Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(434)

Issues Search
    My Issues | Starred     Open | Closed | All

Unified Diff: jidctred.c

Issue 1934113002: Update libjpeg_turbo to 1.4.90 from https://github.com/libjpeg-turbo/ (Closed) Base URL: https://chromium.googlesource.com/chromium/deps/libjpeg_turbo.git@master
Patch Set: Created 4 years, 8 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« jconfigint.h ('K') | « jidctint.c ('k') | jinclude.h » ('j') | simd/jccolext-sse2-64.asm » ('J')
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: jidctred.c
diff --git a/jidctred.c b/jidctred.c
index 421f3c7ca1ed2ebc9335aeb29c990a08840a5b1d..2d5b5466ac7c8a6e7493dec65ecddf8c70ccf339 100644
--- a/jidctred.c
+++ b/jidctred.c
@@ -1,9 +1,12 @@
/*
* jidctred.c
*
+ * This file was part of the Independent JPEG Group's software.
* Copyright (C) 1994-1998, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2015, D. R. Commander
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
*
* This file contains inverse-DCT routines that produce reduced-size output:
* either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block.
@@ -23,7 +26,7 @@
#define JPEG_INTERNALS
#include "jinclude.h"
#include "jpeglib.h"
-#include "jdct.h" /* Private declarations for DCT subsystem */
+#include "jdct.h" /* Private declarations for DCT subsystem */
#ifdef IDCT_SCALING_SUPPORTED
@@ -44,7 +47,7 @@
#define PASS1_BITS 2
#else
#define CONST_BITS 13
-#define PASS1_BITS 1 /* lose a little precision to avoid overflow */
+#define PASS1_BITS 1 /* lose a little precision to avoid overflow */
#endif
/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
@@ -55,20 +58,20 @@
*/
#if CONST_BITS == 13
-#define FIX_0_211164243 ((INT32) 1730) /* FIX(0.211164243) */
-#define FIX_0_509795579 ((INT32) 4176) /* FIX(0.509795579) */
-#define FIX_0_601344887 ((INT32) 4926) /* FIX(0.601344887) */
-#define FIX_0_720959822 ((INT32) 5906) /* FIX(0.720959822) */
-#define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */
-#define FIX_0_850430095 ((INT32) 6967) /* FIX(0.850430095) */
-#define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */
-#define FIX_1_061594337 ((INT32) 8697) /* FIX(1.061594337) */
-#define FIX_1_272758580 ((INT32) 10426) /* FIX(1.272758580) */
-#define FIX_1_451774981 ((INT32) 11893) /* FIX(1.451774981) */
-#define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */
-#define FIX_2_172734803 ((INT32) 17799) /* FIX(2.172734803) */
-#define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */
-#define FIX_3_624509785 ((INT32) 29692) /* FIX(3.624509785) */
+#define FIX_0_211164243 ((JLONG) 1730) /* FIX(0.211164243) */
+#define FIX_0_509795579 ((JLONG) 4176) /* FIX(0.509795579) */
+#define FIX_0_601344887 ((JLONG) 4926) /* FIX(0.601344887) */
+#define FIX_0_720959822 ((JLONG) 5906) /* FIX(0.720959822) */
+#define FIX_0_765366865 ((JLONG) 6270) /* FIX(0.765366865) */
+#define FIX_0_850430095 ((JLONG) 6967) /* FIX(0.850430095) */
+#define FIX_0_899976223 ((JLONG) 7373) /* FIX(0.899976223) */
+#define FIX_1_061594337 ((JLONG) 8697) /* FIX(1.061594337) */
+#define FIX_1_272758580 ((JLONG) 10426) /* FIX(1.272758580) */
+#define FIX_1_451774981 ((JLONG) 11893) /* FIX(1.451774981) */
+#define FIX_1_847759065 ((JLONG) 15137) /* FIX(1.847759065) */
+#define FIX_2_172734803 ((JLONG) 17799) /* FIX(2.172734803) */
+#define FIX_2_562915447 ((JLONG) 20995) /* FIX(2.562915447) */
+#define FIX_3_624509785 ((JLONG) 29692) /* FIX(3.624509785) */
#else
#define FIX_0_211164243 FIX(0.211164243)
#define FIX_0_509795579 FIX(0.509795579)
@@ -87,7 +90,7 @@
#endif
-/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
+/* Multiply a JLONG variable by a JLONG constant to yield a JLONG result.
* For 8-bit samples with the recommended scaling, all the variable
* and constant values involved are no more than 16 bits wide, so a
* 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
@@ -115,19 +118,19 @@
*/
GLOBAL(void)
-jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
- JCOEFPTR coef_block,
- JSAMPARRAY output_buf, JDIMENSION output_col)
+jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block,
+ JSAMPARRAY output_buf, JDIMENSION output_col)
{
- INT32 tmp0, tmp2, tmp10, tmp12;
- INT32 z1, z2, z3, z4;
+ JLONG tmp0, tmp2, tmp10, tmp12;
+ JLONG z1, z2, z3, z4;
JCOEFPTR inptr;
- ISLOW_MULT_TYPE * quantptr;
- int * wsptr;
+ ISLOW_MULT_TYPE *quantptr;
+ int *wsptr;
JSAMPROW outptr;
JSAMPLE *range_limit = IDCT_range_limit(cinfo);
int ctr;
- int workspace[DCTSIZE*4]; /* buffers data between passes */
+ int workspace[DCTSIZE*4]; /* buffers data between passes */
SHIFT_TEMPS
/* Pass 1: process columns from input, store into work array. */
@@ -140,57 +143,58 @@ jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
if (ctr == DCTSIZE-4)
continue;
if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
- inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*5] == 0 &&
- inptr[DCTSIZE*6] == 0 && inptr[DCTSIZE*7] == 0) {
+ inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*5] == 0 &&
+ inptr[DCTSIZE*6] == 0 && inptr[DCTSIZE*7] == 0) {
/* AC terms all zero; we need not examine term 4 for 4x4 output */
- int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
-
+ int dcval = LEFT_SHIFT(DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]),
+ PASS1_BITS);
+
wsptr[DCTSIZE*0] = dcval;
wsptr[DCTSIZE*1] = dcval;
wsptr[DCTSIZE*2] = dcval;
wsptr[DCTSIZE*3] = dcval;
-
+
continue;
}
-
+
/* Even part */
-
+
tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
- tmp0 <<= (CONST_BITS+1);
-
+ tmp0 = LEFT_SHIFT(tmp0, CONST_BITS+1);
+
z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, - FIX_0_765366865);
-
+
tmp10 = tmp0 + tmp2;
tmp12 = tmp0 - tmp2;
-
+
/* Odd part */
-
+
z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
-
+
tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
- + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
- + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
- + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
-
+ + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
+ + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
+ + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
+
tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
- + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
- + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
- + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
+ + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
+ + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
+ + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
/* Final output stage */
-
+
wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITS-PASS1_BITS+1);
wsptr[DCTSIZE*3] = (int) DESCALE(tmp10 - tmp2, CONST_BITS-PASS1_BITS+1);
wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITS-PASS1_BITS+1);
wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 - tmp0, CONST_BITS-PASS1_BITS+1);
}
-
+
/* Pass 2: process 4 rows from work array, store into output array. */
wsptr = workspace;
@@ -200,64 +204,64 @@ jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
#ifndef NO_ZERO_ROW_TEST
if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 &&
- wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
+ wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
/* AC terms all zero */
- JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
- & RANGE_MASK];
-
+ JSAMPLE dcval = range_limit[(int) DESCALE((JLONG) wsptr[0], PASS1_BITS+3)
+ & RANGE_MASK];
+
outptr[0] = dcval;
outptr[1] = dcval;
outptr[2] = dcval;
outptr[3] = dcval;
-
- wsptr += DCTSIZE; /* advance pointer to next row */
+
+ wsptr += DCTSIZE; /* advance pointer to next row */
continue;
}
#endif
-
+
/* Even part */
-
- tmp0 = ((INT32) wsptr[0]) << (CONST_BITS+1);
-
- tmp2 = MULTIPLY((INT32) wsptr[2], FIX_1_847759065)
- + MULTIPLY((INT32) wsptr[6], - FIX_0_765366865);
-
+
+ tmp0 = LEFT_SHIFT((JLONG) wsptr[0], CONST_BITS+1);
+
+ tmp2 = MULTIPLY((JLONG) wsptr[2], FIX_1_847759065)
+ + MULTIPLY((JLONG) wsptr[6], - FIX_0_765366865);
+
tmp10 = tmp0 + tmp2;
tmp12 = tmp0 - tmp2;
-
+
/* Odd part */
-
- z1 = (INT32) wsptr[7];
- z2 = (INT32) wsptr[5];
- z3 = (INT32) wsptr[3];
- z4 = (INT32) wsptr[1];
-
+
+ z1 = (JLONG) wsptr[7];
+ z2 = (JLONG) wsptr[5];
+ z3 = (JLONG) wsptr[3];
+ z4 = (JLONG) wsptr[1];
+
tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
- + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
- + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
- + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
-
+ + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
+ + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
+ + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
+
tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
- + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
- + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
- + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
+ + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
+ + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
+ + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
/* Final output stage */
-
+
outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2,
- CONST_BITS+PASS1_BITS+3+1)
- & RANGE_MASK];
+ CONST_BITS+PASS1_BITS+3+1)
+ & RANGE_MASK];
outptr[3] = range_limit[(int) DESCALE(tmp10 - tmp2,
- CONST_BITS+PASS1_BITS+3+1)
- & RANGE_MASK];
+ CONST_BITS+PASS1_BITS+3+1)
+ & RANGE_MASK];
outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0,
- CONST_BITS+PASS1_BITS+3+1)
- & RANGE_MASK];
+ CONST_BITS+PASS1_BITS+3+1)
+ & RANGE_MASK];
outptr[2] = range_limit[(int) DESCALE(tmp12 - tmp0,
- CONST_BITS+PASS1_BITS+3+1)
- & RANGE_MASK];
-
- wsptr += DCTSIZE; /* advance pointer to next row */
+ CONST_BITS+PASS1_BITS+3+1)
+ & RANGE_MASK];
+
+ wsptr += DCTSIZE; /* advance pointer to next row */
}
}
@@ -268,18 +272,18 @@ jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
*/
GLOBAL(void)
-jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
- JCOEFPTR coef_block,
- JSAMPARRAY output_buf, JDIMENSION output_col)
+jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block,
+ JSAMPARRAY output_buf, JDIMENSION output_col)
{
- INT32 tmp0, tmp10, z1;
+ JLONG tmp0, tmp10, z1;
JCOEFPTR inptr;
- ISLOW_MULT_TYPE * quantptr;
- int * wsptr;
+ ISLOW_MULT_TYPE *quantptr;
+ int *wsptr;
JSAMPROW outptr;
JSAMPLE *range_limit = IDCT_range_limit(cinfo);
int ctr;
- int workspace[DCTSIZE*2]; /* buffers data between passes */
+ int workspace[DCTSIZE*2]; /* buffers data between passes */
SHIFT_TEMPS
/* Pass 1: process columns from input, store into work array. */
@@ -292,21 +296,22 @@ jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
if (ctr == DCTSIZE-2 || ctr == DCTSIZE-4 || ctr == DCTSIZE-6)
continue;
if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*3] == 0 &&
- inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*7] == 0) {
+ inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*7] == 0) {
/* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */
- int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
-
+ int dcval = LEFT_SHIFT(DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]),
+ PASS1_BITS);
+
wsptr[DCTSIZE*0] = dcval;
wsptr[DCTSIZE*1] = dcval;
-
+
continue;
}
-
+
/* Even part */
-
+
z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
- tmp10 = z1 << (CONST_BITS+2);
-
+ tmp10 = LEFT_SHIFT(z1, CONST_BITS+2);
+
/* Odd part */
z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
@@ -319,11 +324,11 @@ jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
/* Final output stage */
-
+
wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITS-PASS1_BITS+2);
wsptr[DCTSIZE*1] = (int) DESCALE(tmp10 - tmp0, CONST_BITS-PASS1_BITS+2);
}
-
+
/* Pass 2: process 2 rows from work array, store into output array. */
wsptr = workspace;
@@ -334,38 +339,38 @@ jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
#ifndef NO_ZERO_ROW_TEST
if (wsptr[1] == 0 && wsptr[3] == 0 && wsptr[5] == 0 && wsptr[7] == 0) {
/* AC terms all zero */
- JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
- & RANGE_MASK];
-
+ JSAMPLE dcval = range_limit[(int) DESCALE((JLONG) wsptr[0], PASS1_BITS+3)
+ & RANGE_MASK];
+
outptr[0] = dcval;
outptr[1] = dcval;
-
- wsptr += DCTSIZE; /* advance pointer to next row */
+
+ wsptr += DCTSIZE; /* advance pointer to next row */
continue;
}
#endif
-
+
/* Even part */
-
- tmp10 = ((INT32) wsptr[0]) << (CONST_BITS+2);
-
+
+ tmp10 = LEFT_SHIFT((JLONG) wsptr[0], CONST_BITS+2);
+
/* Odd part */
- tmp0 = MULTIPLY((INT32) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-c1) */
- + MULTIPLY((INT32) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c7) */
- + MULTIPLY((INT32) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5-c7) */
- + MULTIPLY((INT32) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
+ tmp0 = MULTIPLY((JLONG) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-c1) */
+ + MULTIPLY((JLONG) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c7) */
+ + MULTIPLY((JLONG) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5-c7) */
+ + MULTIPLY((JLONG) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
/* Final output stage */
-
+
outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0,
- CONST_BITS+PASS1_BITS+3+2)
- & RANGE_MASK];
+ CONST_BITS+PASS1_BITS+3+2)
+ & RANGE_MASK];
outptr[1] = range_limit[(int) DESCALE(tmp10 - tmp0,
- CONST_BITS+PASS1_BITS+3+2)
- & RANGE_MASK];
-
- wsptr += DCTSIZE; /* advance pointer to next row */
+ CONST_BITS+PASS1_BITS+3+2)
+ & RANGE_MASK];
+
+ wsptr += DCTSIZE; /* advance pointer to next row */
}
}
@@ -376,12 +381,12 @@ jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
*/
GLOBAL(void)
-jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
- JCOEFPTR coef_block,
- JSAMPARRAY output_buf, JDIMENSION output_col)
+jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block,
+ JSAMPARRAY output_buf, JDIMENSION output_col)
{
int dcval;
- ISLOW_MULT_TYPE * quantptr;
+ ISLOW_MULT_TYPE *quantptr;
JSAMPLE *range_limit = IDCT_range_limit(cinfo);
SHIFT_TEMPS
@@ -390,7 +395,7 @@ jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
*/
quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
- dcval = (int) DESCALE((INT32) dcval, 3);
+ dcval = (int) DESCALE((JLONG) dcval, 3);
output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];
}
« jconfigint.h ('K') | « jidctint.c ('k') | jinclude.h » ('j') | simd/jccolext-sse2-64.asm » ('J')

Powered by Google App Engine
This is Rietveld 408576698