),  sf[4] * sf[5] / (W[44] * S),  sf[5] * sf[5] / (W[45] * S),  sf[6] * sf[5] / (W[46] * S),  sf[7] * sf[5] / (W[47] * S),
-        sf[0] * sf[6] / (W[48] * S),  sf[1] * sf[6] / (W[49] * S),  sf[2] * sf[6] / (W[50] * S),  sf[3] * sf[6] / (W[51] * S),  sf[4] * sf[6] / (W[52] * S),  sf[5] * sf[6] / (W[53] * S),  sf[6] * sf[6] / (W[54] * S),  sf[7] * sf[6] / (W[55] * S),
-        sf[0] * sf[7] / (W[56] * S),  sf[1] * sf[7] / (W[57] * S),  sf[2] * sf[7] / (W[58] * S),  sf[3] * sf[7] / (W[59] * S),  sf[4] * sf[7] / (W[60] * S),  sf[5] * sf[7] / (W[61] * S),  sf[6] * sf[7] / (W[62] * S),  sf[7] * sf[7] / (W[63] * S)
+        1.0 / 64.0, QM(1, 0), QM(2, 0), QM(3, 0), QM(4, 0), QM(5, 0), QM(6, 0), QM(7, 0),
+        QM(0, 1),   QM(1, 1), QM(2, 1), QM(3, 1), QM(4, 1), QM(5, 1), QM(6, 1), QM(7, 1),
+        QM(0, 2),   QM(1, 2), QM(2, 2), QM(3, 2), QM(4, 2), QM(5, 2), QM(6, 2), QM(7, 2),
+        QM(0, 3),   QM(1, 3), QM(2, 3), QM(3, 3), QM(4, 3), QM(5, 3), QM(6, 3), QM(7, 3),
+        QM(0, 4),   QM(1, 4), QM(2, 4), QM(3, 4), QM(4, 4), QM(5, 4), QM(6, 4), QM(7, 4),
+        QM(0, 5),   QM(1, 5), QM(2, 5), QM(3, 5), QM(4, 5), QM(5, 5), QM(6, 5), QM(7, 5),
+        QM(0, 6),   QM(1, 6), QM(2, 6), QM(3, 6), QM(4, 6), QM(5, 6), QM(6, 6), QM(7, 6),
+        QM(0, 7),   QM(1, 7), QM(2, 7), QM(3, 7), QM(4, 7), QM(5, 7), QM(6, 7), QM(7, 7)
 };
 
 // Clamp and pack a 9-bit and a 7-bit signed value into a 16-bit word.