On Thu, 20 Apr 2017, Alexandra Hájková wrote:
From: Seppo Tomperi <[email protected]>
Signed-off-by: Alexandra Hájková <[email protected]>
---
libavcodec/arm/hevc_idct.S | 78 +++++++++++++++++++++++++++++++++++++++
libavcodec/arm/hevcdsp_init_arm.c | 15 ++++++--
2 files changed, 90 insertions(+), 3 deletions(-)
diff --git a/libavcodec/arm/hevc_idct.S b/libavcodec/arm/hevc_idct.S
index 156d476..f0008aa 100644
--- a/libavcodec/arm/hevc_idct.S
+++ b/libavcodec/arm/hevc_idct.S
@@ -1,5 +1,7 @@
/*
* ARM NEON optimised IDCT functions for HEVC decoding
+ *
+ * Copyright (c) 2014 Seppo Tomperi <[email protected]>
* Copyright (c) 2017 Alexandra Hájková
*
* This file is part of Libav.
@@ -28,6 +30,82 @@ const trans, align=4
.short 57, 43, 25, 9
endconst
+function ff_hevc_idct_4x4_dc_8_neon, export=1
+ ldrsh r1, [r0]
+ ldr r2, =0x20
+ add r1, #1
+ asr r1, #1
+ add r1, r2
+ asr r1, #6
+ vdup.16 q0, r1
+ vdup.16 q1, r1
+ vst1.16 {q0, q1}, [r0, :128]
+ bx lr
+endfunc
+
+function ff_hevc_idct_8x8_dc_8_neon, export=1
+ ldrsh r1, [r0]
+ ldr r2, =0x20
+ add r1, #1
+ asr r1, #1
+ add r1, r2
+ asr r1, #6
+ vdup.16 q8, r1
+ vdup.16 q9, r1
+ vmov.16 q10, q8
+ vmov.16 q11, q8
+ vmov.16 q12, q8
+ vmov.16 q13, q8
+ vmov.16 q14, q8
+ vmov.16 q15, q8
+ vstm r0, {q8-q15}
+ bx lr
+endfunc
+
+function ff_hevc_idct_16x16_dc_8_neon, export=1
+ ldrsh r1, [r0]
+ ldr r2, =0x20
+ add r1, #1
+ asr r1, #1
+ add r1, r2
+ asr r1, #6
+ vdup.16 q8, r1
+ vdup.16 q9, r1
+ vmov.16 q10, q8
+ vmov.16 q11, q8
+ vmov.16 q12, q8
+ vmov.16 q13, q8
+ vmov.16 q14, q8
+ vmov.16 q15, q8
+ vstm r0!, {q8-q15}
+ vstm r0!, {q8-q15}
+ vstm r0!, {q8-q15}
+ vstm r0, {q8-q15}
+ bx lr
+endfunc
+
+function ff_hevc_idct_32x32_dc_8_neon, export=1
+ ldrsh r1, [r0]
+ ldr r2, =0x20
+ add r1, #1
+ asr r1, #1
+ add r1, r2
+ asr r1, #6
+ mov r3, #16
+ vdup.16 q8, r1
+ vdup.16 q9, r1
+ vmov.16 q10, q8
+ vmov.16 q11, q8
+ vmov.16 q12, q8
+ vmov.16 q13, q8
+ vmov.16 q14, q8
+ vmov.16 q15, q8
+1: subs r3, #1
+ vstm r0!, {q8-q15}
+ bne 1b
+ bx lr
Misindented bx lr
+endfunc
+
.macro sum_sub out, in, c, op
.ifc \op, +
vmlal.s16 \out, \in, \c
diff --git a/libavcodec/arm/hevcdsp_init_arm.c
b/libavcodec/arm/hevcdsp_init_arm.c
index e61587f..10f60bc 100644
--- a/libavcodec/arm/hevcdsp_init_arm.c
+++ b/libavcodec/arm/hevcdsp_init_arm.c
@@ -26,8 +26,12 @@
#include "libavcodec/hevcdsp.h"
void ff_hevc_idct_4x4_8_neon(int16_t *coeffs, int col_limit);
+void ff_hevc_idct_4x4_dc_8_neon(int16_t *coeffs);
void ff_hevc_idct_8x8_8_neon(int16_t *coeffs, int col_limit);
+void ff_hevc_idct_8x8_dc_8_neon(int16_t *coeffs);
void ff_hevc_idct_16x16_8_neon(int16_t *coeffs, int col_limit);
+void ff_hevc_idct_16x16_dc_8_neon(int16_t *coeffs);
+void ff_hevc_idct_32x32_dc_8_neon(int16_t *coeffs);
void ff_hevc_idct_4x4_10_neon(int16_t *coeffs, int col_limit);
void ff_hevc_idct_8x8_10_neon(int16_t *coeffs, int col_limit);
void ff_hevc_idct_16x16_10_neon(int16_t *coeffs, int col_limit);
As Diego suggested and I said in the previous review, group these with the
idct_dc separately. You do this change in patch 2/2, but make it this way
in the first commit directly instead of changing it later.
@@ -38,9 +42,14 @@ av_cold void ff_hevc_dsp_init_arm(HEVCDSPContext *c, int
bit_depth)
if (have_neon(cpu_flags)) {
if (bit_depth == 8) {
- c->idct[0] = ff_hevc_idct_4x4_8_neon;
- c->idct[1] = ff_hevc_idct_8x8_8_neon;
- c->idct[2] = ff_hevc_idct_16x16_8_neon;
+ c->idct_dc[0] = ff_hevc_idct_4x4_dc_8_neon;
+ c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_neon;
+ c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_neon;
+ c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_neon;
+
+ c->idct[0] = ff_hevc_idct_4x4_8_neon;
+ c->idct[1] = ff_hevc_idct_8x8_8_neon;
+ c->idct[2] = ff_hevc_idct_16x16_8_neon;
I'm not sure if I feel that it's necessary to realign the idct lines here;
since they're grouped separately we could just have them aligned
separately as well. (I'm not sure if Diego agrees here or not.)
// Martin
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
