summaryrefslogtreecommitdiff
path: root/src/libffmpeg/libavcodec/armv4l
diff options
context:
space:
mode:
authorMiguel Freitas <miguelfreitas@users.sourceforge.net>2007-01-13 21:19:52 +0000
committerMiguel Freitas <miguelfreitas@users.sourceforge.net>2007-01-13 21:19:52 +0000
commit6e8ff6e5c232de4b8235626af31ab85345120a93 (patch)
tree25930156aa9f4f2014bf6fe3d65c183262626b8d /src/libffmpeg/libavcodec/armv4l
parent2f5905081ee2040537f043fe4afabbb66d26354e (diff)
downloadxine-lib-6e8ff6e5c232de4b8235626af31ab85345120a93.tar.gz
xine-lib-6e8ff6e5c232de4b8235626af31ab85345120a93.tar.bz2
* ffmpeg update to 51.28.0
* Workaround ffmpeg buggy codecs that don't release their DR1 frames. * Fix several segfaults and freezing problem with H264 streams that use a lot of reference frames (eg. 15) * Initial support to enable/disable ffmpeg codecs. Codecs may be disabled in groups by --disable-ffmpeg-uncommon-codecs/--disable-ffmpeg-popular-codecs Think of "uncommon" codecs what people would never want to play with their PDAs (they will save memory by removing them). Note: currently both uncommon/popular codecs are _build_ but disabled. that is, build system still need some improvements to really save memory. warning: non-autoconf guru playing with the build system, likely breakage. CVS patchset: 8499 CVS date: 2007/01/13 21:19:52
Diffstat (limited to 'src/libffmpeg/libavcodec/armv4l')
-rw-r--r--src/libffmpeg/libavcodec/armv4l/Makefile.am7
-rw-r--r--src/libffmpeg/libavcodec/armv4l/dsputil_arm_s.S696
-rw-r--r--src/libffmpeg/libavcodec/armv4l/dsputil_iwmmxt.c188
-rw-r--r--src/libffmpeg/libavcodec/armv4l/dsputil_iwmmxt_rnd.h1114
-rw-r--r--src/libffmpeg/libavcodec/armv4l/mathops.h49
-rw-r--r--src/libffmpeg/libavcodec/armv4l/mpegvideo_armv5te.c213
-rw-r--r--src/libffmpeg/libavcodec/armv4l/mpegvideo_iwmmxt.c119
-rw-r--r--src/libffmpeg/libavcodec/armv4l/simple_idct_armv5te.S718
8 files changed, 3103 insertions, 1 deletions
diff --git a/src/libffmpeg/libavcodec/armv4l/Makefile.am b/src/libffmpeg/libavcodec/armv4l/Makefile.am
index 0f3d230f6..33e0882c9 100644
--- a/src/libffmpeg/libavcodec/armv4l/Makefile.am
+++ b/src/libffmpeg/libavcodec/armv4l/Makefile.am
@@ -6,7 +6,12 @@ ASFLAGS =
noinst_LTLIBRARIES = libavcodec_armv4l.la
-libavcodec_armv4l_src = dsputil_arm.c jrevdct_arm.S mpegvideo_arm.c simple_idct_arm.S
+libavcodec_armv4l_src = dsputil_arm.c jrevdct_arm.S mpegvideo_arm.c simple_idct_arm.S \
+ dsputil_arm_s.S dsputil_iwmmxt.c dsputil_iwmmxt_rnd.h \
+ mpegvideo_armv5te.c mpegvideo_iwmmxt.c simple_idct_armv5te.S
+
+noinst_HEADERS = mathops.h
+
libavcodec_armv4l_dummy = libavcodec_armv4l_dummy.c
EXTRA_DIST = $(libavcodec_armv4l_src) $(libavcodec_armv4l_dummy)
diff --git a/src/libffmpeg/libavcodec/armv4l/dsputil_arm_s.S b/src/libffmpeg/libavcodec/armv4l/dsputil_arm_s.S
new file mode 100644
index 000000000..2a3ee9c50
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/dsputil_arm_s.S
@@ -0,0 +1,696 @@
+@
+@ ARMv4L optimized DSP utils
+@ Copyright (c) 2004 AGAWA Koji <i (AT) atty (DOT) jp>
+@
+@ This file is part of FFmpeg.
+@
+@ FFmpeg is free software; you can redistribute it and/or
+@ modify it under the terms of the GNU Lesser General Public
+@ License as published by the Free Software Foundation; either
+@ version 2.1 of the License, or (at your option) any later version.
+@
+@ FFmpeg is distributed in the hope that it will be useful,
+@ but WITHOUT ANY WARRANTY; without even the implied warranty of
+@ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+@ Lesser General Public License for more details.
+@
+@ You should have received a copy of the GNU Lesser General Public
+@ License along with FFmpeg; if not, write to the Free Software
+@ Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+@
+
+.macro ADJ_ALIGN_QUADWORD_D shift, Rd0, Rd1, Rd2, Rd3, Rn0, Rn1, Rn2, Rn3, Rn4
+ mov \Rd0, \Rn0, lsr #(\shift * 8)
+ mov \Rd1, \Rn1, lsr #(\shift * 8)
+ mov \Rd2, \Rn2, lsr #(\shift * 8)
+ mov \Rd3, \Rn3, lsr #(\shift * 8)
+ orr \Rd0, \Rd0, \Rn1, lsl #(32 - \shift * 8)
+ orr \Rd1, \Rd1, \Rn2, lsl #(32 - \shift * 8)
+ orr \Rd2, \Rd2, \Rn3, lsl #(32 - \shift * 8)
+ orr \Rd3, \Rd3, \Rn4, lsl #(32 - \shift * 8)
+.endm
+.macro ADJ_ALIGN_DOUBLEWORD shift, R0, R1, R2
+ mov \R0, \R0, lsr #(\shift * 8)
+ orr \R0, \R0, \R1, lsl #(32 - \shift * 8)
+ mov \R1, \R1, lsr #(\shift * 8)
+ orr \R1, \R1, \R2, lsl #(32 - \shift * 8)
+.endm
+.macro ADJ_ALIGN_DOUBLEWORD_D shift, Rdst0, Rdst1, Rsrc0, Rsrc1, Rsrc2
+ mov \Rdst0, \Rsrc0, lsr #(\shift * 8)
+ mov \Rdst1, \Rsrc1, lsr #(\shift * 8)
+ orr \Rdst0, \Rdst0, \Rsrc1, lsl #(32 - (\shift * 8))
+ orr \Rdst1, \Rdst1, \Rsrc2, lsl #(32 - (\shift * 8))
+.endm
+
+.macro RND_AVG32 Rd0, Rd1, Rn0, Rn1, Rm0, Rm1, Rmask
+ @ Rd = (Rn | Rm) - (((Rn ^ Rm) & ~0x01010101) >> 1)
+ @ Rmask = 0xFEFEFEFE
+ @ Rn = destroy
+ eor \Rd0, \Rn0, \Rm0
+ eor \Rd1, \Rn1, \Rm1
+ orr \Rn0, \Rn0, \Rm0
+ orr \Rn1, \Rn1, \Rm1
+ and \Rd0, \Rd0, \Rmask
+ and \Rd1, \Rd1, \Rmask
+ sub \Rd0, \Rn0, \Rd0, lsr #1
+ sub \Rd1, \Rn1, \Rd1, lsr #1
+.endm
+
+.macro NO_RND_AVG32 Rd0, Rd1, Rn0, Rn1, Rm0, Rm1, Rmask
+ @ Rd = (Rn & Rm) - (((Rn ^ Rm) & ~0x01010101) >> 1)
+ @ Rmask = 0xFEFEFEFE
+ @ Rn = destroy
+ eor \Rd0, \Rn0, \Rm0
+ eor \Rd1, \Rn1, \Rm1
+ and \Rn0, \Rn0, \Rm0
+ and \Rn1, \Rn1, \Rm1
+ and \Rd0, \Rd0, \Rmask
+ and \Rd1, \Rd1, \Rmask
+ add \Rd0, \Rn0, \Rd0, lsr #1
+ add \Rd1, \Rn1, \Rd1, lsr #1
+.endm
+
+@ ----------------------------------------------------------------
+ .align 8
+ .global put_pixels16_arm
+put_pixels16_arm:
+ @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+ @ block = word aligned, pixles = unaligned
+ pld [r1]
+ stmfd sp!, {r4-r11, lr} @ R14 is also called LR
+ adr r5, 5f
+ ands r4, r1, #3
+ bic r1, r1, #3
+ add r5, r5, r4, lsl #2
+ ldrne pc, [r5]
+1:
+ ldmia r1, {r4-r7}
+ add r1, r1, r2
+ stmia r0, {r4-r7}
+ pld [r1]
+ subs r3, r3, #1
+ add r0, r0, r2
+ bne 1b
+ ldmfd sp!, {r4-r11, pc}
+ .align 8
+2:
+ ldmia r1, {r4-r8}
+ add r1, r1, r2
+ ADJ_ALIGN_QUADWORD_D 1, r9, r10, r11, r12, r4, r5, r6, r7, r8
+ pld [r1]
+ subs r3, r3, #1
+ stmia r0, {r9-r12}
+ add r0, r0, r2
+ bne 2b
+ ldmfd sp!, {r4-r11, pc}
+ .align 8
+3:
+ ldmia r1, {r4-r8}
+ add r1, r1, r2
+ ADJ_ALIGN_QUADWORD_D 2, r9, r10, r11, r12, r4, r5, r6, r7, r8
+ pld [r1]
+ subs r3, r3, #1
+ stmia r0, {r9-r12}
+ add r0, r0, r2
+ bne 3b
+ ldmfd sp!, {r4-r11, pc}
+ .align 8
+4:
+ ldmia r1, {r4-r8}
+ add r1, r1, r2
+ ADJ_ALIGN_QUADWORD_D 3, r9, r10, r11, r12, r4, r5, r6, r7, r8
+ pld [r1]
+ subs r3, r3, #1
+ stmia r0, {r9-r12}
+ add r0, r0, r2
+ bne 4b
+ ldmfd sp!, {r4-r11,pc}
+ .align 8
+5:
+ .word 1b
+ .word 2b
+ .word 3b
+ .word 4b
+
+@ ----------------------------------------------------------------
+ .align 8
+ .global put_pixels8_arm
+put_pixels8_arm:
+ @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+ @ block = word aligned, pixles = unaligned
+ pld [r1]
+ stmfd sp!, {r4-r5,lr} @ R14 is also called LR
+ adr r5, 5f
+ ands r4, r1, #3
+ bic r1, r1, #3
+ add r5, r5, r4, lsl #2
+ ldrne pc, [r5]
+1:
+ ldmia r1, {r4-r5}
+ add r1, r1, r2
+ subs r3, r3, #1
+ pld [r1]
+ stmia r0, {r4-r5}
+ add r0, r0, r2
+ bne 1b
+ ldmfd sp!, {r4-r5,pc}
+ .align 8
+2:
+ ldmia r1, {r4-r5, r12}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r12
+ pld [r1]
+ subs r3, r3, #1
+ stmia r0, {r4-r5}
+ add r0, r0, r2
+ bne 2b
+ ldmfd sp!, {r4-r5,pc}
+ .align 8
+3:
+ ldmia r1, {r4-r5, r12}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r12
+ pld [r1]
+ subs r3, r3, #1
+ stmia r0, {r4-r5}
+ add r0, r0, r2
+ bne 3b
+ ldmfd sp!, {r4-r5,pc}
+ .align 8
+4:
+ ldmia r1, {r4-r5, r12}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r12
+ pld [r1]
+ subs r3, r3, #1
+ stmia r0, {r4-r5}
+ add r0, r0, r2
+ bne 4b
+ ldmfd sp!, {r4-r5,pc}
+ .align 8
+5:
+ .word 1b
+ .word 2b
+ .word 3b
+ .word 4b
+
+@ ----------------------------------------------------------------
+ .align 8
+ .global put_pixels8_x2_arm
+put_pixels8_x2_arm:
+ @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+ @ block = word aligned, pixles = unaligned
+ pld [r1]
+ stmfd sp!, {r4-r10,lr} @ R14 is also called LR
+ adr r5, 5f
+ ands r4, r1, #3
+ ldr r12, [r5]
+ add r5, r5, r4, lsl #2
+ bic r1, r1, #3
+ ldrne pc, [r5]
+1:
+ ldmia r1, {r4-r5, r10}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD_D 1, r6, r7, r4, r5, r10
+ pld [r1]
+ RND_AVG32 r8, r9, r4, r5, r6, r7, r12
+ subs r3, r3, #1
+ stmia r0, {r8-r9}
+ add r0, r0, r2
+ bne 1b
+ ldmfd sp!, {r4-r10,pc}
+ .align 8
+2:
+ ldmia r1, {r4-r5, r10}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD_D 1, r6, r7, r4, r5, r10
+ ADJ_ALIGN_DOUBLEWORD_D 2, r8, r9, r4, r5, r10
+ pld [r1]
+ RND_AVG32 r4, r5, r6, r7, r8, r9, r12
+ subs r3, r3, #1
+ stmia r0, {r4-r5}
+ add r0, r0, r2
+ bne 2b
+ ldmfd sp!, {r4-r10,pc}
+ .align 8
+3:
+ ldmia r1, {r4-r5, r10}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD_D 2, r6, r7, r4, r5, r10
+ ADJ_ALIGN_DOUBLEWORD_D 3, r8, r9, r4, r5, r10
+ pld [r1]
+ RND_AVG32 r4, r5, r6, r7, r8, r9, r12
+ subs r3, r3, #1
+ stmia r0, {r4-r5}
+ add r0, r0, r2
+ bne 3b
+ ldmfd sp!, {r4-r10,pc}
+ .align 8
+4:
+ ldmia r1, {r4-r5, r10}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD_D 3, r6, r7, r4, r5, r10
+ pld [r1]
+ RND_AVG32 r8, r9, r6, r7, r5, r10, r12
+ subs r3, r3, #1
+ stmia r0, {r8-r9}
+ add r0, r0, r2
+ bne 4b
+ ldmfd sp!, {r4-r10,pc} @@ update PC with LR content.
+ .align 8
+5:
+ .word 0xFEFEFEFE
+ .word 2b
+ .word 3b
+ .word 4b
+
+ .align 8
+ .global put_no_rnd_pixels8_x2_arm
+put_no_rnd_pixels8_x2_arm:
+ @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+ @ block = word aligned, pixles = unaligned
+ pld [r1]
+ stmfd sp!, {r4-r10,lr} @ R14 is also called LR
+ adr r5, 5f
+ ands r4, r1, #3
+ ldr r12, [r5]
+ add r5, r5, r4, lsl #2
+ bic r1, r1, #3
+ ldrne pc, [r5]
+1:
+ ldmia r1, {r4-r5, r10}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD_D 1, r6, r7, r4, r5, r10
+ pld [r1]
+ NO_RND_AVG32 r8, r9, r4, r5, r6, r7, r12
+ subs r3, r3, #1
+ stmia r0, {r8-r9}
+ add r0, r0, r2
+ bne 1b
+ ldmfd sp!, {r4-r10,pc}
+ .align 8
+2:
+ ldmia r1, {r4-r5, r10}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD_D 1, r6, r7, r4, r5, r10
+ ADJ_ALIGN_DOUBLEWORD_D 2, r8, r9, r4, r5, r10
+ pld [r1]
+ NO_RND_AVG32 r4, r5, r6, r7, r8, r9, r12
+ subs r3, r3, #1
+ stmia r0, {r4-r5}
+ add r0, r0, r2
+ bne 2b
+ ldmfd sp!, {r4-r10,pc}
+ .align 8
+3:
+ ldmia r1, {r4-r5, r10}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD_D 2, r6, r7, r4, r5, r10
+ ADJ_ALIGN_DOUBLEWORD_D 3, r8, r9, r4, r5, r10
+ pld [r1]
+ NO_RND_AVG32 r4, r5, r6, r7, r8, r9, r12
+ subs r3, r3, #1
+ stmia r0, {r4-r5}
+ add r0, r0, r2
+ bne 3b
+ ldmfd sp!, {r4-r10,pc}
+ .align 8
+4:
+ ldmia r1, {r4-r5, r10}
+ add r1, r1, r2
+ ADJ_ALIGN_DOUBLEWORD_D 3, r6, r7, r4, r5, r10
+ pld [r1]
+ NO_RND_AVG32 r8, r9, r6, r7, r5, r10, r12
+ subs r3, r3, #1
+ stmia r0, {r8-r9}
+ add r0, r0, r2
+ bne 4b
+ ldmfd sp!, {r4-r10,pc} @@ update PC with LR content.
+ .align 8
+5:
+ .word 0xFEFEFEFE
+ .word 2b
+ .word 3b
+ .word 4b
+
+
+@ ----------------------------------------------------------------
+ .align 8
+ .global put_pixels8_y2_arm
+put_pixels8_y2_arm:
+ @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+ @ block = word aligned, pixles = unaligned
+ pld [r1]
+ stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+ adr r5, 5f
+ ands r4, r1, #3
+ mov r3, r3, lsr #1
+ ldr r12, [r5]
+ add r5, r5, r4, lsl #2
+ bic r1, r1, #3
+ ldrne pc, [r5]
+1:
+ ldmia r1, {r4-r5}
+ add r1, r1, r2
+6: ldmia r1, {r6-r7}
+ add r1, r1, r2
+ pld [r1]
+ RND_AVG32 r8, r9, r4, r5, r6, r7, r12
+ ldmia r1, {r4-r5}
+ add r1, r1, r2
+ stmia r0, {r8-r9}
+ add r0, r0, r2
+ pld [r1]
+ RND_AVG32 r8, r9, r6, r7, r4, r5, r12
+ subs r3, r3, #1
+ stmia r0, {r8-r9}
+ add r0, r0, r2
+ bne 6b
+ ldmfd sp!, {r4-r11,pc}
+ .align 8
+2:
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r6
+6: ldmia r1, {r7-r9}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 1, r7, r8, r9
+ RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r6
+ subs r3, r3, #1
+ RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ bne 6b
+ ldmfd sp!, {r4-r11,pc}
+ .align 8
+3:
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r6
+6: ldmia r1, {r7-r9}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 2, r7, r8, r9
+ RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r6
+ subs r3, r3, #1
+ RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ bne 6b
+ ldmfd sp!, {r4-r11,pc}
+ .align 8
+4:
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r6
+6: ldmia r1, {r7-r9}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 3, r7, r8, r9
+ RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r6
+ subs r3, r3, #1
+ RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ bne 6b
+ ldmfd sp!, {r4-r11,pc}
+
+ .align 8
+5:
+ .word 0xFEFEFEFE
+ .word 2b
+ .word 3b
+ .word 4b
+
+ .align 8
+ .global put_no_rnd_pixels8_y2_arm
+put_no_rnd_pixels8_y2_arm:
+ @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+ @ block = word aligned, pixles = unaligned
+ pld [r1]
+ stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+ adr r5, 5f
+ ands r4, r1, #3
+ mov r3, r3, lsr #1
+ ldr r12, [r5]
+ add r5, r5, r4, lsl #2
+ bic r1, r1, #3
+ ldrne pc, [r5]
+1:
+ ldmia r1, {r4-r5}
+ add r1, r1, r2
+6: ldmia r1, {r6-r7}
+ add r1, r1, r2
+ pld [r1]
+ NO_RND_AVG32 r8, r9, r4, r5, r6, r7, r12
+ ldmia r1, {r4-r5}
+ add r1, r1, r2
+ stmia r0, {r8-r9}
+ add r0, r0, r2
+ pld [r1]
+ NO_RND_AVG32 r8, r9, r6, r7, r4, r5, r12
+ subs r3, r3, #1
+ stmia r0, {r8-r9}
+ add r0, r0, r2
+ bne 6b
+ ldmfd sp!, {r4-r11,pc}
+ .align 8
+2:
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r6
+6: ldmia r1, {r7-r9}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 1, r7, r8, r9
+ NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r6
+ subs r3, r3, #1
+ NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ bne 6b
+ ldmfd sp!, {r4-r11,pc}
+ .align 8
+3:
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r6
+6: ldmia r1, {r7-r9}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 2, r7, r8, r9
+ NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r6
+ subs r3, r3, #1
+ NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ bne 6b
+ ldmfd sp!, {r4-r11,pc}
+ .align 8
+4:
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r6
+6: ldmia r1, {r7-r9}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 3, r7, r8, r9
+ NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ ldmia r1, {r4-r6}
+ add r1, r1, r2
+ pld [r1]
+ ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r6
+ subs r3, r3, #1
+ NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+ stmia r0, {r10-r11}
+ add r0, r0, r2
+ bne 6b
+ ldmfd sp!, {r4-r11,pc}
+ .align 8
+5:
+ .word 0xFEFEFEFE
+ .word 2b
+ .word 3b
+ .word 4b
+
+@ ----------------------------------------------------------------
+.macro RND_XY2_IT align, rnd
+ @ l1= (a & 0x03030303) + (b & 0x03030303) ?(+ 0x02020202)
+ @ h1= ((a & 0xFCFCFCFCUL) >> 2) + ((b & 0xFCFCFCFCUL) >> 2)
+.if \align == 0
+ ldmia r1, {r6-r8}
+.elseif \align == 3
+ ldmia r1, {r5-r7}
+.else
+ ldmia r1, {r8-r10}
+.endif
+ add r1, r1, r2
+ pld [r1]
+.if \align == 0
+ ADJ_ALIGN_DOUBLEWORD_D 1, r4, r5, r6, r7, r8
+.elseif \align == 1
+ ADJ_ALIGN_DOUBLEWORD_D 1, r4, r5, r8, r9, r10
+ ADJ_ALIGN_DOUBLEWORD_D 2, r6, r7, r8, r9, r10
+.elseif \align == 2
+ ADJ_ALIGN_DOUBLEWORD_D 2, r4, r5, r8, r9, r10
+ ADJ_ALIGN_DOUBLEWORD_D 3, r6, r7, r8, r9, r10
+.elseif \align == 3
+ ADJ_ALIGN_DOUBLEWORD_D 3, r4, r5, r5, r6, r7
+.endif
+ ldr r14, [r12, #0] @ 0x03030303
+ tst r3, #1
+ and r8, r4, r14
+ and r9, r5, r14
+ and r10, r6, r14
+ and r11, r7, r14
+.if \rnd == 1
+ ldreq r14, [r12, #16] @ 0x02020202
+.else
+ ldreq r14, [r12, #28] @ 0x01010101
+.endif
+ add r8, r8, r10
+ add r9, r9, r11
+ addeq r8, r8, r14
+ addeq r9, r9, r14
+ ldr r14, [r12, #20] @ 0xFCFCFCFC >> 2
+ and r4, r14, r4, lsr #2
+ and r5, r14, r5, lsr #2
+ and r6, r14, r6, lsr #2
+ and r7, r14, r7, lsr #2
+ add r10, r4, r6
+ add r11, r5, r7
+.endm
+
+.macro RND_XY2_EXPAND align, rnd
+ RND_XY2_IT \align, \rnd
+6: stmfd sp!, {r8-r11}
+ RND_XY2_IT \align, \rnd
+ ldmfd sp!, {r4-r7}
+ add r4, r4, r8
+ add r5, r5, r9
+ add r6, r6, r10
+ add r7, r7, r11
+ ldr r14, [r12, #24] @ 0x0F0F0F0F
+ and r4, r14, r4, lsr #2
+ and r5, r14, r5, lsr #2
+ add r4, r4, r6
+ add r5, r5, r7
+ subs r3, r3, #1
+ stmia r0, {r4-r5}
+ add r0, r0, r2
+ bne 6b
+ ldmfd sp!, {r4-r11,pc}
+.endm
+
+ .align 8
+ .global put_pixels8_xy2_arm
+put_pixels8_xy2_arm:
+ @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+ @ block = word aligned, pixles = unaligned
+ pld [r1]
+ stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+ adrl r12, 5f
+ ands r4, r1, #3
+ add r5, r12, r4, lsl #2
+ bic r1, r1, #3
+ ldrne pc, [r5]
+1:
+ RND_XY2_EXPAND 0, 1
+
+ .align 8
+2:
+ RND_XY2_EXPAND 1, 1
+
+ .align 8
+3:
+ RND_XY2_EXPAND 2, 1
+
+ .align 8
+4:
+ RND_XY2_EXPAND 3, 1
+
+5:
+ .word 0x03030303
+ .word 2b
+ .word 3b
+ .word 4b
+ .word 0x02020202
+ .word 0xFCFCFCFC >> 2
+ .word 0x0F0F0F0F
+ .word 0x01010101
+
+ .align 8
+ .global put_no_rnd_pixels8_xy2_arm
+put_no_rnd_pixels8_xy2_arm:
+ @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+ @ block = word aligned, pixles = unaligned
+ pld [r1]
+ stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+ adrl r12, 5f
+ ands r4, r1, #3
+ add r5, r12, r4, lsl #2
+ bic r1, r1, #3
+ ldrne pc, [r5]
+1:
+ RND_XY2_EXPAND 0, 0
+
+ .align 8
+2:
+ RND_XY2_EXPAND 1, 0
+
+ .align 8
+3:
+ RND_XY2_EXPAND 2, 0
+
+ .align 8
+4:
+ RND_XY2_EXPAND 3, 0
+
+5:
+ .word 0x03030303
+ .word 2b
+ .word 3b
+ .word 4b
+ .word 0x02020202
+ .word 0xFCFCFCFC >> 2
+ .word 0x0F0F0F0F
+ .word 0x01010101
diff --git a/src/libffmpeg/libavcodec/armv4l/dsputil_iwmmxt.c b/src/libffmpeg/libavcodec/armv4l/dsputil_iwmmxt.c
new file mode 100644
index 000000000..d7401e760
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/dsputil_iwmmxt.c
@@ -0,0 +1,188 @@
+/*
+ * iWMMXt optimized DSP utils
+ * Copyright (c) 2004 AGAWA Koji
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "../dsputil.h"
+
+#define DEF(x, y) x ## _no_rnd_ ## y ##_iwmmxt
+#define SET_RND(regd) __asm__ __volatile__ ("mov r12, #1 \n\t tbcsth " #regd ", r12":::"r12");
+#define WAVG2B "wavg2b"
+#include "dsputil_iwmmxt_rnd.h"
+#undef DEF
+#undef SET_RND
+#undef WAVG2B
+
+#define DEF(x, y) x ## _ ## y ##_iwmmxt
+#define SET_RND(regd) __asm__ __volatile__ ("mov r12, #2 \n\t tbcsth " #regd ", r12":::"r12");
+#define WAVG2B "wavg2br"
+#include "dsputil_iwmmxt_rnd.h"
+#undef DEF
+#undef SET_RND
+#undef WAVG2BR
+
+// need scheduling
+#define OP(AVG) \
+ asm volatile ( \
+ /* alignment */ \
+ "and r12, %[pixels], #7 \n\t" \
+ "bic %[pixels], %[pixels], #7 \n\t" \
+ "tmcr wcgr1, r12 \n\t" \
+ \
+ "wldrd wr0, [%[pixels]] \n\t" \
+ "wldrd wr1, [%[pixels], #8] \n\t" \
+ "add %[pixels], %[pixels], %[line_size] \n\t" \
+ "walignr1 wr4, wr0, wr1 \n\t" \
+ \
+ "1: \n\t" \
+ \
+ "wldrd wr2, [%[pixels]] \n\t" \
+ "wldrd wr3, [%[pixels], #8] \n\t" \
+ "add %[pixels], %[pixels], %[line_size] \n\t" \
+ "pld [%[pixels]] \n\t" \
+ "walignr1 wr5, wr2, wr3 \n\t" \
+ AVG " wr6, wr4, wr5 \n\t" \
+ "wstrd wr6, [%[block]] \n\t" \
+ "add %[block], %[block], %[line_size] \n\t" \
+ \
+ "wldrd wr0, [%[pixels]] \n\t" \
+ "wldrd wr1, [%[pixels], #8] \n\t" \
+ "add %[pixels], %[pixels], %[line_size] \n\t" \
+ "walignr1 wr4, wr0, wr1 \n\t" \
+ "pld [%[pixels]] \n\t" \
+ AVG " wr6, wr4, wr5 \n\t" \
+ "wstrd wr6, [%[block]] \n\t" \
+ "add %[block], %[block], %[line_size] \n\t" \
+ \
+ "subs %[h], %[h], #2 \n\t" \
+ "bne 1b \n\t" \
+ : [block]"+r"(block), [pixels]"+r"(pixels), [h]"+r"(h) \
+ : [line_size]"r"(line_size) \
+ : "memory", "r12");
+void put_pixels8_y2_iwmmxt(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ OP("wavg2br");
+}
+void put_no_rnd_pixels8_y2_iwmmxt(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ OP("wavg2b");
+}
+#undef OP
+
+void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size)
+{
+ uint8_t *pixels2 = pixels + line_size;
+
+ __asm__ __volatile__ (
+ "mov r12, #4 \n\t"
+ "1: \n\t"
+ "pld [%[pixels], %[line_size2]] \n\t"
+ "pld [%[pixels2], %[line_size2]] \n\t"
+ "wldrd wr4, [%[pixels]] \n\t"
+ "wldrd wr5, [%[pixels2]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "wunpckelub wr6, wr4 \n\t"
+ "wldrd wr0, [%[block]] \n\t"
+ "wunpckehub wr7, wr4 \n\t"
+ "wldrd wr1, [%[block], #8] \n\t"
+ "wunpckelub wr8, wr5 \n\t"
+ "wldrd wr2, [%[block], #16] \n\t"
+ "wunpckehub wr9, wr5 \n\t"
+ "wldrd wr3, [%[block], #24] \n\t"
+ "add %[block], %[block], #32 \n\t"
+ "waddhss wr10, wr0, wr6 \n\t"
+ "waddhss wr11, wr1, wr7 \n\t"
+ "waddhss wr12, wr2, wr8 \n\t"
+ "waddhss wr13, wr3, wr9 \n\t"
+ "wpackhus wr14, wr10, wr11 \n\t"
+ "wpackhus wr15, wr12, wr13 \n\t"
+ "wstrd wr14, [%[pixels]] \n\t"
+ "add %[pixels], %[pixels], %[line_size2] \n\t"
+ "subs r12, r12, #1 \n\t"
+ "wstrd wr15, [%[pixels2]] \n\t"
+ "add %[pixels2], %[pixels2], %[line_size2] \n\t"
+ "bne 1b \n\t"
+ : [block]"+r"(block), [pixels]"+r"(pixels), [pixels2]"+r"(pixels2)
+ : [line_size2]"r"(line_size << 1)
+ : "cc", "memory", "r12");
+}
+
+static void nop(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+ return;
+}
+
+int mm_flags; /* multimedia extension flags */
+
+int mm_support(void)
+{
+ return 0; /* TODO, implement proper detection */
+}
+
+void dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx)
+{
+ mm_flags = mm_support();
+
+ if (avctx->dsp_mask) {
+ if (avctx->dsp_mask & FF_MM_FORCE)
+ mm_flags |= (avctx->dsp_mask & 0xffff);
+ else
+ mm_flags &= ~(avctx->dsp_mask & 0xffff);
+ }
+
+ if (!(mm_flags & MM_IWMMXT)) return;
+
+ c->add_pixels_clamped = add_pixels_clamped_iwmmxt;
+
+ c->put_pixels_tab[0][0] = put_pixels16_iwmmxt;
+ c->put_pixels_tab[0][1] = put_pixels16_x2_iwmmxt;
+ c->put_pixels_tab[0][2] = put_pixels16_y2_iwmmxt;
+ c->put_pixels_tab[0][3] = put_pixels16_xy2_iwmmxt;
+ c->put_no_rnd_pixels_tab[0][0] = put_pixels16_iwmmxt;
+ c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_iwmmxt;
+ c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_iwmmxt;
+ c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_iwmmxt;
+
+ c->put_pixels_tab[1][0] = put_pixels8_iwmmxt;
+ c->put_pixels_tab[1][1] = put_pixels8_x2_iwmmxt;
+ c->put_pixels_tab[1][2] = put_pixels8_y2_iwmmxt;
+ c->put_pixels_tab[1][3] = put_pixels8_xy2_iwmmxt;
+ c->put_no_rnd_pixels_tab[1][0] = put_pixels8_iwmmxt;
+ c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_iwmmxt;
+ c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_iwmmxt;
+ c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_iwmmxt;
+
+ c->avg_pixels_tab[0][0] = avg_pixels16_iwmmxt;
+ c->avg_pixels_tab[0][1] = avg_pixels16_x2_iwmmxt;
+ c->avg_pixels_tab[0][2] = avg_pixels16_y2_iwmmxt;
+ c->avg_pixels_tab[0][3] = avg_pixels16_xy2_iwmmxt;
+ c->avg_no_rnd_pixels_tab[0][0] = avg_pixels16_iwmmxt;
+ c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_iwmmxt;
+ c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_iwmmxt;
+ c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_iwmmxt;
+
+ c->avg_pixels_tab[1][0] = avg_pixels8_iwmmxt;
+ c->avg_pixels_tab[1][1] = avg_pixels8_x2_iwmmxt;
+ c->avg_pixels_tab[1][2] = avg_pixels8_y2_iwmmxt;
+ c->avg_pixels_tab[1][3] = avg_pixels8_xy2_iwmmxt;
+ c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_iwmmxt;
+ c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_iwmmxt;
+ c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_iwmmxt;
+ c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_iwmmxt;
+}
diff --git a/src/libffmpeg/libavcodec/armv4l/dsputil_iwmmxt_rnd.h b/src/libffmpeg/libavcodec/armv4l/dsputil_iwmmxt_rnd.h
new file mode 100644
index 000000000..51ba61c47
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/dsputil_iwmmxt_rnd.h
@@ -0,0 +1,1114 @@
+/*
+ * iWMMXt optimized DSP utils
+ * copyright (c) 2004 AGAWA Koji
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+void DEF(put, pixels8)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ __asm__ __volatile__ (
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r4, %[pixels], %[line_size] \n\t"
+ "add r5, %[block], %[line_size] \n\t"
+ "mov %[line_size], %[line_size], lsl #1 \n\t"
+ "1: \n\t"
+ "wldrd wr0, [%[pixels]] \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "wldrd wr1, [%[pixels], #8] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "wldrd wr3, [r4] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "wldrd wr4, [r4, #8] \n\t"
+ "add r4, r4, %[line_size] \n\t"
+ "walignr1 wr8, wr0, wr1 \n\t"
+ "pld [r4] \n\t"
+ "pld [r4, #32] \n\t"
+ "walignr1 wr10, wr3, wr4 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "wstrd wr10, [r5] \n\t"
+ "add r5, r5, %[line_size] \n\t"
+ "bne 1b \n\t"
+ : [block]"+r"(block), [pixels]"+r"(pixels), [line_size]"+r"(stride), [h]"+r"(h)
+ :
+ : "memory", "r4", "r5", "r12");
+}
+
+void DEF(avg, pixels8)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ __asm__ __volatile__ (
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r4, %[pixels], %[line_size] \n\t"
+ "add r5, %[block], %[line_size] \n\t"
+ "mov %[line_size], %[line_size], lsl #1 \n\t"
+ "1: \n\t"
+ "wldrd wr0, [%[pixels]] \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "wldrd wr1, [%[pixels], #8] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "wldrd wr3, [r4] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "wldrd wr4, [r4, #8] \n\t"
+ "add r4, r4, %[line_size] \n\t"
+ "walignr1 wr8, wr0, wr1 \n\t"
+ "wldrd wr0, [%[block]] \n\t"
+ "wldrd wr2, [r5] \n\t"
+ "pld [r4] \n\t"
+ "pld [r4, #32] \n\t"
+ "walignr1 wr10, wr3, wr4 \n\t"
+ WAVG2B" wr8, wr8, wr0 \n\t"
+ WAVG2B" wr10, wr10, wr2 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "wstrd wr10, [r5] \n\t"
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "add r5, r5, %[line_size] \n\t"
+ "pld [r5] \n\t"
+ "pld [r5, #32] \n\t"
+ "bne 1b \n\t"
+ : [block]"+r"(block), [pixels]"+r"(pixels), [line_size]"+r"(stride), [h]"+r"(h)
+ :
+ : "memory", "r4", "r5", "r12");
+}
+
+void DEF(put, pixels16)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ __asm__ __volatile__ (
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r4, %[pixels], %[line_size] \n\t"
+ "add r5, %[block], %[line_size] \n\t"
+ "mov %[line_size], %[line_size], lsl #1 \n\t"
+ "1: \n\t"
+ "wldrd wr0, [%[pixels]] \n\t"
+ "wldrd wr1, [%[pixels], #8] \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "wldrd wr2, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "wldrd wr3, [r4] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr8, wr0, wr1 \n\t"
+ "wldrd wr4, [r4, #8] \n\t"
+ "walignr1 wr9, wr1, wr2 \n\t"
+ "wldrd wr5, [r4, #16] \n\t"
+ "add r4, r4, %[line_size] \n\t"
+ "pld [r4] \n\t"
+ "pld [r4, #32] \n\t"
+ "walignr1 wr10, wr3, wr4 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "walignr1 wr11, wr4, wr5 \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "wstrd wr10, [r5] \n\t"
+ "wstrd wr11, [r5, #8] \n\t"
+ "add r5, r5, %[line_size] \n\t"
+ "bne 1b \n\t"
+ : [block]"+r"(block), [pixels]"+r"(pixels), [line_size]"+r"(stride), [h]"+r"(h)
+ :
+ : "memory", "r4", "r5", "r12");
+}
+
+void DEF(avg, pixels16)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ __asm__ __volatile__ (
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r4, %[pixels], %[line_size]\n\t"
+ "add r5, %[block], %[line_size] \n\t"
+ "mov %[line_size], %[line_size], lsl #1 \n\t"
+ "1: \n\t"
+ "wldrd wr0, [%[pixels]] \n\t"
+ "wldrd wr1, [%[pixels], #8] \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "wldrd wr2, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "wldrd wr3, [r4] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr8, wr0, wr1 \n\t"
+ "wldrd wr4, [r4, #8] \n\t"
+ "walignr1 wr9, wr1, wr2 \n\t"
+ "wldrd wr5, [r4, #16] \n\t"
+ "add r4, r4, %[line_size] \n\t"
+ "wldrd wr0, [%[block]] \n\t"
+ "pld [r4] \n\t"
+ "wldrd wr1, [%[block], #8] \n\t"
+ "pld [r4, #32] \n\t"
+ "wldrd wr2, [r5] \n\t"
+ "walignr1 wr10, wr3, wr4 \n\t"
+ "wldrd wr3, [r5, #8] \n\t"
+ WAVG2B" wr8, wr8, wr0 \n\t"
+ WAVG2B" wr9, wr9, wr1 \n\t"
+ WAVG2B" wr10, wr10, wr2 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "walignr1 wr11, wr4, wr5 \n\t"
+ WAVG2B" wr11, wr11, wr3 \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "wstrd wr10, [r5] \n\t"
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "wstrd wr11, [r5, #8] \n\t"
+ "add r5, r5, %[line_size] \n\t"
+ "pld [r5] \n\t"
+ "pld [r5, #32] \n\t"
+ "bne 1b \n\t"
+ : [block]"+r"(block), [pixels]"+r"(pixels), [line_size]"+r"(stride), [h]"+r"(h)
+ :
+ : "memory", "r4", "r5", "r12");
+}
+
+void DEF(put, pixels8_x2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
+ __asm__ __volatile__(
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r12, r12, #1 \n\t"
+ "add r4, %[pixels], %[line_size]\n\t"
+ "tmcr wcgr2, r12 \n\t"
+ "add r5, %[block], %[line_size] \n\t"
+ "mov %[line_size], %[line_size], lsl #1 \n\t"
+
+ "1: \n\t"
+ "wldrd wr10, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "wldrd wr13, [r4] \n\t"
+ "pld [%[pixels]] \n\t"
+ "wldrd wr14, [r4, #8] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "add r4, r4, %[line_size] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "pld [r4] \n\t"
+ "pld [r4, #32] \n\t"
+ "walignr1 wr2, wr13, wr14 \n\t"
+ "wmoveq wr4, wr11 \n\t"
+ "wmoveq wr6, wr14 \n\t"
+ "walignr2ne wr4, wr10, wr11 \n\t"
+ "walignr2ne wr6, wr13, wr14 \n\t"
+ WAVG2B" wr0, wr0, wr4 \n\t"
+ WAVG2B" wr2, wr2, wr6 \n\t"
+ "wstrd wr0, [%[block]] \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "wstrd wr2, [r5] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "add r5, r5, %[line_size] \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
+ :
+ : "r4", "r5", "r12", "memory");
+}
+
+void DEF(put, pixels16_x2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
+ __asm__ __volatile__(
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r12, r12, #1 \n\t"
+ "add r4, %[pixels], %[line_size]\n\t"
+ "tmcr wcgr2, r12 \n\t"
+ "add r5, %[block], %[line_size] \n\t"
+ "mov %[line_size], %[line_size], lsl #1 \n\t"
+
+ "1: \n\t"
+ "wldrd wr10, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "wldrd wr12, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "wldrd wr13, [r4] \n\t"
+ "pld [%[pixels]] \n\t"
+ "wldrd wr14, [r4, #8] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "wldrd wr15, [r4, #16] \n\t"
+ "add r4, r4, %[line_size] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "pld [r4] \n\t"
+ "pld [r4, #32] \n\t"
+ "walignr1 wr1, wr11, wr12 \n\t"
+ "walignr1 wr2, wr13, wr14 \n\t"
+ "walignr1 wr3, wr14, wr15 \n\t"
+ "wmoveq wr4, wr11 \n\t"
+ "wmoveq wr5, wr12 \n\t"
+ "wmoveq wr6, wr14 \n\t"
+ "wmoveq wr7, wr15 \n\t"
+ "walignr2ne wr4, wr10, wr11 \n\t"
+ "walignr2ne wr5, wr11, wr12 \n\t"
+ "walignr2ne wr6, wr13, wr14 \n\t"
+ "walignr2ne wr7, wr14, wr15 \n\t"
+ WAVG2B" wr0, wr0, wr4 \n\t"
+ WAVG2B" wr1, wr1, wr5 \n\t"
+ "wstrd wr0, [%[block]] \n\t"
+ WAVG2B" wr2, wr2, wr6 \n\t"
+ "wstrd wr1, [%[block], #8] \n\t"
+ WAVG2B" wr3, wr3, wr7 \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "wstrd wr2, [r5] \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "wstrd wr3, [r5, #8] \n\t"
+ "add r5, r5, %[line_size] \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
+ :
+ : "r4", "r5", "r12", "memory");
+}
+
+void DEF(avg, pixels8_x2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
+ __asm__ __volatile__(
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r12, r12, #1 \n\t"
+ "add r4, %[pixels], %[line_size]\n\t"
+ "tmcr wcgr2, r12 \n\t"
+ "add r5, %[block], %[line_size] \n\t"
+ "mov %[line_size], %[line_size], lsl #1 \n\t"
+ "pld [r5] \n\t"
+ "pld [r5, #32] \n\t"
+
+ "1: \n\t"
+ "wldrd wr10, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "wldrd wr13, [r4] \n\t"
+ "pld [%[pixels]] \n\t"
+ "wldrd wr14, [r4, #8] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "add r4, r4, %[line_size] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "pld [r4] \n\t"
+ "pld [r4, #32] \n\t"
+ "walignr1 wr2, wr13, wr14 \n\t"
+ "wmoveq wr4, wr11 \n\t"
+ "wmoveq wr6, wr14 \n\t"
+ "walignr2ne wr4, wr10, wr11 \n\t"
+ "wldrd wr10, [%[block]] \n\t"
+ "walignr2ne wr6, wr13, wr14 \n\t"
+ "wldrd wr12, [r5] \n\t"
+ WAVG2B" wr0, wr0, wr4 \n\t"
+ WAVG2B" wr2, wr2, wr6 \n\t"
+ WAVG2B" wr0, wr0, wr10 \n\t"
+ WAVG2B" wr2, wr2, wr12 \n\t"
+ "wstrd wr0, [%[block]] \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "wstrd wr2, [r5] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "add r5, r5, %[line_size] \n\t"
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "pld [r5] \n\t"
+ "pld [r5, #32] \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
+ :
+ : "r4", "r5", "r12", "memory");
+}
+
+void DEF(avg, pixels16_x2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
+ __asm__ __volatile__(
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r12, r12, #1 \n\t"
+ "add r4, %[pixels], %[line_size]\n\t"
+ "tmcr wcgr2, r12 \n\t"
+ "add r5, %[block], %[line_size] \n\t"
+ "mov %[line_size], %[line_size], lsl #1 \n\t"
+ "pld [r5] \n\t"
+ "pld [r5, #32] \n\t"
+
+ "1: \n\t"
+ "wldrd wr10, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "wldrd wr12, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "wldrd wr13, [r4] \n\t"
+ "pld [%[pixels]] \n\t"
+ "wldrd wr14, [r4, #8] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "wldrd wr15, [r4, #16] \n\t"
+ "add r4, r4, %[line_size] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "pld [r4] \n\t"
+ "pld [r4, #32] \n\t"
+ "walignr1 wr1, wr11, wr12 \n\t"
+ "walignr1 wr2, wr13, wr14 \n\t"
+ "walignr1 wr3, wr14, wr15 \n\t"
+ "wmoveq wr4, wr11 \n\t"
+ "wmoveq wr5, wr12 \n\t"
+ "wmoveq wr6, wr14 \n\t"
+ "wmoveq wr7, wr15 \n\t"
+ "walignr2ne wr4, wr10, wr11 \n\t"
+ "walignr2ne wr5, wr11, wr12 \n\t"
+ "walignr2ne wr6, wr13, wr14 \n\t"
+ "walignr2ne wr7, wr14, wr15 \n\t"
+ "wldrd wr10, [%[block]] \n\t"
+ WAVG2B" wr0, wr0, wr4 \n\t"
+ "wldrd wr11, [%[block], #8] \n\t"
+ WAVG2B" wr1, wr1, wr5 \n\t"
+ "wldrd wr12, [r5] \n\t"
+ WAVG2B" wr2, wr2, wr6 \n\t"
+ "wldrd wr13, [r5, #8] \n\t"
+ WAVG2B" wr3, wr3, wr7 \n\t"
+ WAVG2B" wr0, wr0, wr10 \n\t"
+ WAVG2B" wr1, wr1, wr11 \n\t"
+ WAVG2B" wr2, wr2, wr12 \n\t"
+ WAVG2B" wr3, wr3, wr13 \n\t"
+ "wstrd wr0, [%[block]] \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "wstrd wr1, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "wstrd wr2, [r5] \n\t"
+ "pld [%[block]] \n\t"
+ "wstrd wr3, [r5, #8] \n\t"
+ "add r5, r5, %[line_size] \n\t"
+ "pld [%[block], #32] \n\t"
+ "pld [r5] \n\t"
+ "pld [r5, #32] \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
+ :
+ :"r4", "r5", "r12", "memory");
+}
+
+void DEF(avg, pixels8_y2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ __asm__ __volatile__(
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "and r12, %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+
+ "wldrd wr10, [%[pixels]] \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "pld [%[block]] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+
+ "1: \n\t"
+ "wldrd wr10, [%[pixels]] \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr4, wr10, wr11 \n\t"
+ "wldrd wr10, [%[block]] \n\t"
+ WAVG2B" wr8, wr0, wr4 \n\t"
+ WAVG2B" wr8, wr8, wr10 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ "wldrd wr10, [%[pixels]] \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "pld [%[block]] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "wldrd wr10, [%[block]] \n\t"
+ WAVG2B" wr8, wr0, wr4 \n\t"
+ WAVG2B" wr8, wr8, wr10 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ "subs %[h], %[h], #2 \n\t"
+ "pld [%[block]] \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
+ :
+ : "cc", "memory", "r12");
+}
+
+void DEF(put, pixels16_y2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ __asm__ __volatile__(
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "and r12, %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+
+ "wldrd wr10, [%[pixels]] \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "wldrd wr12, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "walignr1 wr1, wr11, wr12 \n\t"
+
+ "1: \n\t"
+ "wldrd wr10, [%[pixels]] \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "wldrd wr12, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr4, wr10, wr11 \n\t"
+ "walignr1 wr5, wr11, wr12 \n\t"
+ WAVG2B" wr8, wr0, wr4 \n\t"
+ WAVG2B" wr9, wr1, wr5 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ "wldrd wr10, [%[pixels]] \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "wldrd wr12, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "walignr1 wr1, wr11, wr12 \n\t"
+ WAVG2B" wr8, wr0, wr4 \n\t"
+ WAVG2B" wr9, wr1, wr5 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ "subs %[h], %[h], #2 \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
+ :
+ : "r4", "r5", "r12", "memory");
+}
+
+void DEF(avg, pixels16_y2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ int stride = line_size;
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ __asm__ __volatile__(
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "and r12, %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+
+ "wldrd wr10, [%[pixels]] \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "pld [%[block]] \n\t"
+ "wldrd wr12, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "walignr1 wr1, wr11, wr12 \n\t"
+
+ "1: \n\t"
+ "wldrd wr10, [%[pixels]] \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "wldrd wr12, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr4, wr10, wr11 \n\t"
+ "walignr1 wr5, wr11, wr12 \n\t"
+ "wldrd wr10, [%[block]] \n\t"
+ "wldrd wr11, [%[block], #8] \n\t"
+ WAVG2B" wr8, wr0, wr4 \n\t"
+ WAVG2B" wr9, wr1, wr5 \n\t"
+ WAVG2B" wr8, wr8, wr10 \n\t"
+ WAVG2B" wr9, wr9, wr11 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ "wldrd wr10, [%[pixels]] \n\t"
+ "wldrd wr11, [%[pixels], #8] \n\t"
+ "pld [%[block]] \n\t"
+ "wldrd wr12, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr0, wr10, wr11 \n\t"
+ "walignr1 wr1, wr11, wr12 \n\t"
+ "wldrd wr10, [%[block]] \n\t"
+ "wldrd wr11, [%[block], #8] \n\t"
+ WAVG2B" wr8, wr0, wr4 \n\t"
+ WAVG2B" wr9, wr1, wr5 \n\t"
+ WAVG2B" wr8, wr8, wr10 \n\t"
+ WAVG2B" wr9, wr9, wr11 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ "subs %[h], %[h], #2 \n\t"
+ "pld [%[block]] \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
+ :
+ : "r4", "r5", "r12", "memory");
+}
+
+void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
+ __asm__ __volatile__(
+ "pld [%[pixels]] \n\t"
+ "mov r12, #2 \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "tmcr wcgr0, r12 \n\t" /* for shift value */
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+
+ // [wr0 wr1 wr2 wr3] <= *
+ // [wr4 wr5 wr6 wr7]
+ "wldrd wr12, [%[pixels]] \n\t"
+ "add r12, r12, #1 \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "tmcr wcgr2, r12 \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "cmp r12, #8 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr2, wr12, wr13 \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "wunpckelub wr0, wr2 \n\t"
+ "wunpckehub wr1, wr2 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "waddhus wr0, wr0, wr8 \n\t"
+ "waddhus wr1, wr1, wr9 \n\t"
+
+ "1: \n\t"
+ // [wr0 wr1 wr2 wr3]
+ // [wr4 wr5 wr6 wr7] <= *
+ "wldrd wr12, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "walignr1 wr6, wr12, wr13 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "wunpckelub wr4, wr6 \n\t"
+ "wunpckehub wr5, wr6 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "waddhus wr4, wr4, wr8 \n\t"
+ "waddhus wr5, wr5, wr9 \n\t"
+ "waddhus wr8, wr0, wr4 \n\t"
+ "waddhus wr9, wr1, wr5 \n\t"
+ "waddhus wr8, wr8, wr15 \n\t"
+ "waddhus wr9, wr9, wr15 \n\t"
+ "wsrlhg wr8, wr8, wcgr0 \n\t"
+ "wsrlhg wr9, wr9, wcgr0 \n\t"
+ "wpackhus wr8, wr8, wr9 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ // [wr0 wr1 wr2 wr3] <= *
+ // [wr4 wr5 wr6 wr7]
+ "wldrd wr12, [%[pixels]] \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "walignr1 wr2, wr12, wr13 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "wunpckelub wr0, wr2 \n\t"
+ "wunpckehub wr1, wr2 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "waddhus wr0, wr0, wr8 \n\t"
+ "waddhus wr1, wr1, wr9 \n\t"
+ "waddhus wr8, wr0, wr4 \n\t"
+ "waddhus wr9, wr1, wr5 \n\t"
+ "waddhus wr8, wr8, wr15 \n\t"
+ "waddhus wr9, wr9, wr15 \n\t"
+ "wsrlhg wr8, wr8, wcgr0 \n\t"
+ "wsrlhg wr9, wr9, wcgr0 \n\t"
+ "wpackhus wr8, wr8, wr9 \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block)
+ : [line_size]"r"(line_size)
+ : "r12", "memory");
+}
+
+void DEF(put, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
+ __asm__ __volatile__(
+ "pld [%[pixels]] \n\t"
+ "mov r12, #2 \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "tmcr wcgr0, r12 \n\t" /* for shift value */
+ /* alignment */
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r12, r12, #1 \n\t"
+ "tmcr wcgr2, r12 \n\t"
+
+ // [wr0 wr1 wr2 wr3] <= *
+ // [wr4 wr5 wr6 wr7]
+ "wldrd wr12, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "wldrd wr14, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "walignr1 wr2, wr12, wr13 \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr3, wr13, wr14 \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "wmoveq wr11, wr14 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "walignr2ne wr11, wr13, wr14 \n\t"
+ "wunpckelub wr0, wr2 \n\t"
+ "wunpckehub wr1, wr2 \n\t"
+ "wunpckelub wr2, wr3 \n\t"
+ "wunpckehub wr3, wr3 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "wunpckelub wr10, wr11 \n\t"
+ "wunpckehub wr11, wr11 \n\t"
+ "waddhus wr0, wr0, wr8 \n\t"
+ "waddhus wr1, wr1, wr9 \n\t"
+ "waddhus wr2, wr2, wr10 \n\t"
+ "waddhus wr3, wr3, wr11 \n\t"
+
+ "1: \n\t"
+ // [wr0 wr1 wr2 wr3]
+ // [wr4 wr5 wr6 wr7] <= *
+ "wldrd wr12, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "wldrd wr14, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "walignr1 wr6, wr12, wr13 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr7, wr13, wr14 \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "wmoveq wr11, wr14 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "walignr2ne wr11, wr13, wr14 \n\t"
+ "wunpckelub wr4, wr6 \n\t"
+ "wunpckehub wr5, wr6 \n\t"
+ "wunpckelub wr6, wr7 \n\t"
+ "wunpckehub wr7, wr7 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "wunpckelub wr10, wr11 \n\t"
+ "wunpckehub wr11, wr11 \n\t"
+ "waddhus wr4, wr4, wr8 \n\t"
+ "waddhus wr5, wr5, wr9 \n\t"
+ "waddhus wr6, wr6, wr10 \n\t"
+ "waddhus wr7, wr7, wr11 \n\t"
+ "waddhus wr8, wr0, wr4 \n\t"
+ "waddhus wr9, wr1, wr5 \n\t"
+ "waddhus wr10, wr2, wr6 \n\t"
+ "waddhus wr11, wr3, wr7 \n\t"
+ "waddhus wr8, wr8, wr15 \n\t"
+ "waddhus wr9, wr9, wr15 \n\t"
+ "waddhus wr10, wr10, wr15 \n\t"
+ "waddhus wr11, wr11, wr15 \n\t"
+ "wsrlhg wr8, wr8, wcgr0 \n\t"
+ "wsrlhg wr9, wr9, wcgr0 \n\t"
+ "wsrlhg wr10, wr10, wcgr0 \n\t"
+ "wsrlhg wr11, wr11, wcgr0 \n\t"
+ "wpackhus wr8, wr8, wr9 \n\t"
+ "wpackhus wr9, wr10, wr11 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ // [wr0 wr1 wr2 wr3] <= *
+ // [wr4 wr5 wr6 wr7]
+ "wldrd wr12, [%[pixels]] \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "wldrd wr14, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "walignr1 wr2, wr12, wr13 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr3, wr13, wr14 \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "wmoveq wr11, wr14 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "walignr2ne wr11, wr13, wr14 \n\t"
+ "wunpckelub wr0, wr2 \n\t"
+ "wunpckehub wr1, wr2 \n\t"
+ "wunpckelub wr2, wr3 \n\t"
+ "wunpckehub wr3, wr3 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "wunpckelub wr10, wr11 \n\t"
+ "wunpckehub wr11, wr11 \n\t"
+ "waddhus wr0, wr0, wr8 \n\t"
+ "waddhus wr1, wr1, wr9 \n\t"
+ "waddhus wr2, wr2, wr10 \n\t"
+ "waddhus wr3, wr3, wr11 \n\t"
+ "waddhus wr8, wr0, wr4 \n\t"
+ "waddhus wr9, wr1, wr5 \n\t"
+ "waddhus wr10, wr2, wr6 \n\t"
+ "waddhus wr11, wr3, wr7 \n\t"
+ "waddhus wr8, wr8, wr15 \n\t"
+ "waddhus wr9, wr9, wr15 \n\t"
+ "waddhus wr10, wr10, wr15 \n\t"
+ "waddhus wr11, wr11, wr15 \n\t"
+ "wsrlhg wr8, wr8, wcgr0 \n\t"
+ "wsrlhg wr9, wr9, wcgr0 \n\t"
+ "wsrlhg wr10, wr10, wcgr0 \n\t"
+ "wsrlhg wr11, wr11, wcgr0 \n\t"
+ "wpackhus wr8, wr8, wr9 \n\t"
+ "wpackhus wr9, wr10, wr11 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ "subs %[h], %[h], #2 \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block)
+ : [line_size]"r"(line_size)
+ : "r12", "memory");
+}
+
+void DEF(avg, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
+ __asm__ __volatile__(
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "pld [%[pixels]] \n\t"
+ "mov r12, #2 \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "tmcr wcgr0, r12 \n\t" /* for shift value */
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+
+ // [wr0 wr1 wr2 wr3] <= *
+ // [wr4 wr5 wr6 wr7]
+ "wldrd wr12, [%[pixels]] \n\t"
+ "add r12, r12, #1 \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "tmcr wcgr2, r12 \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "cmp r12, #8 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr2, wr12, wr13 \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "wunpckelub wr0, wr2 \n\t"
+ "wunpckehub wr1, wr2 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "waddhus wr0, wr0, wr8 \n\t"
+ "waddhus wr1, wr1, wr9 \n\t"
+
+ "1: \n\t"
+ // [wr0 wr1 wr2 wr3]
+ // [wr4 wr5 wr6 wr7] <= *
+ "wldrd wr12, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "walignr1 wr6, wr12, wr13 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "wunpckelub wr4, wr6 \n\t"
+ "wunpckehub wr5, wr6 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "waddhus wr4, wr4, wr8 \n\t"
+ "waddhus wr5, wr5, wr9 \n\t"
+ "waddhus wr8, wr0, wr4 \n\t"
+ "waddhus wr9, wr1, wr5 \n\t"
+ "waddhus wr8, wr8, wr15 \n\t"
+ "waddhus wr9, wr9, wr15 \n\t"
+ "wldrd wr12, [%[block]] \n\t"
+ "wsrlhg wr8, wr8, wcgr0 \n\t"
+ "wsrlhg wr9, wr9, wcgr0 \n\t"
+ "wpackhus wr8, wr8, wr9 \n\t"
+ WAVG2B" wr8, wr8, wr12 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "wldrd wr12, [%[pixels]] \n\t"
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+
+ // [wr0 wr1 wr2 wr3] <= *
+ // [wr4 wr5 wr6 wr7]
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "walignr1 wr2, wr12, wr13 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "wunpckelub wr0, wr2 \n\t"
+ "wunpckehub wr1, wr2 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "waddhus wr0, wr0, wr8 \n\t"
+ "waddhus wr1, wr1, wr9 \n\t"
+ "waddhus wr8, wr0, wr4 \n\t"
+ "waddhus wr9, wr1, wr5 \n\t"
+ "waddhus wr8, wr8, wr15 \n\t"
+ "waddhus wr9, wr9, wr15 \n\t"
+ "wldrd wr12, [%[block]] \n\t"
+ "wsrlhg wr8, wr8, wcgr0 \n\t"
+ "wsrlhg wr9, wr9, wcgr0 \n\t"
+ "wpackhus wr8, wr8, wr9 \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ WAVG2B" wr8, wr8, wr12 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block)
+ : [line_size]"r"(line_size)
+ : "r12", "memory");
+}
+
+void DEF(avg, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+ // [wr0 wr1 wr2 wr3] for previous line
+ // [wr4 wr5 wr6 wr7] for current line
+ SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
+ __asm__ __volatile__(
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "pld [%[pixels]] \n\t"
+ "mov r12, #2 \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "tmcr wcgr0, r12 \n\t" /* for shift value */
+ /* alignment */
+ "and r12, %[pixels], #7 \n\t"
+ "bic %[pixels], %[pixels], #7 \n\t"
+ "tmcr wcgr1, r12 \n\t"
+ "add r12, r12, #1 \n\t"
+ "tmcr wcgr2, r12 \n\t"
+
+ // [wr0 wr1 wr2 wr3] <= *
+ // [wr4 wr5 wr6 wr7]
+ "wldrd wr12, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "wldrd wr14, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "pld [%[pixels]] \n\t"
+ "walignr1 wr2, wr12, wr13 \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr3, wr13, wr14 \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "wmoveq wr11, wr14 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "walignr2ne wr11, wr13, wr14 \n\t"
+ "wunpckelub wr0, wr2 \n\t"
+ "wunpckehub wr1, wr2 \n\t"
+ "wunpckelub wr2, wr3 \n\t"
+ "wunpckehub wr3, wr3 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "wunpckelub wr10, wr11 \n\t"
+ "wunpckehub wr11, wr11 \n\t"
+ "waddhus wr0, wr0, wr8 \n\t"
+ "waddhus wr1, wr1, wr9 \n\t"
+ "waddhus wr2, wr2, wr10 \n\t"
+ "waddhus wr3, wr3, wr11 \n\t"
+
+ "1: \n\t"
+ // [wr0 wr1 wr2 wr3]
+ // [wr4 wr5 wr6 wr7] <= *
+ "wldrd wr12, [%[pixels]] \n\t"
+ "cmp r12, #8 \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "wldrd wr14, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "walignr1 wr6, wr12, wr13 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr7, wr13, wr14 \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "wmoveq wr11, wr14 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "walignr2ne wr11, wr13, wr14 \n\t"
+ "wunpckelub wr4, wr6 \n\t"
+ "wunpckehub wr5, wr6 \n\t"
+ "wunpckelub wr6, wr7 \n\t"
+ "wunpckehub wr7, wr7 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "wunpckelub wr10, wr11 \n\t"
+ "wunpckehub wr11, wr11 \n\t"
+ "waddhus wr4, wr4, wr8 \n\t"
+ "waddhus wr5, wr5, wr9 \n\t"
+ "waddhus wr6, wr6, wr10 \n\t"
+ "waddhus wr7, wr7, wr11 \n\t"
+ "waddhus wr8, wr0, wr4 \n\t"
+ "waddhus wr9, wr1, wr5 \n\t"
+ "waddhus wr10, wr2, wr6 \n\t"
+ "waddhus wr11, wr3, wr7 \n\t"
+ "waddhus wr8, wr8, wr15 \n\t"
+ "waddhus wr9, wr9, wr15 \n\t"
+ "waddhus wr10, wr10, wr15 \n\t"
+ "waddhus wr11, wr11, wr15 \n\t"
+ "wsrlhg wr8, wr8, wcgr0 \n\t"
+ "wsrlhg wr9, wr9, wcgr0 \n\t"
+ "wldrd wr12, [%[block]] \n\t"
+ "wldrd wr13, [%[block], #8] \n\t"
+ "wsrlhg wr10, wr10, wcgr0 \n\t"
+ "wsrlhg wr11, wr11, wcgr0 \n\t"
+ "wpackhus wr8, wr8, wr9 \n\t"
+ "wpackhus wr9, wr10, wr11 \n\t"
+ WAVG2B" wr8, wr8, wr12 \n\t"
+ WAVG2B" wr9, wr9, wr13 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+
+ // [wr0 wr1 wr2 wr3] <= *
+ // [wr4 wr5 wr6 wr7]
+ "wldrd wr12, [%[pixels]] \n\t"
+ "pld [%[block]] \n\t"
+ "wldrd wr13, [%[pixels], #8] \n\t"
+ "pld [%[block], #32] \n\t"
+ "wldrd wr14, [%[pixels], #16] \n\t"
+ "add %[pixels], %[pixels], %[line_size] \n\t"
+ "walignr1 wr2, wr12, wr13 \n\t"
+ "pld [%[pixels]] \n\t"
+ "pld [%[pixels], #32] \n\t"
+ "walignr1 wr3, wr13, wr14 \n\t"
+ "wmoveq wr10, wr13 \n\t"
+ "wmoveq wr11, wr14 \n\t"
+ "walignr2ne wr10, wr12, wr13 \n\t"
+ "walignr2ne wr11, wr13, wr14 \n\t"
+ "wunpckelub wr0, wr2 \n\t"
+ "wunpckehub wr1, wr2 \n\t"
+ "wunpckelub wr2, wr3 \n\t"
+ "wunpckehub wr3, wr3 \n\t"
+ "wunpckelub wr8, wr10 \n\t"
+ "wunpckehub wr9, wr10 \n\t"
+ "wunpckelub wr10, wr11 \n\t"
+ "wunpckehub wr11, wr11 \n\t"
+ "waddhus wr0, wr0, wr8 \n\t"
+ "waddhus wr1, wr1, wr9 \n\t"
+ "waddhus wr2, wr2, wr10 \n\t"
+ "waddhus wr3, wr3, wr11 \n\t"
+ "waddhus wr8, wr0, wr4 \n\t"
+ "waddhus wr9, wr1, wr5 \n\t"
+ "waddhus wr10, wr2, wr6 \n\t"
+ "waddhus wr11, wr3, wr7 \n\t"
+ "waddhus wr8, wr8, wr15 \n\t"
+ "waddhus wr9, wr9, wr15 \n\t"
+ "waddhus wr10, wr10, wr15 \n\t"
+ "waddhus wr11, wr11, wr15 \n\t"
+ "wsrlhg wr8, wr8, wcgr0 \n\t"
+ "wsrlhg wr9, wr9, wcgr0 \n\t"
+ "wldrd wr12, [%[block]] \n\t"
+ "wldrd wr13, [%[block], #8] \n\t"
+ "wsrlhg wr10, wr10, wcgr0 \n\t"
+ "wsrlhg wr11, wr11, wcgr0 \n\t"
+ "wpackhus wr8, wr8, wr9 \n\t"
+ "wpackhus wr9, wr10, wr11 \n\t"
+ WAVG2B" wr8, wr8, wr12 \n\t"
+ WAVG2B" wr9, wr9, wr13 \n\t"
+ "wstrd wr8, [%[block]] \n\t"
+ "wstrd wr9, [%[block], #8] \n\t"
+ "add %[block], %[block], %[line_size] \n\t"
+ "subs %[h], %[h], #2 \n\t"
+ "pld [%[block]] \n\t"
+ "pld [%[block], #32] \n\t"
+ "bne 1b \n\t"
+ : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block)
+ : [line_size]"r"(line_size)
+ : "r12", "memory");
+}
diff --git a/src/libffmpeg/libavcodec/armv4l/mathops.h b/src/libffmpeg/libavcodec/armv4l/mathops.h
new file mode 100644
index 000000000..7ddd0ec6e
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/mathops.h
@@ -0,0 +1,49 @@
+/*
+ * simple math operations
+ * Copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at> et al
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifdef FRAC_BITS
+# define MULL(a, b) \
+ ({ int lo, hi;\
+ asm("smull %0, %1, %2, %3 \n\t"\
+ "mov %0, %0, lsr %4\n\t"\
+ "add %1, %0, %1, lsl %5\n\t"\
+ : "=&r"(lo), "=&r"(hi)\
+ : "r"(b), "r"(a), "i"(FRAC_BITS), "i"(32-FRAC_BITS));\
+ hi; })
+#endif
+
+#define MULH(a, b) \
+ ({ int lo, hi;\
+ asm ("smull %0, %1, %2, %3" : "=&r"(lo), "=&r"(hi) : "r"(b), "r"(a));\
+ hi; })
+
+#if defined(HAVE_ARMV5TE)
+
+/* signed 16x16 -> 32 multiply add accumulate */
+# define MAC16(rt, ra, rb) \
+ asm ("smlabb %0, %2, %3, %0" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
+/* signed 16x16 -> 32 multiply */
+# define MUL16(ra, rb) \
+ ({ int __rt; \
+ asm ("smulbb %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); \
+ __rt; })
+
+#endif
diff --git a/src/libffmpeg/libavcodec/armv4l/mpegvideo_armv5te.c b/src/libffmpeg/libavcodec/armv4l/mpegvideo_armv5te.c
new file mode 100644
index 000000000..a8d09b8ce
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/mpegvideo_armv5te.c
@@ -0,0 +1,213 @@
+/*
+ * Optimization of some functions from mpegvideo.c for armv5te
+ * Copyright (c) 2007 Siarhei Siamashka <ssvb@users.sourceforge.net>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/*
+ * Some useful links for those who may be interested in optimizing code for ARM.
+ * ARM Architecture Reference Manual: http://www.arm.com/community/academy/resources.html
+ * Instructions timings and optimization guide for ARM9E: http://www.arm.com/pdfs/DDI0222B_9EJS_r1p2.pdf
+ */
+
+#include "../dsputil.h"
+#include "../mpegvideo.h"
+#include "../avcodec.h"
+
+
+#ifdef ENABLE_ARM_TESTS
+/**
+ * h263 dequantizer supplementary function, it is performance critical and needs to
+ * have optimized implementations for each architecture. Is also used as a reference
+ * implementation in regression tests
+ */
+static inline void dct_unquantize_h263_helper_c(DCTELEM *block, int qmul, int qadd, int count)
+{
+ int i, level;
+ for (i = 0; i < count; i++) {
+ level = block[i];
+ if (level) {
+ if (level < 0) {
+ level = level * qmul - qadd;
+ } else {
+ level = level * qmul + qadd;
+ }
+ block[i] = level;
+ }
+ }
+}
+#endif
+
+/* GCC 3.1 or higher is required to support symbolic names in assembly code */
+#if (__GNUC__ > 3) || ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 1))
+
+/**
+ * Special optimized version of dct_unquantize_h263_helper_c, it requires the block
+ * to be at least 8 bytes aligned, and may process more elements than requested.
+ * But it is guaranteed to never process more than 64 elements provided that
+ * xxcount argument is <= 64, so it is safe. This macro is optimized for a common
+ * distribution of values for nCoeffs (they are mostly multiple of 8 plus one or
+ * two extra elements). So this macro processes data as 8 elements per loop iteration
+ * and contains optional 2 elements processing in the end.
+ *
+ * Inner loop should take 6 cycles per element on arm926ej-s (Nokia 770)
+ */
+#define dct_unquantize_h263_special_helper_armv5te(xxblock, xxqmul, xxqadd, xxcount) \
+({ DCTELEM *xblock = xxblock; \
+ int xqmul = xxqmul, xqadd = xxqadd, xcount = xxcount, xtmp; \
+ int xdata1, xdata2; \
+__asm__ __volatile__( \
+ "subs %[count], #2 \n\t" \
+ "ble 2f \n\t" \
+ "ldrd r4, [%[block], #0] \n\t" \
+ "1: \n\t" \
+ "ldrd r6, [%[block], #8] \n\t" \
+\
+ "rsbs %[data1], %[zero], r4, asr #16 \n\t" \
+ "addgt %[data1], %[qadd], #0 \n\t" \
+ "rsblt %[data1], %[qadd], #0 \n\t" \
+ "smlatbne %[data1], r4, %[qmul], %[data1] \n\t" \
+\
+ "rsbs %[data2], %[zero], r5, asr #16 \n\t" \
+ "addgt %[data2], %[qadd], #0 \n\t" \
+ "rsblt %[data2], %[qadd], #0 \n\t" \
+ "smlatbne %[data2], r5, %[qmul], %[data2] \n\t" \
+\
+ "rsbs %[tmp], %[zero], r4, asl #16 \n\t" \
+ "addgt %[tmp], %[qadd], #0 \n\t" \
+ "rsblt %[tmp], %[qadd], #0 \n\t" \
+ "smlabbne r4, r4, %[qmul], %[tmp] \n\t" \
+\
+ "rsbs %[tmp], %[zero], r5, asl #16 \n\t" \
+ "addgt %[tmp], %[qadd], #0 \n\t" \
+ "rsblt %[tmp], %[qadd], #0 \n\t" \
+ "smlabbne r5, r5, %[qmul], %[tmp] \n\t" \
+\
+ "strh r4, [%[block]], #2 \n\t" \
+ "strh %[data1], [%[block]], #2 \n\t" \
+ "strh r5, [%[block]], #2 \n\t" \
+ "strh %[data2], [%[block]], #2 \n\t" \
+\
+ "rsbs %[data1], %[zero], r6, asr #16 \n\t" \
+ "addgt %[data1], %[qadd], #0 \n\t" \
+ "rsblt %[data1], %[qadd], #0 \n\t" \
+ "smlatbne %[data1], r6, %[qmul], %[data1] \n\t" \
+\
+ "rsbs %[data2], %[zero], r7, asr #16 \n\t" \
+ "addgt %[data2], %[qadd], #0 \n\t" \
+ "rsblt %[data2], %[qadd], #0 \n\t" \
+ "smlatbne %[data2], r7, %[qmul], %[data2] \n\t" \
+\
+ "rsbs %[tmp], %[zero], r6, asl #16 \n\t" \
+ "addgt %[tmp], %[qadd], #0 \n\t" \
+ "rsblt %[tmp], %[qadd], #0 \n\t" \
+ "smlabbne r6, r6, %[qmul], %[tmp] \n\t" \
+\
+ "rsbs %[tmp], %[zero], r7, asl #16 \n\t" \
+ "addgt %[tmp], %[qadd], #0 \n\t" \
+ "rsblt %[tmp], %[qadd], #0 \n\t" \
+ "smlabbne r7, r7, %[qmul], %[tmp] \n\t" \
+\
+ "strh r6, [%[block]], #2 \n\t" \
+ "strh %[data1], [%[block]], #2 \n\t" \
+ "strh r7, [%[block]], #2 \n\t" \
+ "strh %[data2], [%[block]], #2 \n\t" \
+\
+ "subs %[count], #8 \n\t" \
+ "ldrgtd r4, [%[block], #0] \n\t" /* load data early to avoid load/use pipeline stall */ \
+ "bgt 1b \n\t" \
+\
+ "adds %[count], #2 \n\t" \
+ "ble 3f \n\t" \
+ "2: \n\t" \
+ "ldrsh %[data1], [%[block], #0] \n\t" \
+ "ldrsh %[data2], [%[block], #2] \n\t" \
+ "mov %[tmp], %[qadd] \n\t" \
+ "cmp %[data1], #0 \n\t" \
+ "rsblt %[tmp], %[qadd], #0 \n\t" \
+ "smlabbne %[data1], %[data1], %[qmul], %[tmp] \n\t" \
+ "mov %[tmp], %[qadd] \n\t" \
+ "cmp %[data2], #0 \n\t" \
+ "rsblt %[tmp], %[qadd], #0 \n\t" \
+ "smlabbne %[data2], %[data2], %[qmul], %[tmp] \n\t" \
+ "strh %[data1], [%[block]], #2 \n\t" \
+ "strh %[data2], [%[block]], #2 \n\t" \
+ "3: \n\t" \
+ : [block] "+&r" (xblock), [count] "+&r" (xcount), [tmp] "=&r" (xtmp), \
+ [data1] "=&r" (xdata1), [data2] "=&r" (xdata2) \
+ : [qmul] "r" (xqmul), [qadd] "r" (xqadd), [zero] "r" (0) \
+ : "r4", "r5", "r6", "r7", "cc", "memory" \
+); \
+})
+
+static void dct_unquantize_h263_intra_armv5te(MpegEncContext *s,
+ DCTELEM *block, int n, int qscale)
+{
+ int i, level, qmul, qadd;
+ int nCoeffs;
+
+ assert(s->block_last_index[n]>=0);
+
+ qmul = qscale << 1;
+
+ if (!s->h263_aic) {
+ if (n < 4)
+ level = block[0] * s->y_dc_scale;
+ else
+ level = block[0] * s->c_dc_scale;
+ qadd = (qscale - 1) | 1;
+ }else{
+ qadd = 0;
+ level = block[0];
+ }
+ if(s->ac_pred)
+ nCoeffs=63;
+ else
+ nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+
+ dct_unquantize_h263_special_helper_armv5te(block, qmul, qadd, nCoeffs + 1);
+ block[0] = level;
+}
+
+static void dct_unquantize_h263_inter_armv5te(MpegEncContext *s,
+ DCTELEM *block, int n, int qscale)
+{
+ int i, level, qmul, qadd;
+ int nCoeffs;
+
+ assert(s->block_last_index[n]>=0);
+
+ qadd = (qscale - 1) | 1;
+ qmul = qscale << 1;
+
+ nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+
+ dct_unquantize_h263_special_helper_armv5te(block, qmul, qadd, nCoeffs + 1);
+}
+
+#define HAVE_DCT_UNQUANTIZE_H263_ARMV5TE_OPTIMIZED
+
+#endif
+
+void MPV_common_init_armv5te(MpegEncContext *s)
+{
+#ifdef HAVE_DCT_UNQUANTIZE_H263_ARMV5TE_OPTIMIZED
+ s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_armv5te;
+ s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_armv5te;
+#endif
+}
diff --git a/src/libffmpeg/libavcodec/armv4l/mpegvideo_iwmmxt.c b/src/libffmpeg/libavcodec/armv4l/mpegvideo_iwmmxt.c
new file mode 100644
index 000000000..1336ac5f8
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/mpegvideo_iwmmxt.c
@@ -0,0 +1,119 @@
+/*
+ * copyright (c) 2004 AGAWA Koji
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "../dsputil.h"
+#include "../mpegvideo.h"
+#include "../avcodec.h"
+
+static void dct_unquantize_h263_intra_iwmmxt(MpegEncContext *s,
+ DCTELEM *block, int n, int qscale)
+{
+ int level, qmul, qadd;
+ int nCoeffs;
+ DCTELEM *block_orig = block;
+
+ assert(s->block_last_index[n]>=0);
+
+ qmul = qscale << 1;
+
+ if (!s->h263_aic) {
+ if (n < 4)
+ level = block[0] * s->y_dc_scale;
+ else
+ level = block[0] * s->c_dc_scale;
+ qadd = (qscale - 1) | 1;
+ }else{
+ qadd = 0;
+ level = block[0];
+ }
+ if(s->ac_pred)
+ nCoeffs=63;
+ else
+ nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+
+ __asm__ __volatile__ (
+/* "movd %1, %%mm6 \n\t" //qmul */
+/* "packssdw %%mm6, %%mm6 \n\t" */
+/* "packssdw %%mm6, %%mm6 \n\t" */
+ "tbcsth wr6, %[qmul] \n\t"
+/* "movd %2, %%mm5 \n\t" //qadd */
+/* "packssdw %%mm5, %%mm5 \n\t" */
+/* "packssdw %%mm5, %%mm5 \n\t" */
+ "tbcsth wr5, %[qadd] \n\t"
+ "wzero wr7 \n\t" /* "pxor %%mm7, %%mm7 \n\t" */
+ "wzero wr4 \n\t" /* "pxor %%mm4, %%mm4 \n\t" */
+ "wsubh wr7, wr5, wr7 \n\t" /* "psubw %%mm5, %%mm7 \n\t" */
+ "1: \n\t"
+ "wldrd wr2, [%[block]] \n\t" /* "movq (%0, %3), %%mm0 \n\t" */
+ "wldrd wr3, [%[block], #8] \n\t" /* "movq 8(%0, %3), %%mm1 \n\t" */
+ "wmulsl wr0, wr6, wr2 \n\t" /* "pmullw %%mm6, %%mm0 \n\t" */
+ "wmulsl wr1, wr6, wr3 \n\t" /* "pmullw %%mm6, %%mm1 \n\t" */
+/* "movq (%0, %3), %%mm2 \n\t" */
+/* "movq 8(%0, %3), %%mm3 \n\t" */
+ "wcmpgtsh wr2, wr4, wr2 \n\t" /* "pcmpgtw %%mm4, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 */
+ "wcmpgtsh wr3, wr4, wr2 \n\t" /* "pcmpgtw %%mm4, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 */
+ "wxor wr0, wr2, wr0 \n\t" /* "pxor %%mm2, %%mm0 \n\t" */
+ "wxor wr1, wr3, wr1 \n\t" /* "pxor %%mm3, %%mm1 \n\t" */
+ "waddh wr0, wr7, wr0 \n\t" /* "paddw %%mm7, %%mm0 \n\t" */
+ "waddh wr1, wr7, wr1 \n\t" /* "paddw %%mm7, %%mm1 \n\t" */
+ "wxor wr2, wr0, wr2 \n\t" /* "pxor %%mm0, %%mm2 \n\t" */
+ "wxor wr3, wr1, wr3 \n\t" /* "pxor %%mm1, %%mm3 \n\t" */
+ "wcmpeqh wr0, wr7, wr0 \n\t" /* "pcmpeqw %%mm7, %%mm0 \n\t" // block[i] == 0 ? -1 : 0 */
+ "wcmpeqh wr1, wr7, wr1 \n\t" /* "pcmpeqw %%mm7, %%mm1 \n\t" // block[i] == 0 ? -1 : 0 */
+ "wandn wr0, wr2, wr0 \n\t" /* "pandn %%mm2, %%mm0 \n\t" */
+ "wandn wr1, wr3, wr1 \n\t" /* "pandn %%mm3, %%mm1 \n\t" */
+ "wstrd wr0, [%[block]] \n\t" /* "movq %%mm0, (%0, %3) \n\t" */
+ "wstrd wr1, [%[block], #8] \n\t" /* "movq %%mm1, 8(%0, %3) \n\t" */
+ "add %[block], %[block], #16 \n\t" /* "addl $16, %3 \n\t" */
+ "subs %[i], %[i], #1 \n\t"
+ "bne 1b \n\t" /* "jng 1b \n\t" */
+ :[block]"+r"(block)
+ :[i]"r"((nCoeffs + 8) / 8), [qmul]"r"(qmul), [qadd]"r"(qadd)
+ :"memory");
+
+ block_orig[0] = level;
+}
+
+#if 0
+static void dct_unquantize_h263_inter_iwmmxt(MpegEncContext *s,
+ DCTELEM *block, int n, int qscale)
+{
+ int nCoeffs;
+
+ assert(s->block_last_index[n]>=0);
+
+ if(s->ac_pred)
+ nCoeffs=63;
+ else
+ nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+
+ ippiQuantInvInter_Compact_H263_16s_I(block, nCoeffs+1, qscale);
+}
+#endif
+
+void MPV_common_init_iwmmxt(MpegEncContext *s)
+{
+ if (!(mm_flags & MM_IWMMXT)) return;
+
+ s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_iwmmxt;
+#if 0
+ s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_iwmmxt;
+#endif
+}
diff --git a/src/libffmpeg/libavcodec/armv4l/simple_idct_armv5te.S b/src/libffmpeg/libavcodec/armv4l/simple_idct_armv5te.S
new file mode 100644
index 000000000..28bee0643
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/simple_idct_armv5te.S
@@ -0,0 +1,718 @@
+/*
+ * Simple IDCT
+ *
+ * Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
+ * Copyright (c) 2006 Mans Rullgard <mru@inprovide.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#define W1 22725 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
+#define W2 21407 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
+#define W3 19266 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
+#define W4 16383 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
+#define W5 12873 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
+#define W6 8867 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
+#define W7 4520 /* cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5 */
+#define ROW_SHIFT 11
+#define COL_SHIFT 20
+
+#define W13 (W1 | (W3 << 16))
+#define W26 (W2 | (W6 << 16))
+#define W57 (W5 | (W7 << 16))
+
+ .text
+ .align
+w13: .long W13
+w26: .long W26
+w57: .long W57
+
+ .align
+ .func idct_row_armv5te
+idct_row_armv5te:
+ str lr, [sp, #-4]!
+
+ ldrd v1, [a1, #8]
+ ldrd a3, [a1] /* a3 = row[1:0], a4 = row[3:2] */
+ orrs v1, v1, v2
+ cmpeq v1, a4
+ cmpeq v1, a3, lsr #16
+ beq row_dc_only
+
+ mov v1, #(1<<(ROW_SHIFT-1))
+ mov ip, #16384
+ sub ip, ip, #1 /* ip = W4 */
+ smlabb v1, ip, a3, v1 /* v1 = W4*row[0]+(1<<(RS-1)) */
+ ldr ip, [pc, #(w26-.-8)] /* ip = W2 | (W6 << 16) */
+ smultb a2, ip, a4
+ smulbb lr, ip, a4
+ add v2, v1, a2
+ sub v3, v1, a2
+ sub v4, v1, lr
+ add v1, v1, lr
+
+ ldr ip, [pc, #(w13-.-8)] /* ip = W1 | (W3 << 16) */
+ ldr lr, [pc, #(w57-.-8)] /* lr = W5 | (W7 << 16) */
+ smulbt v5, ip, a3
+ smultt v6, lr, a4
+ smlatt v5, ip, a4, v5
+ smultt a2, ip, a3
+ smulbt v7, lr, a3
+ sub v6, v6, a2
+ smulbt a2, ip, a4
+ smultt fp, lr, a3
+ sub v7, v7, a2
+ smulbt a2, lr, a4
+ ldrd a3, [a1, #8] /* a3=row[5:4] a4=row[7:6] */
+ sub fp, fp, a2
+
+ orrs a2, a3, a4
+ beq 1f
+
+ smlabt v5, lr, a3, v5
+ smlabt v6, ip, a3, v6
+ smlatt v5, lr, a4, v5
+ smlabt v6, lr, a4, v6
+ smlatt v7, lr, a3, v7
+ smlatt fp, ip, a3, fp
+ smulbt a2, ip, a4
+ smlatt v7, ip, a4, v7
+ sub fp, fp, a2
+
+ ldr ip, [pc, #(w26-.-8)] /* ip = W2 | (W6 << 16) */
+ mov a2, #16384
+ sub a2, a2, #1 /* a2 = W4 */
+ smulbb a2, a2, a3 /* a2 = W4*row[4] */
+ smultb lr, ip, a4 /* lr = W6*row[6] */
+ add v1, v1, a2 /* v1 += W4*row[4] */
+ add v1, v1, lr /* v1 += W6*row[6] */
+ add v4, v4, a2 /* v4 += W4*row[4] */
+ sub v4, v4, lr /* v4 -= W6*row[6] */
+ smulbb lr, ip, a4 /* lr = W2*row[6] */
+ sub v2, v2, a2 /* v2 -= W4*row[4] */
+ sub v2, v2, lr /* v2 -= W2*row[6] */
+ sub v3, v3, a2 /* v3 -= W4*row[4] */
+ add v3, v3, lr /* v3 += W2*row[6] */
+
+1: add a2, v1, v5
+ mov a3, a2, lsr #11
+ bic a3, a3, #0x1f0000
+ sub a2, v2, v6
+ mov a2, a2, lsr #11
+ add a3, a3, a2, lsl #16
+ add a2, v3, v7
+ mov a4, a2, lsr #11
+ bic a4, a4, #0x1f0000
+ add a2, v4, fp
+ mov a2, a2, lsr #11
+ add a4, a4, a2, lsl #16
+ strd a3, [a1]
+
+ sub a2, v4, fp
+ mov a3, a2, lsr #11
+ bic a3, a3, #0x1f0000
+ sub a2, v3, v7
+ mov a2, a2, lsr #11
+ add a3, a3, a2, lsl #16
+ add a2, v2, v6
+ mov a4, a2, lsr #11
+ bic a4, a4, #0x1f0000
+ sub a2, v1, v5
+ mov a2, a2, lsr #11
+ add a4, a4, a2, lsl #16
+ strd a3, [a1, #8]
+
+ ldr pc, [sp], #4
+
+row_dc_only:
+ orr a3, a3, a3, lsl #16
+ bic a3, a3, #0xe000
+ mov a3, a3, lsl #3
+ mov a4, a3
+ strd a3, [a1]
+ strd a3, [a1, #8]
+
+ ldr pc, [sp], #4
+ .endfunc
+
+ .macro idct_col
+ ldr a4, [a1] /* a4 = col[1:0] */
+ mov ip, #16384
+ sub ip, ip, #1 /* ip = W4 */
+#if 0
+ mov v1, #(1<<(COL_SHIFT-1))
+ smlabt v2, ip, a4, v1 /* v2 = W4*col[1] + (1<<(COL_SHIFT-1)) */
+ smlabb v1, ip, a4, v1 /* v1 = W4*col[0] + (1<<(COL_SHIFT-1)) */
+ ldr a4, [a1, #(16*4)]
+#else
+ mov v1, #((1<<(COL_SHIFT-1))/W4) /* this matches the C version */
+ add v2, v1, a4, asr #16
+ rsb v2, v2, v2, lsl #14
+ mov a4, a4, lsl #16
+ add v1, v1, a4, asr #16
+ ldr a4, [a1, #(16*4)]
+ rsb v1, v1, v1, lsl #14
+#endif
+
+ smulbb lr, ip, a4
+ smulbt a3, ip, a4
+ sub v3, v1, lr
+ sub v5, v1, lr
+ add v7, v1, lr
+ add v1, v1, lr
+ sub v4, v2, a3
+ sub v6, v2, a3
+ add fp, v2, a3
+ ldr ip, [pc, #(w26-.-8)]
+ ldr a4, [a1, #(16*2)]
+ add v2, v2, a3
+
+ smulbb lr, ip, a4
+ smultb a3, ip, a4
+ add v1, v1, lr
+ sub v7, v7, lr
+ add v3, v3, a3
+ sub v5, v5, a3
+ smulbt lr, ip, a4
+ smultt a3, ip, a4
+ add v2, v2, lr
+ sub fp, fp, lr
+ add v4, v4, a3
+ ldr a4, [a1, #(16*6)]
+ sub v6, v6, a3
+
+ smultb lr, ip, a4
+ smulbb a3, ip, a4
+ add v1, v1, lr
+ sub v7, v7, lr
+ sub v3, v3, a3
+ add v5, v5, a3
+ smultt lr, ip, a4
+ smulbt a3, ip, a4
+ add v2, v2, lr
+ sub fp, fp, lr
+ sub v4, v4, a3
+ add v6, v6, a3
+
+ stmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp}
+
+ ldr ip, [pc, #(w13-.-8)]
+ ldr a4, [a1, #(16*1)]
+ ldr lr, [pc, #(w57-.-8)]
+ smulbb v1, ip, a4
+ smultb v3, ip, a4
+ smulbb v5, lr, a4
+ smultb v7, lr, a4
+ smulbt v2, ip, a4
+ smultt v4, ip, a4
+ smulbt v6, lr, a4
+ smultt fp, lr, a4
+ rsb v4, v4, #0
+ ldr a4, [a1, #(16*3)]
+ rsb v3, v3, #0
+
+ smlatb v1, ip, a4, v1
+ smlatb v3, lr, a4, v3
+ smulbb a3, ip, a4
+ smulbb a2, lr, a4
+ sub v5, v5, a3
+ sub v7, v7, a2
+ smlatt v2, ip, a4, v2
+ smlatt v4, lr, a4, v4
+ smulbt a3, ip, a4
+ smulbt a2, lr, a4
+ sub v6, v6, a3
+ ldr a4, [a1, #(16*5)]
+ sub fp, fp, a2
+
+ smlabb v1, lr, a4, v1
+ smlabb v3, ip, a4, v3
+ smlatb v5, lr, a4, v5
+ smlatb v7, ip, a4, v7
+ smlabt v2, lr, a4, v2
+ smlabt v4, ip, a4, v4
+ smlatt v6, lr, a4, v6
+ ldr a3, [a1, #(16*7)]
+ smlatt fp, ip, a4, fp
+
+ smlatb v1, lr, a3, v1
+ smlabb v3, lr, a3, v3
+ smlatb v5, ip, a3, v5
+ smulbb a4, ip, a3
+ smlatt v2, lr, a3, v2
+ sub v7, v7, a4
+ smlabt v4, lr, a3, v4
+ smulbt a4, ip, a3
+ smlatt v6, ip, a3, v6
+ sub fp, fp, a4
+ .endm
+
+ .align
+ .func idct_col_armv5te
+idct_col_armv5te:
+ str lr, [sp, #-4]!
+
+ idct_col
+
+ ldmfd sp!, {a3, a4}
+ adds a2, a3, v1
+ mov a2, a2, lsr #20
+ orrmi a2, a2, #0xf000
+ add ip, a4, v2
+ mov ip, ip, asr #20
+ orr a2, a2, ip, lsl #16
+ str a2, [a1]
+ subs a3, a3, v1
+ mov a2, a3, lsr #20
+ orrmi a2, a2, #0xf000
+ sub a4, a4, v2
+ mov a4, a4, asr #20
+ orr a2, a2, a4, lsl #16
+ ldmfd sp!, {a3, a4}
+ str a2, [a1, #(16*7)]
+
+ subs a2, a3, v3
+ mov a2, a2, lsr #20
+ orrmi a2, a2, #0xf000
+ sub ip, a4, v4
+ mov ip, ip, asr #20
+ orr a2, a2, ip, lsl #16
+ str a2, [a1, #(16*1)]
+ adds a3, a3, v3
+ mov a2, a3, lsr #20
+ orrmi a2, a2, #0xf000
+ add a4, a4, v4
+ mov a4, a4, asr #20
+ orr a2, a2, a4, lsl #16
+ ldmfd sp!, {a3, a4}
+ str a2, [a1, #(16*6)]
+
+ adds a2, a3, v5
+ mov a2, a2, lsr #20
+ orrmi a2, a2, #0xf000
+ add ip, a4, v6
+ mov ip, ip, asr #20
+ orr a2, a2, ip, lsl #16
+ str a2, [a1, #(16*2)]
+ subs a3, a3, v5
+ mov a2, a3, lsr #20
+ orrmi a2, a2, #0xf000
+ sub a4, a4, v6
+ mov a4, a4, asr #20
+ orr a2, a2, a4, lsl #16
+ ldmfd sp!, {a3, a4}
+ str a2, [a1, #(16*5)]
+
+ adds a2, a3, v7
+ mov a2, a2, lsr #20
+ orrmi a2, a2, #0xf000
+ add ip, a4, fp
+ mov ip, ip, asr #20
+ orr a2, a2, ip, lsl #16
+ str a2, [a1, #(16*3)]
+ subs a3, a3, v7
+ mov a2, a3, lsr #20
+ orrmi a2, a2, #0xf000
+ sub a4, a4, fp
+ mov a4, a4, asr #20
+ orr a2, a2, a4, lsl #16
+ str a2, [a1, #(16*4)]
+
+ ldr pc, [sp], #4
+ .endfunc
+
+ .align
+ .func idct_col_put_armv5te
+idct_col_put_armv5te:
+ str lr, [sp, #-4]!
+
+ idct_col
+
+ ldmfd sp!, {a3, a4}
+ ldr lr, [sp, #32]
+ add a2, a3, v1
+ movs a2, a2, asr #20
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ add ip, a4, v2
+ movs ip, ip, asr #20
+ movmi ip, #0
+ cmp ip, #255
+ movgt ip, #255
+ orr a2, a2, ip, lsl #8
+ sub a3, a3, v1
+ movs a3, a3, asr #20
+ movmi a3, #0
+ cmp a3, #255
+ movgt a3, #255
+ sub a4, a4, v2
+ movs a4, a4, asr #20
+ movmi a4, #0
+ cmp a4, #255
+ ldr v1, [sp, #28]
+ movgt a4, #255
+ strh a2, [v1]
+ add a2, v1, #2
+ str a2, [sp, #28]
+ orr a2, a3, a4, lsl #8
+ rsb v2, lr, lr, lsl #3
+ ldmfd sp!, {a3, a4}
+ strh a2, [v2, v1]!
+
+ sub a2, a3, v3
+ movs a2, a2, asr #20
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ sub ip, a4, v4
+ movs ip, ip, asr #20
+ movmi ip, #0
+ cmp ip, #255
+ movgt ip, #255
+ orr a2, a2, ip, lsl #8
+ strh a2, [v1, lr]!
+ add a3, a3, v3
+ movs a2, a3, asr #20
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ add a4, a4, v4
+ movs a4, a4, asr #20
+ movmi a4, #0
+ cmp a4, #255
+ movgt a4, #255
+ orr a2, a2, a4, lsl #8
+ ldmfd sp!, {a3, a4}
+ strh a2, [v2, -lr]!
+
+ add a2, a3, v5
+ movs a2, a2, asr #20
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ add ip, a4, v6
+ movs ip, ip, asr #20
+ movmi ip, #0
+ cmp ip, #255
+ movgt ip, #255
+ orr a2, a2, ip, lsl #8
+ strh a2, [v1, lr]!
+ sub a3, a3, v5
+ movs a2, a3, asr #20
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ sub a4, a4, v6
+ movs a4, a4, asr #20
+ movmi a4, #0
+ cmp a4, #255
+ movgt a4, #255
+ orr a2, a2, a4, lsl #8
+ ldmfd sp!, {a3, a4}
+ strh a2, [v2, -lr]!
+
+ add a2, a3, v7
+ movs a2, a2, asr #20
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ add ip, a4, fp
+ movs ip, ip, asr #20
+ movmi ip, #0
+ cmp ip, #255
+ movgt ip, #255
+ orr a2, a2, ip, lsl #8
+ strh a2, [v1, lr]
+ sub a3, a3, v7
+ movs a2, a3, asr #20
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ sub a4, a4, fp
+ movs a4, a4, asr #20
+ movmi a4, #0
+ cmp a4, #255
+ movgt a4, #255
+ orr a2, a2, a4, lsl #8
+ strh a2, [v2, -lr]
+
+ ldr pc, [sp], #4
+ .endfunc
+
+ .align
+ .func idct_col_add_armv5te
+idct_col_add_armv5te:
+ str lr, [sp, #-4]!
+
+ idct_col
+
+ ldr lr, [sp, #36]
+
+ ldmfd sp!, {a3, a4}
+ ldrh ip, [lr]
+ add a2, a3, v1
+ mov a2, a2, asr #20
+ sub a3, a3, v1
+ and v1, ip, #255
+ adds a2, a2, v1
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ add v1, a4, v2
+ mov v1, v1, asr #20
+ adds v1, v1, ip, lsr #8
+ movmi v1, #0
+ cmp v1, #255
+ movgt v1, #255
+ orr a2, a2, v1, lsl #8
+ ldr v1, [sp, #32]
+ sub a4, a4, v2
+ rsb v2, v1, v1, lsl #3
+ ldrh ip, [v2, lr]!
+ strh a2, [lr]
+ mov a3, a3, asr #20
+ and a2, ip, #255
+ adds a3, a3, a2
+ movmi a3, #0
+ cmp a3, #255
+ movgt a3, #255
+ mov a4, a4, asr #20
+ adds a4, a4, ip, lsr #8
+ movmi a4, #0
+ cmp a4, #255
+ movgt a4, #255
+ add a2, lr, #2
+ str a2, [sp, #28]
+ orr a2, a3, a4, lsl #8
+ strh a2, [v2]
+
+ ldmfd sp!, {a3, a4}
+ ldrh ip, [lr, v1]!
+ sub a2, a3, v3
+ mov a2, a2, asr #20
+ add a3, a3, v3
+ and v3, ip, #255
+ adds a2, a2, v3
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ sub v3, a4, v4
+ mov v3, v3, asr #20
+ adds v3, v3, ip, lsr #8
+ movmi v3, #0
+ cmp v3, #255
+ movgt v3, #255
+ orr a2, a2, v3, lsl #8
+ add a4, a4, v4
+ ldrh ip, [v2, -v1]!
+ strh a2, [lr]
+ mov a3, a3, asr #20
+ and a2, ip, #255
+ adds a3, a3, a2
+ movmi a3, #0
+ cmp a3, #255
+ movgt a3, #255
+ mov a4, a4, asr #20
+ adds a4, a4, ip, lsr #8
+ movmi a4, #0
+ cmp a4, #255
+ movgt a4, #255
+ orr a2, a3, a4, lsl #8
+ strh a2, [v2]
+
+ ldmfd sp!, {a3, a4}
+ ldrh ip, [lr, v1]!
+ add a2, a3, v5
+ mov a2, a2, asr #20
+ sub a3, a3, v5
+ and v3, ip, #255
+ adds a2, a2, v3
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ add v3, a4, v6
+ mov v3, v3, asr #20
+ adds v3, v3, ip, lsr #8
+ movmi v3, #0
+ cmp v3, #255
+ movgt v3, #255
+ orr a2, a2, v3, lsl #8
+ sub a4, a4, v6
+ ldrh ip, [v2, -v1]!
+ strh a2, [lr]
+ mov a3, a3, asr #20
+ and a2, ip, #255
+ adds a3, a3, a2
+ movmi a3, #0
+ cmp a3, #255
+ movgt a3, #255
+ mov a4, a4, asr #20
+ adds a4, a4, ip, lsr #8
+ movmi a4, #0
+ cmp a4, #255
+ movgt a4, #255
+ orr a2, a3, a4, lsl #8
+ strh a2, [v2]
+
+ ldmfd sp!, {a3, a4}
+ ldrh ip, [lr, v1]!
+ add a2, a3, v7
+ mov a2, a2, asr #20
+ sub a3, a3, v7
+ and v3, ip, #255
+ adds a2, a2, v3
+ movmi a2, #0
+ cmp a2, #255
+ movgt a2, #255
+ add v3, a4, fp
+ mov v3, v3, asr #20
+ adds v3, v3, ip, lsr #8
+ movmi v3, #0
+ cmp v3, #255
+ movgt v3, #255
+ orr a2, a2, v3, lsl #8
+ sub a4, a4, fp
+ ldrh ip, [v2, -v1]!
+ strh a2, [lr]
+ mov a3, a3, asr #20
+ and a2, ip, #255
+ adds a3, a3, a2
+ movmi a3, #0
+ cmp a3, #255
+ movgt a3, #255
+ mov a4, a4, asr #20
+ adds a4, a4, ip, lsr #8
+ movmi a4, #0
+ cmp a4, #255
+ movgt a4, #255
+ orr a2, a3, a4, lsl #8
+ strh a2, [v2]
+
+ ldr pc, [sp], #4
+ .endfunc
+
+ .align
+ .global simple_idct_armv5te
+ .func simple_idct_armv5te
+simple_idct_armv5te:
+ stmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp, lr}
+
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+
+ sub a1, a1, #(16*7)
+
+ bl idct_col_armv5te
+ add a1, a1, #4
+ bl idct_col_armv5te
+ add a1, a1, #4
+ bl idct_col_armv5te
+ add a1, a1, #4
+ bl idct_col_armv5te
+
+ ldmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp, pc}
+ .endfunc
+
+ .align
+ .global simple_idct_add_armv5te
+ .func simple_idct_add_armv5te
+simple_idct_add_armv5te:
+ stmfd sp!, {a1, a2, v1, v2, v3, v4, v5, v6, v7, fp, lr}
+
+ mov a1, a3
+
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+
+ sub a1, a1, #(16*7)
+
+ bl idct_col_add_armv5te
+ add a1, a1, #4
+ bl idct_col_add_armv5te
+ add a1, a1, #4
+ bl idct_col_add_armv5te
+ add a1, a1, #4
+ bl idct_col_add_armv5te
+
+ add sp, sp, #8
+ ldmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp, pc}
+ .endfunc
+
+ .align
+ .global simple_idct_put_armv5te
+ .func simple_idct_put_armv5te
+simple_idct_put_armv5te:
+ stmfd sp!, {a1, a2, v1, v2, v3, v4, v5, v6, v7, fp, lr}
+
+ mov a1, a3
+
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+ add a1, a1, #16
+ bl idct_row_armv5te
+
+ sub a1, a1, #(16*7)
+
+ bl idct_col_put_armv5te
+ add a1, a1, #4
+ bl idct_col_put_armv5te
+ add a1, a1, #4
+ bl idct_col_put_armv5te
+ add a1, a1, #4
+ bl idct_col_put_armv5te
+
+ add sp, sp, #8
+ ldmfd sp!, {v1, v2, v3, v4, v5, v6, v7, fp, pc}
+ .endfunc
generated by cgit v1.2.3 (git 2.25.1) at 2025-11-15 23:58:39 +0000