ARM NEON SIMD support for YCC-to-RGB565 conversion, and optimizations to the existing YCC-to-RGB conversion code:

-----

aee36252be.patch

From aee36252be20054afce371a92406fc66ba6627b5 Mon Sep 17 00:00:00 2001
From: Siarhei Siamashka <siarhei.siamashka@gmail.com>
Date: Wed, 13 Aug 2014 03:50:22 +0300
Subject: [PATCH] ARM: Faster NEON yuv->rgb conversion for Krait and Cortex-A15

The older code was developed and tested only on ARM Cortex-A8 and ARM Cortex-A9.
Tuning it for newer ARM processors can introduce some speed-up (up to 20%).

The performance of the inner loop (conversion of 8 pixels) improves from
~27 cycles down to ~22 cycles on Qualcomm Krait 300, and from ~20 cycles
down to ~18 cycles on ARM Cortex-A15.

The performance remains exactly the same on ARM Cortex-A7 (~58 cycles),
ARM Cortex-A8 (~25 cycles) and ARM Cortex-A9 (~30 cycles) processors.

Also use larger indentation in the source code for separating two independent
instruction streams.

-----

a5efdbf22c.patch

From a5efdbf22ce9c1acd4b14a353cec863c2c57557e Mon Sep 17 00:00:00 2001
From: Siarhei Siamashka <siarhei.siamashka@gmail.com>
Date: Wed, 13 Aug 2014 07:23:09 +0300
Subject: [PATCH] ARM: NEON optimized yuv->rgb565 conversion

The performance of the inner loop (conversion of 8 pixels):
* ARM Cortex-A7:  ~55 cycles
* ARM Cortex-A8:  ~28 cycles
* ARM Cortex-A9:  ~32 cycles
* ARM Cortex-A15: ~20 cycles
* Qualcomm Krait: ~24 cycles

Based on the Linaro rgb565 patch from
    https://sourceforge.net/p/libjpeg-turbo/patches/24/
but implements better instructions scheduling.


git-svn-id: svn+ssh://svn.code.sf.net/p/libjpeg-turbo/code/trunk@1385 632fc199-4ca6-4c93-a231-07263d6284db

ChangeLog.txt

@@ -68,7 +68,8 @@ accuracy than the slow integer DCT/IDCT algorithms, and they are quite a bit
 slower.
 
 [8] Added a new output colorspace (JCS_RGB565) to the libjpeg API that allows
-for decompressing JPEG images into RGB565 (16-bit) pixels.
+for decompressing JPEG images into RGB565 (16-bit) pixels.  If dithering is not
+used, then this code path is SIMD-accelerated on ARM platforms.
 
 [9] Numerous obsolete features, such as support for non-ANSI compilers and
 support for the MS-DOS memory model, were removed from the libjpeg code,

jdcolor.c

@@ -6,7 +6,7 @@
  * Modified 2011 by Guido Vollbeding.
  * libjpeg-turbo Modifications:
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright (C) 2009, 2011-2012, D. R. Commander.
+ * Copyright (C) 2009, 2011-2012, 2014, D. R. Commander.
  * Copyright (C) 2013, Linaro Limited.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -657,8 +657,12 @@ jinit_color_deconverter (j_decompress_ptr cinfo)
     cinfo->out_color_components = 3;
     if (cinfo->dither_mode == JDITHER_NONE) {
       if (cinfo->jpeg_color_space == JCS_YCbCr) {
-        cconvert->pub.color_convert = ycc_rgb565_convert;
+         if (jsimd_can_ycc_rgb565())
-        build_ycc_rgb_table(cinfo);
+           cconvert->pub.color_convert = jsimd_ycc_rgb565_convert;
+         else {
+           cconvert->pub.color_convert = ycc_rgb565_convert;
+           build_ycc_rgb_table(cinfo);
+        }
       } else if (cinfo->jpeg_color_space == JCS_GRAYSCALE) {
         cconvert->pub.color_convert = gray_rgb565_convert;
       } else if (cinfo->jpeg_color_space == JCS_RGB) {

jsimd.h

@@ -2,7 +2,7 @@
  * jsimd.h
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright 2011 D. R. Commander
+ * Copyright 2011, 2014 D. R. Commander
  *
  * Based on the x86 SIMD extension for IJG JPEG library,
  * Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -13,6 +13,7 @@
 EXTERN(int) jsimd_can_rgb_ycc (void);
 EXTERN(int) jsimd_can_rgb_gray (void);
 EXTERN(int) jsimd_can_ycc_rgb (void);
+EXTERN(int) jsimd_can_ycc_rgb565 (void);
 EXTERN(int) jsimd_c_can_null_convert (void);
 
 EXTERN(void) jsimd_rgb_ycc_convert
@@ -24,6 +25,9 @@ EXTERN(void) jsimd_rgb_gray_convert
 EXTERN(void) jsimd_ycc_rgb_convert
         (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION input_row,
          JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_rgb565_convert
+        (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION input_row,
+         JSAMPARRAY output_buf, int num_rows);
 EXTERN(void) jsimd_c_null_convert
         (j_compress_ptr cinfo, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
          JDIMENSION output_row, int num_rows);

jsimd_none.c

@@ -2,7 +2,7 @@
  * jsimd_none.c
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright 2009-2011 D. R. Commander
+ * Copyright 2009-2011, 2014 D. R. Commander
  *
  * Based on the x86 SIMD extension for IJG JPEG library,
  * Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -36,6 +36,12 @@ jsimd_can_ycc_rgb (void)
   return 0;
 }
 
+GLOBAL(int)
+jsimd_can_ycc_rgb565 (void)
+{
+  return 0;
+}
+
 GLOBAL(int)
 jsimd_c_can_null_convert (void)
 {
@@ -63,6 +69,13 @@ jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
 {
 }
 
+GLOBAL(void)
+jsimd_ycc_rgb565_convert (j_decompress_ptr cinfo,
+                          JSAMPIMAGE input_buf, JDIMENSION input_row,
+                          JSAMPARRAY output_buf, int num_rows)
+{
+}
+
 GLOBAL(void)
 jsimd_c_null_convert (j_compress_ptr cinfo,
                       JSAMPARRAY input_buf, JSAMPIMAGE output_buf,

simd/jsimd.h

@@ -4,6 +4,7 @@
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * Copyright 2011 D. R. Commander
  * Copyright (C) 2013-2014, MIPS Technologies, Inc., California
+ * Copyright (C) 2014 Linaro Limited
  *
  * Based on the x86 SIMD extension for IJG JPEG library,
  * Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -249,6 +250,9 @@ EXTERN(void) jsimd_ycc_extxbgr_convert_neon
 EXTERN(void) jsimd_ycc_extxrgb_convert_neon
         (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
          JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_rgb565_convert_neon
+        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+         JSAMPARRAY output_buf, int num_rows);
 
 EXTERN(void) jsimd_ycc_rgb_convert_mips_dspr2
         (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,

simd/jsimd_arm.c

@@ -2,7 +2,7 @@
  * jsimd_arm.c
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright 2009-2011, 2013 D. R. Commander
+ * Copyright 2009-2011, 2013-2014 D. R. Commander
  *
  * Based on the x86 SIMD extension for IJG JPEG library,
  * Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -175,6 +175,23 @@ jsimd_can_ycc_rgb (void)
   return 0;
 }
 
+GLOBAL(int)
+jsimd_can_ycc_rgb565 (void)
+{
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (simd_support & JSIMD_ARM_NEON)
+    return 1;
+
+  return 0;
+}
+
 GLOBAL(void)
 jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
                        JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
@@ -251,7 +268,7 @@ jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
     case JCS_EXT_ARGB:
       neonfct=jsimd_ycc_extxrgb_convert_neon;
       break;
-  default:
+    default:
       neonfct=jsimd_ycc_extrgb_convert_neon;
       break;
   }
@@ -260,6 +277,16 @@ jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
     neonfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
 }
 
+GLOBAL(void)
+jsimd_ycc_rgb565_convert (j_decompress_ptr cinfo,
+                          JSAMPIMAGE input_buf, JDIMENSION input_row,
+                          JSAMPARRAY output_buf, int num_rows)
+{
+  if (simd_support & JSIMD_ARM_NEON)
+    jsimd_ycc_rgb565_convert_neon(cinfo->output_width, input_buf, input_row,
+                                  output_buf, num_rows);
+}
+
 GLOBAL(int)
 jsimd_can_h2v2_downsample (void)
 {

simd/jsimd_arm64.c

@@ -95,6 +95,12 @@ jsimd_can_ycc_rgb (void)
   return 0;
 }
 
+GLOBAL(int)
+jsimd_can_ycc_rgb565 (void)
+{
+  return 0;
+}
+
 GLOBAL(void)
 jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
                        JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
@@ -148,6 +154,13 @@ jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
     neonfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
 }
 
+GLOBAL(void)
+jsimd_ycc_rgb565_convert (j_decompress_ptr cinfo,
+                          JSAMPIMAGE input_buf, JDIMENSION input_row,
+                          JSAMPARRAY output_buf, int num_rows)
+{
+}
+
 GLOBAL(int)
 jsimd_can_h2v2_downsample (void)
 {

simd/jsimd_arm_neon.S

@@ -4,6 +4,7 @@
  * Copyright (C) 2009-2011 Nokia Corporation and/or its subsidiary(-ies).
  * All rights reserved.
  * Author: Siarhei Siamashka <siarhei.siamashka@nokia.com>
+ * Copyright (C) 2014 Linaro Limited.  All Rights Reserved.
  *
  * This software is provided 'as-is', without any express or implied
  * warranty.  In no event will the authors be held liable for any damages
@@ -1346,6 +1347,19 @@ asm_function jsimd_idct_2x2_neon
         .else
             .error unsupported macroblock size
         .endif
+    .elseif \bpp == 16
+        .if \size == 8
+            vst1.16  {q15}, [RGB]!
+        .elseif \size == 4
+            vst1.16  {d30}, [RGB]!
+        .elseif \size == 2
+            vst1.16  {d31[0]}, [RGB]!
+            vst1.16  {d31[1]}, [RGB]!
+        .elseif \size == 1
+            vst1.16  {d31[2]}, [RGB]!
+        .else
+            .error unsupported macroblock size
+        .endif
     .else
         .error unsupported bpp
     .endif
@@ -1377,44 +1391,71 @@ asm_function jsimd_idct_2x2_neon
     vrshrn.s32      d25, q13, #14
     vrshrn.s32      d28, q14, #14
     vrshrn.s32      d29, q15, #14
-    vaddw.u8        q10, q10, d0
+    vaddw.u8        q11, q10, d0
     vaddw.u8        q12, q12, d0
     vaddw.u8        q14, q14, d0
-    vqmovun.s16     d1\g_offs, q10
+.if \bpp != 16
+    vqmovun.s16     d1\g_offs, q11
     vqmovun.s16     d1\r_offs, q12
     vqmovun.s16     d1\b_offs, q14
+.else /* rgb565 */
+    vqshlu.s16      q13, q11, #8
+    vqshlu.s16      q15, q12, #8
+    vqshlu.s16      q14, q14, #8
+    vsri.u16        q15, q13, #5
+    vsri.u16        q15, q14, #11
+.endif
 .endm
 
 .macro do_yuv_to_rgb_stage2_store_load_stage1
-    vld1.8          {d4}, [U, :64]!
+      /* "do_yuv_to_rgb_stage2" and "store" */
       vrshrn.s32      d20, q10, #15
+    /* "load" and "do_yuv_to_rgb_stage1" */
+    pld             [U, #64]
       vrshrn.s32      d21, q11, #15
+    pld             [V, #64]
       vrshrn.s32      d24, q12, #14
       vrshrn.s32      d25, q13, #14
+    vld1.8          {d4}, [U, :64]!
       vrshrn.s32      d28, q14, #14
     vld1.8          {d5}, [V, :64]!
       vrshrn.s32      d29, q15, #14
-      vaddw.u8        q10, q10, d0
-      vaddw.u8        q12, q12, d0
-      vaddw.u8        q14, q14, d0
-      vqmovun.s16     d1\g_offs, q10
-    vld1.8          {d0}, [Y, :64]!
-      vqmovun.s16     d1\r_offs, q12
-    pld             [U, #64]
-    pld             [V, #64]
-    pld             [Y, #64]
-      vqmovun.s16     d1\b_offs, q14
     vaddw.u8        q3, q1, d4     /* q3 = u - 128 */
     vaddw.u8        q4, q1, d5     /* q2 = v - 128 */
-      do_store        \bpp, 8
+      vaddw.u8        q11, q10, d0
     vmull.s16       q10, d6, d1[1] /* multiply by -11277 */
     vmlal.s16       q10, d8, d1[2] /* multiply by -23401 */
+      vaddw.u8        q12, q12, d0
+      vaddw.u8        q14, q14, d0
+.if \bpp != 16 /**************** rgb24/rgb32 *********************************/
+      vqmovun.s16     d1\g_offs, q11
+    pld             [Y, #64]
+      vqmovun.s16     d1\r_offs, q12
+    vld1.8          {d0}, [Y, :64]!
+      vqmovun.s16     d1\b_offs, q14
     vmull.s16       q11, d7, d1[1] /* multiply by -11277 */
     vmlal.s16       q11, d9, d1[2] /* multiply by -23401 */
+      do_store        \bpp, 8
     vmull.s16       q12, d8, d1[0] /* multiply by 22971 */
     vmull.s16       q13, d9, d1[0] /* multiply by 22971 */
     vmull.s16       q14, d6, d1[3] /* multiply by 29033 */
     vmull.s16       q15, d7, d1[3] /* multiply by 29033 */
+.else /**************************** rgb565 ***********************************/
+      vqshlu.s16      q13, q11, #8
+    pld             [Y, #64]
+      vqshlu.s16      q15, q12, #8
+      vqshlu.s16      q14, q14, #8
+    vld1.8          {d0}, [Y, :64]!
+    vmull.s16       q11, d7, d1[1]
+    vmlal.s16       q11, d9, d1[2]
+      vsri.u16        q15, q13, #5
+    vmull.s16       q12, d8, d1[0]
+      vsri.u16        q15, q14, #11
+    vmull.s16       q13, d9, d1[0]
+    vmull.s16       q14, d6, d1[3]
+      do_store        \bpp, 8
+    vmull.s16       q15, d7, d1[3]
+.endif
 .endm
 
 .macro do_yuv_to_rgb
@@ -1556,6 +1597,7 @@ generate_jsimd_ycc_rgb_convert_neon extrgbx, 32, 0, 1, 2
 generate_jsimd_ycc_rgb_convert_neon extbgrx, 32, 2, 1, 0
 generate_jsimd_ycc_rgb_convert_neon extxbgr, 32, 3, 2, 1
 generate_jsimd_ycc_rgb_convert_neon extxrgb, 32, 1, 2, 3
+generate_jsimd_ycc_rgb_convert_neon rgb565,  16, 0, 0, 0
 
 .purgem do_load
 .purgem do_store

simd/jsimd_i386.c

@@ -2,7 +2,7 @@
  * jsimd_i386.c
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright 2009-2011, 2013 D. R. Commander
+ * Copyright 2009-2011, 2013-2014 D. R. Commander
  *
  * Based on the x86 SIMD extension for IJG JPEG library,
  * Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -130,6 +130,12 @@ jsimd_can_ycc_rgb (void)
   return 0;
 }
 
+GLOBAL(int)
+jsimd_can_ycc_rgb565 (void)
+{
+  return 0;
+}
+
 GLOBAL(void)
 jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
                        JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
@@ -280,6 +286,13 @@ jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
     mmxfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
 }
 
+GLOBAL(void)
+jsimd_ycc_rgb565_convert (j_decompress_ptr cinfo,
+                          JSAMPIMAGE input_buf, JDIMENSION input_row,
+                          JSAMPARRAY output_buf, int num_rows)
+{
+}
+
 GLOBAL(int)
 jsimd_can_h2v2_downsample (void)
 {

simd/jsimd_mips.c

@@ -2,7 +2,7 @@
  * jsimd_mips.c
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright 2009-2011 D. R. Commander
+ * Copyright 2009-2011, 2014 D. R. Commander
  * Copyright (C) 2013-2014, MIPS Technologies, Inc., California
  *
  * Based on the x86 SIMD extension for IJG JPEG library,
@@ -161,6 +161,12 @@ jsimd_can_ycc_rgb (void)
   return 0;
 }
 
+GLOBAL(int)
+jsimd_can_ycc_rgb565 (void)
+{
+  return 0;
+}
+
 GLOBAL(int)
 jsimd_c_can_null_convert (void)
 {
@@ -299,6 +305,13 @@ jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
                  num_rows);
 }
 
+GLOBAL(void)
+jsimd_ycc_rgb565_convert (j_decompress_ptr cinfo,
+                          JSAMPIMAGE input_buf, JDIMENSION input_row,
+                          JSAMPARRAY output_buf, int num_rows)
+{
+}
+
 GLOBAL(void)
 jsimd_c_null_convert (j_compress_ptr cinfo,
                       JSAMPARRAY input_buf, JSAMPIMAGE output_buf,

simd/jsimd_x86_64.c

@@ -2,7 +2,7 @@
  * jsimd_x86_64.c
  *
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
- * Copyright 2009-2011 D. R. Commander
+ * Copyright 2009-2011, 2014 D. R. Commander
  *
  * Based on the x86 SIMD extension for IJG JPEG library,
  * Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -80,6 +80,12 @@ jsimd_can_ycc_rgb (void)
   return 1;
 }
 
+GLOBAL(int)
+jsimd_can_ycc_rgb565 (void)
+{
+  return 0;
+}
+
 GLOBAL(void)
 jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
                        JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
@@ -194,6 +200,13 @@ jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
   sse2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
 }
 
+GLOBAL(void)
+jsimd_ycc_rgb565_convert (j_decompress_ptr cinfo,
+                          JSAMPIMAGE input_buf, JDIMENSION input_row,
+                          JSAMPARRAY output_buf, int num_rows)
+{
+}
+
 GLOBAL(int)
 jsimd_can_h2v2_downsample (void)
 {