AltiVec SIMD implementation of YCC-to-RGB color conversion

git-svn-id: svn://svn.code.sf.net/p/libjpeg-turbo/code/trunk@1489 632fc199-4ca6-4c93-a231-07263d6284db

8 files changed, 430 insertions, 10 deletions

diff --git a/simd/Makefile.am b/simd/Makefile.am
index cdea264..c4474f5 100644
--- a/simd/Makefile.am
+++ b/simd/Makefile.am
@@ -74,6 +74,7 @@ if SIMD_POWERPC
 
 libsimd_la_SOURCES = jsimd_powerpc.c \
 	jccolor-altivec.c     jcgray-altivec.c      jcsample-altivec.c \
+	jdcolor-altivec.c \
 	jfdctfst-altivec.c    jfdctint-altivec.c \
 	jidctfst-altivec.c    jidctint-altivec.c \
 	jquanti-altivec.c
diff --git a/simd/jccolor-altivec.c b/simd/jccolor-altivec.c
index acb0357..04b8708 100644
--- a/simd/jccolor-altivec.c
+++ b/simd/jccolor-altivec.c
@@ -25,11 +25,6 @@
 #include "jsimd_altivec.h"
 
 
-#ifndef min
-#define min(a,b) ((a) < (b) ? (a) : (b))
-#endif
-
-
 #define F_0_081 5329                 /* FIX(0.08131) */
 #define F_0_114 7471                 /* FIX(0.11400) */
 #define F_0_168 11059                /* FIX(0.16874) */
diff --git a/simd/jcgray-altivec.c b/simd/jcgray-altivec.c
index 7fc0e47..b52fade 100644
--- a/simd/jcgray-altivec.c
+++ b/simd/jcgray-altivec.c
@@ -25,11 +25,6 @@
 #include "jsimd_altivec.h"
 
 
-#ifndef min
-#define min(a,b) ((a) < (b) ? (a) : (b))
-#endif
-
-
 #define F_0_114 7471                 /* FIX(0.11400) */
 #define F_0_250 16384                /* FIX(0.25000) */
 #define F_0_299 19595                /* FIX(0.29900) */
diff --git a/simd/jdcolext-altivec.c b/simd/jdcolext-altivec.c
new file mode 100644
index 0000000..80dc94c
--- /dev/null
+++ b/simd/jdcolext-altivec.c
@@ -0,0 +1,263 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2015, D. R. Commander.
+ * 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
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* This file is included by jdcolor-altivec.c */
+
+
+void jsimd_ycc_rgb_convert_altivec (JDIMENSION out_width, JSAMPIMAGE input_buf,
+                                    JDIMENSION input_row,
+                                    JSAMPARRAY output_buf, int num_rows)
+{
+  JSAMPROW outptr, inptr0, inptr1, inptr2;
+  int pitch = out_width * RGB_PIXELSIZE, offset, num_cols;
+  unsigned char __attribute__((aligned(16))) tmpbuf[RGB_PIXELSIZE * 16];
+
+  __vector unsigned char rgb0, rgb1, rgb2, rgbx0, rgbx1, rgbx2, rgbx3,
+    y, cb, cr, edgel, edgeh, edges, out0, out1, out2, out3;
+#if RGB_PIXELSIZE == 4
+  __vector unsigned char rgb3, out4;
+#endif
+  __vector short rg0, rg1, rg2, rg3, bx0, bx1, bx2, bx3, yl, yh, cbl, cbh,
+    crl, crh, rl, rh, gl, gh, bl, bh, g0s, g1s, g2s, g3s;
+  __vector int g0, g1, g2, g3;
+
+  /* Constants
+   * NOTE: The >> 1 is to compensate for the fact that vec_madds() returns 17
+   * high-order bits, not 16.
+   */
+  __vector short pw_f0402 = { __8X(F_0_402 >> 1) },
+    pw_mf0228 = { __8X(-F_0_228 >> 1) },
+    pw_mf0344_f0285 = { __4X2(-F_0_344, F_0_285) },
+    pw_one = { __8X(1) },
+    pw_255 = { __8X(255) },
+    pw_cj = { __8X(CENTERJSAMPLE) };
+  __vector int pd_onehalf = { __4X(ONE_HALF) };
+  __vector unsigned char zero = { __16X(0) },
+    shift_pack_index =
+      { 0, 1, 4, 5, 8, 9, 12, 13, 16, 17, 20, 21, 24, 25, 28, 29};
+
+  while (--num_rows >= 0) {
+    inptr0 = input_buf[0][input_row];
+    inptr1 = input_buf[1][input_row];
+    inptr2 = input_buf[2][input_row];
+    input_row++;
+    outptr = *output_buf++;
+
+    for (num_cols = pitch; num_cols > 0;
+         num_cols -= RGB_PIXELSIZE * 16, outptr += RGB_PIXELSIZE * 16,
+         inptr0 += 16, inptr1 += 16, inptr2 += 16) {
+
+      y = vec_ld(0, inptr0);
+      /* NOTE: We have to use vec_merge*() here because vec_unpack*() doesn't
+       * support unsigned vectors.
+       */
+      yl = (__vector signed short)vec_mergeh(zero, y);
+      yh = (__vector signed short)vec_mergel(zero, y);
+
+      cb = vec_ld(0, inptr1);
+      cbl = (__vector signed short)vec_mergeh(zero, cb);
+      cbh = (__vector signed short)vec_mergel(zero, cb);
+      cbl = vec_sub(cbl, pw_cj);
+      cbh = vec_sub(cbh, pw_cj);
+
+      cr = vec_ld(0, inptr2);
+      crl = (__vector signed short)vec_mergeh(zero, cr);
+      crh = (__vector signed short)vec_mergel(zero, cr);
+      crl = vec_sub(crl, pw_cj);
+      crh = vec_sub(crh, pw_cj);
+
+      /* (Original)
+       * R = Y                + 1.40200 * Cr
+       * G = Y - 0.34414 * Cb - 0.71414 * Cr
+       * B = Y + 1.77200 * Cb
+       *
+       * (This implementation)
+       * R = Y                + 0.40200 * Cr + Cr
+       * G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+       * B = Y - 0.22800 * Cb + Cb + Cb
+       */
+      bl = vec_add(cbl, cbl);
+      bh = vec_add(cbh, cbh);
+      bl = vec_madds(bl, pw_mf0228, pw_one);
+      bh = vec_madds(bh, pw_mf0228, pw_one);
+      bl = vec_sra(bl, (__vector unsigned short)pw_one);
+      bh = vec_sra(bh, (__vector unsigned short)pw_one);
+      bl = vec_add(bl, cbl);
+      bh = vec_add(bh, cbh);
+      bl = vec_add(bl, cbl);
+      bh = vec_add(bh, cbh);
+      bl = vec_add(bl, yl);
+      bh = vec_add(bh, yh);
+
+      rl = vec_add(crl, crl);
+      rh = vec_add(crh, crh);
+      rl = vec_madds(rl, pw_f0402, pw_one);
+      rh = vec_madds(rh, pw_f0402, pw_one);
+      rl = vec_sra(rl, (__vector unsigned short)pw_one);
+      rh = vec_sra(rh, (__vector unsigned short)pw_one);
+      rl = vec_add(rl, crl);
+      rh = vec_add(rh, crh);
+      rl = vec_add(rl, yl);
+      rh = vec_add(rh, yh);
+
+      g0s = vec_mergeh(cbl, crl);
+      g1s = vec_mergel(cbl, crl);
+      g0 = vec_msums(g0s, pw_mf0344_f0285, pd_onehalf);
+      g1 = vec_msums(g1s, pw_mf0344_f0285, pd_onehalf);
+      g2s = vec_mergeh(cbh, crh);
+      g3s = vec_mergel(cbh, crh);
+      g2 = vec_msums(g2s, pw_mf0344_f0285, pd_onehalf);
+      g3 = vec_msums(g3s, pw_mf0344_f0285, pd_onehalf);
+      /* Clever way to avoid 4 shifts + 2 packs.  This packs the high word from
+       * each dword into a new 16-bit vector, which is the equivalent of
+       * descaling the 32-bit results (right-shifting by 16 bits) and then
+       * packing them.
+       */
+      gl = (__vector short)vec_perm((__vector unsigned short)g0,
+                                    (__vector unsigned short)g1,
+                                    shift_pack_index);
+      gh = (__vector short)vec_perm((__vector unsigned short)g2,
+                                    (__vector unsigned short)g3,
+                                    shift_pack_index);
+      gl = vec_sub(gl, crl);
+      gh = vec_sub(gh, crh);
+      gl = vec_add(gl, yl);
+      gh = vec_add(gh, yh);
+
+      rg0 = vec_mergeh(rl, gl);
+      bx0 = vec_mergeh(bl, pw_255);
+      rg1 = vec_mergel(rl, gl);
+      bx1 = vec_mergel(bl, pw_255);
+      rg2 = vec_mergeh(rh, gh);
+      bx2 = vec_mergeh(bh, pw_255);
+      rg3 = vec_mergel(rh, gh);
+      bx3 = vec_mergel(bh, pw_255);
+
+      rgbx0 = vec_packsu(rg0, bx0);
+      rgbx1 = vec_packsu(rg1, bx1);
+      rgbx2 = vec_packsu(rg2, bx2);
+      rgbx3 = vec_packsu(rg3, bx3);
+
+#if RGB_PIXELSIZE == 3
+      /* rgbx0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 X0 B1 X1 B2 X2 B3 X3
+       * rgbx1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 X4 B5 X5 B6 X6 B7 X7
+       * rgbx2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 X8 B9 X9 Ba Xa Bb Xb
+       * rgbx3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Xc Bd Xd Be Xe Bf Xf
+       *
+       * rgb0 = R0 G0 B0 R1 G1 B1 R2 G2 B2 R3 G3 B3 R4 G4 B4 R5
+       * rgb1 = G5 B5 R6 G6 B6 R7 G7 B7 R8 G8 B8 R9 G9 B9 Ra Ga
+       * rgb2 = Ba Rb Gb Bb Rc Gc Bc Rd Gd Bd Re Ge Be Rf Gf Bf
+       */
+      rgb0 = vec_perm(rgbx0, rgbx1, (__vector unsigned char)RGB_INDEX0);
+      rgb1 = vec_perm(rgbx1, rgbx2, (__vector unsigned char)RGB_INDEX1);
+      rgb2 = vec_perm(rgbx2, rgbx3, (__vector unsigned char)RGB_INDEX2);
+#else
+      /* rgbx0 = R0 G0 R1 G1 R2 G2 R3 G3 B0 X0 B1 X1 B2 X2 B3 X3
+       * rgbx1 = R4 G4 R5 G5 R6 G6 R7 G7 B4 X4 B5 X5 B6 X6 B7 X7
+       * rgbx2 = R8 G8 R9 G9 Ra Ga Rb Gb B8 X8 B9 X9 Ba Xa Bb Xb
+       * rgbx3 = Rc Gc Rd Gd Re Ge Rf Gf Bc Xc Bd Xd Be Xe Bf Xf
+       *
+       * rgb0 = R0 G0 B0 X0 R1 G1 B1 X1 R2 G2 B2 X2 R3 G3 B3 X3
+       * rgb1 = R4 G4 B4 X4 R5 G5 B5 X5 R6 G6 B6 X6 R7 G7 B7 X7
+       * rgb2 = R8 G8 B8 X8 R9 G9 B9 X9 Ra Ga Ba Xa Rb Gb Bb Xb
+       * rgb3 = Rc Gc Bc Xc Rd Gd Bd Xd Re Ge Be Xe Rf Gf Bf Xf
+       */
+      rgb0 = vec_perm(rgbx0, rgbx0, (__vector unsigned char)RGB_INDEX);
+      rgb1 = vec_perm(rgbx1, rgbx1, (__vector unsigned char)RGB_INDEX);
+      rgb2 = vec_perm(rgbx2, rgbx2, (__vector unsigned char)RGB_INDEX);
+      rgb3 = vec_perm(rgbx3, rgbx3, (__vector unsigned char)RGB_INDEX);
+#endif
+
+      offset = (size_t)outptr & 15;
+      if (offset) {
+        __vector unsigned char unaligned_shift_index;
+        int bytes = num_cols + offset;
+
+        if (bytes < (RGB_PIXELSIZE + 1) * 16 && (bytes & 15)) {
+          /* Slow path to prevent buffer overwrite.  Since there is no way to
+           * write a partial AltiVec register, overwrite would occur on the
+           * last chunk of the last image row if the right edge is not on a
+           * 16-byte boundary.  It could also occur on other rows if the bytes
+           * per row is low enough.  Since we can't determine whether we're on
+           * the last image row, we have to assume every row is the last.
+           */
+          vec_st(rgb0, 0, tmpbuf);
+          vec_st(rgb1, 16, tmpbuf);
+          vec_st(rgb2, 32, tmpbuf);
+#if RGB_PIXELSIZE == 4
+          vec_st(rgb3, 48, tmpbuf);
+#endif
+          memcpy(outptr, tmpbuf, min(num_cols, RGB_PIXELSIZE * 16));
+        } else {
+          /* Fast path */
+          unaligned_shift_index = vec_lvsl(0, outptr);
+          edgel = vec_ld(0, outptr);
+          edgeh = vec_ld(min(num_cols, RGB_PIXELSIZE * 16) + offset, outptr);
+          edges = vec_perm(edgeh, edgel, unaligned_shift_index);
+          unaligned_shift_index = vec_lvsr(0, outptr);
+          out0 = vec_perm(edges, rgb0, unaligned_shift_index);
+          out1 = vec_perm(rgb0, rgb1, unaligned_shift_index);
+          out2 = vec_perm(rgb1, rgb2, unaligned_shift_index);
+#if RGB_PIXELSIZE == 4
+          out3 = vec_perm(rgb2, rgb3, unaligned_shift_index);
+          out4 = vec_perm(rgb3, edges, unaligned_shift_index);
+#else
+          out3 = vec_perm(rgb2, edges, unaligned_shift_index);
+#endif
+          vec_st(out0, 0, outptr);
+          if (bytes > 16)
+            vec_st(out1, 16, outptr);
+          if (bytes > 32)
+            vec_st(out2, 32, outptr);
+          if (bytes > 48)
+            vec_st(out3, 48, outptr);
+#if RGB_PIXELSIZE == 4
+          if (bytes > 64)
+            vec_st(out4, 64, outptr);
+#endif
+        }
+      } else {
+        if (num_cols < RGB_PIXELSIZE * 16 && (num_cols & 15)) {
+          /* Slow path */
+          vec_st(rgb0, 0, tmpbuf);
+          vec_st(rgb1, 16, tmpbuf);
+          vec_st(rgb2, 32, tmpbuf);
+#if RGB_PIXELSIZE == 4
+          vec_st(rgb3, 48, tmpbuf);
+#endif
+          memcpy(outptr, tmpbuf, min(num_cols, RGB_PIXELSIZE * 16));
+        } else {
+          /* Fast path */
+          vec_st(rgb0, 0, outptr);
+          if (num_cols > 16)
+            vec_st(rgb1, 16, outptr);
+          if (num_cols > 32)
+            vec_st(rgb2, 32, outptr);
+#if RGB_PIXELSIZE == 4
+          if (num_cols > 48)
+            vec_st(rgb3, 48, outptr);
+#endif
+        }
+      }
+    }
+  }
+}
diff --git a/simd/jdcolor-altivec.c b/simd/jdcolor-altivec.c
new file mode 100644
index 0000000..e0892d8
--- /dev/null
+++ b/simd/jdcolor-altivec.c
@@ -0,0 +1,96 @@
+/*
+ * AltiVec optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2015, D. R. Commander.
+ * 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
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* YCC --> RGB CONVERSION */
+
+#include "jsimd_altivec.h"
+
+
+#define F_0_344 22554              /* FIX(0.34414) */
+#define F_0_714 46802              /* FIX(0.71414) */
+#define F_1_402 91881              /* FIX(1.40200) */
+#define F_1_772 116130             /* FIX(1.77200) */
+#define F_0_402 (F_1_402 - 65536)  /* FIX(1.40200) - FIX(1) */
+#define F_0_285 (65536 - F_0_714)  /* FIX(1) - FIX(0.71414) */
+#define F_0_228 (131072 - F_1_772) /* FIX(2) - FIX(1.77200) */
+
+#define SCALEBITS 16
+#define ONE_HALF (1 << (SCALEBITS - 1))
+
+#define RGB_INDEX0 {0,1,8,2,3,10,4,5,12,6,7,14,16,17,24,18}
+#define RGB_INDEX1 {3,10,4,5,12,6,7,14,16,17,24,18,19,26,20,21}
+#define RGB_INDEX2 {12,6,7,14,16,17,24,18,19,26,20,21,28,22,23,30}
+#include "jdcolext-altivec.c"
+#undef RGB_PIXELSIZE
+
+#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+#define jsimd_ycc_rgb_convert_altivec jsimd_ycc_extrgb_convert_altivec
+#include "jdcolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGB_INDEX0
+#undef RGB_INDEX1
+#undef RGB_INDEX2
+#undef jsimd_ycc_rgb_convert_altivec
+
+#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+#define RGB_INDEX {0,1,8,9,2,3,10,11,4,5,12,13,6,7,14,15}
+#define jsimd_ycc_rgb_convert_altivec jsimd_ycc_extrgbx_convert_altivec
+#include "jdcolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGB_INDEX
+#undef jsimd_ycc_rgb_convert_altivec
+
+#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+#define RGB_INDEX0 {8,1,0,10,3,2,12,5,4,14,7,6,24,17,16,26}
+#define RGB_INDEX1 {3,2,12,5,4,14,7,6,24,17,16,26,19,18,28,21}
+#define RGB_INDEX2 {4,14,7,6,24,17,16,26,19,18,28,21,20,30,23,22}
+#define jsimd_ycc_rgb_convert_altivec jsimd_ycc_extbgr_convert_altivec
+#include "jdcolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGB_INDEX0
+#undef RGB_INDEX1
+#undef RGB_INDEX2
+#undef jsimd_ycc_rgb_convert_altivec
+
+#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+#define RGB_INDEX {8,1,0,9,10,3,2,11,12,5,4,13,14,7,6,15}
+#define jsimd_ycc_rgb_convert_altivec jsimd_ycc_extbgrx_convert_altivec
+#include "jdcolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGB_INDEX
+#undef jsimd_ycc_rgb_convert_altivec
+
+#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+#define RGB_INDEX {9,8,1,0,11,10,3,2,13,12,5,4,15,14,7,6}
+#define jsimd_ycc_rgb_convert_altivec jsimd_ycc_extxbgr_convert_altivec
+#include "jdcolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGB_INDEX
+#undef jsimd_ycc_rgb_convert_altivec
+
+#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+#define RGB_INDEX {9,0,1,8,11,2,3,10,13,4,5,12,15,6,7,14}
+#define jsimd_ycc_rgb_convert_altivec jsimd_ycc_extxrgb_convert_altivec
+#include "jdcolext-altivec.c"
+#undef RGB_PIXELSIZE
+#undef RGB_INDEX
+#undef jsimd_ycc_rgb_convert_altivec
diff --git a/simd/jsimd.h b/simd/jsimd.h
index 905032e..402f44b 100644
--- a/simd/jsimd.h
+++ b/simd/jsimd.h
@@ -321,6 +321,28 @@ EXTERN(void) jsimd_ycc_extxrgb_convert_mips_dspr2
         (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
          JSAMPARRAY output_buf, int num_rows);
 
+EXTERN(void) jsimd_ycc_rgb_convert_altivec
+        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+         JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extrgb_convert_altivec
+        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+         JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extrgbx_convert_altivec
+        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+         JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extbgr_convert_altivec
+        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+         JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extbgrx_convert_altivec
+        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+         JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extxbgr_convert_altivec
+        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+         JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extxrgb_convert_altivec
+        (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+         JSAMPARRAY output_buf, int num_rows);
+
 /* NULL Colorspace Conversion */
 EXTERN(void) jsimd_c_null_convert_mips_dspr2
         (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
diff --git a/simd/jsimd_altivec.h b/simd/jsimd_altivec.h
index 7870556..e800206 100644
--- a/simd/jsimd_altivec.h
+++ b/simd/jsimd_altivec.h
@@ -74,3 +74,7 @@
   col##6 = vec_mergeh(col67e, col67o); /* col6=(06 16 26 36 46 56 66 76) */   \
   col##7 = vec_mergel(col67e, col67o); /* col7=(07 17 27 37 47 57 67 77) */   \
 }
+
+#ifndef min
+#define min(a,b) ((a) < (b) ? (a) : (b))
+#endif
diff --git a/simd/jsimd_powerpc.c b/simd/jsimd_powerpc.c
index a6bcb69..431510a 100644
--- a/simd/jsimd_powerpc.c
+++ b/simd/jsimd_powerpc.c
@@ -80,6 +80,19 @@ jsimd_can_rgb_gray (void)
 GLOBAL(int)
 jsimd_can_ycc_rgb (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 ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
@@ -170,6 +183,37 @@ jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
                        JSAMPIMAGE input_buf, JDIMENSION input_row,
                        JSAMPARRAY output_buf, int num_rows)
 {
+  void (*altivecfct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
+
+  switch(cinfo->out_color_space) {
+    case JCS_EXT_RGB:
+      altivecfct=jsimd_ycc_extrgb_convert_altivec;
+      break;
+    case JCS_EXT_RGBX:
+    case JCS_EXT_RGBA:
+      altivecfct=jsimd_ycc_extrgbx_convert_altivec;
+      break;
+    case JCS_EXT_BGR:
+      altivecfct=jsimd_ycc_extbgr_convert_altivec;
+      break;
+    case JCS_EXT_BGRX:
+    case JCS_EXT_BGRA:
+      altivecfct=jsimd_ycc_extbgrx_convert_altivec;
+      break;
+    case JCS_EXT_XBGR:
+    case JCS_EXT_ABGR:
+      altivecfct=jsimd_ycc_extxbgr_convert_altivec;
+      break;
+    case JCS_EXT_XRGB:
+    case JCS_EXT_ARGB:
+      altivecfct=jsimd_ycc_extxrgb_convert_altivec;
+      break;
+    default:
+      altivecfct=jsimd_ycc_rgb_convert_altivec;
+      break;
+  }
+
+  altivecfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
 }
 
 GLOBAL(void)