Jeff Muizelaar posted some changes to the idct/reconstruction c code.
This is the equivalent update for the arm assembly.
This shows a good boost on v6, and a minor boost on neon.
Here are some numbers for highway in qcif, 2641 frames:
HEAD neon: ~161 fps
new neon: ~162 fps
HEAD v6: ~102 fps
new v6: ~106 fps
The following functions have been updated for armv6 and neon:
vp8_dc_only_idct_add
vp8_dequant_idct_add
vp8_dequant_dc_idct_add
Conflicts:
vp8/decoder/arm/armv6/dequantdcidct_v6.asm
vp8/decoder/arm/armv6/dequantidct_v6.asm
Resolved by removing these files. When I rewrote the functions, I also
moved the files to dequant_dc_idct_v6.asm/dequant_idct_v6.asm
Change-Id: Ie3300df824d52474eca1a5134cf22d8b7809a5d4
--- /dev/null
+;
+; Copyright (c) 2010 The VP8 project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license and patent
+; grant that can be found in the LICENSE file in the root of the source
+; tree. All contributing project authors may be found in the AUTHORS
+; file in the root of the source tree.
+;
+
+ EXPORT |vp8_dc_only_idct_add_v6|
+
+ AREA |.text|, CODE, READONLY
+
+;void vp8_dc_only_idct_add_v6(short input_dc, unsigned char *pred_ptr,
+; unsigned char *dst_ptr, int pitch, int stride)
+; r0 input_dc
+; r1 pred_ptr
+; r2 dest_ptr
+; r3 pitch
+; sp stride
+
+|vp8_dc_only_idct_add_v6| PROC
+ stmdb sp!, {r4 - r7, lr}
+
+ add r0, r0, #4 ; input_dc += 4
+ ldr r12, c0x0000FFFF
+ ldr r4, [r1], r3
+ ldr r6, [r1], r3
+ and r0, r12, r0, asr #3 ; input_dc >> 3 + mask
+ ldr lr, [sp, #20]
+ orr r0, r0, r0, lsl #16 ; a1 | a1
+
+ uxtab16 r5, r0, r4 ; a1+2 | a1+0
+ uxtab16 r4, r0, r4, ror #8 ; a1+3 | a1+1
+ uxtab16 r7, r0, r6
+ uxtab16 r6, r0, r6, ror #8
+ usat16 r5, #8, r5
+ usat16 r4, #8, r4
+ usat16 r7, #8, r7
+ usat16 r6, #8, r6
+ orr r5, r5, r4, lsl #8
+ orr r7, r7, r6, lsl #8
+ ldr r4, [r1], r3
+ ldr r6, [r1]
+ str r5, [r2], lr
+ str r7, [r2], lr
+
+ uxtab16 r5, r0, r4
+ uxtab16 r4, r0, r4, ror #8
+ uxtab16 r7, r0, r6
+ uxtab16 r6, r0, r6, ror #8
+ usat16 r5, #8, r5
+ usat16 r4, #8, r4
+ usat16 r7, #8, r7
+ usat16 r6, #8, r6
+ orr r5, r5, r4, lsl #8
+ orr r7, r7, r6, lsl #8
+ str r5, [r2], lr
+ str r7, [r2]
+
+ ldmia sp!, {r4 - r7, pc}
+
+ ENDP ; |vp8_dc_only_idct_add_v6|
+
+; Constant Pool
+c0x0000FFFF DCD 0x0000FFFF
+ END
EXPORT |vp8_short_idct4x4llm_v6_scott|
EXPORT |vp8_short_idct4x4llm_v6_dual|
- EXPORT |vp8_dc_only_idct_armv6|
-
AREA |.text|, CODE, READONLY
;********************************************************************************
ldmia sp!, {r4 - r11, pc} ; replace vars, return restore
ENDP
-
-; sjl added 10/17/08
-;void dc_only_idct_armv6(short input_dc, short *output, int pitch)
-|vp8_dc_only_idct_armv6| PROC
- stmdb sp!, {r4 - r6, lr}
-
- add r0, r0, #0x4
- add r4, r1, r2 ; output + shortpitch
- mov r0, r0, ASR #0x3 ;aka a1
- add r5, r1, r2, LSL #1 ; output + shortpitch * 2
- pkhbt r0, r0, r0, lsl #16 ; a1 | a1
- add r6, r5, r2 ; output + shortpitch * 3
-
- str r0, [r1, #0]
- str r0, [r1, #4]
-
- str r0, [r4, #0]
- str r0, [r4, #4]
-
- str r0, [r5, #0]
- str r0, [r5, #4]
-
- str r0, [r6, #0]
- str r0, [r6, #4]
-
-
- ldmia sp!, {r4 - r6, pc}
-
- ENDP ; |vp8_dc_only_idct_armv6|
-
END
; be found in the AUTHORS file in the root of the source tree.
;
- EXPORT |vp8_short_inv_walsh4x4_armv6|
- EXPORT |vp8_short_inv_walsh4x4_1_armv6|
+ EXPORT |vp8_short_inv_walsh4x4_v6|
+ EXPORT |vp8_short_inv_walsh4x4_1_v6|
ARM
REQUIRE8
AREA |.text|, CODE, READONLY ; name this block of code
-;short vp8_short_inv_walsh4x4_armv6(short *input, short *output)
-|vp8_short_inv_walsh4x4_armv6| PROC
+;short vp8_short_inv_walsh4x4_v6(short *input, short *output)
+|vp8_short_inv_walsh4x4_v6| PROC
stmdb sp!, {r4 - r11, lr}
str r5, [r1]
ldmia sp!, {r4 - r11, pc}
- ENDP ; |vp8_short_inv_walsh4x4_armv6|
+ ENDP ; |vp8_short_inv_walsh4x4_v6|
-;short vp8_short_inv_walsh4x4_1_armv6(short *input, short *output)
-|vp8_short_inv_walsh4x4_1_armv6| PROC
+;short vp8_short_inv_walsh4x4_1_v6(short *input, short *output)
+|vp8_short_inv_walsh4x4_1_v6| PROC
ldrsh r2, [r0] ; [0]
add r2, r2, #3 ; [0] + 3
str r2, [r1]
bx lr
- ENDP ; |vp8_short_inv_walsh4x4_1_armv6|
+ ENDP ; |vp8_short_inv_walsh4x4_1_v6|
; Constant Pool
c0x00030003 DCD 0x00030003
#if HAVE_ARMV6
extern prototype_idct(vp8_short_idct4x4llm_1_v6);
extern prototype_idct(vp8_short_idct4x4llm_v6_dual);
-extern prototype_second_order(vp8_short_inv_walsh4x4_1_armv6);
-extern prototype_second_order(vp8_short_inv_walsh4x4_armv6);
+extern prototype_idct_scalar_add(vp8_dc_only_idct_add_v6);
+extern prototype_second_order(vp8_short_inv_walsh4x4_1_v6);
+extern prototype_second_order(vp8_short_inv_walsh4x4_v6);
#undef vp8_idct_idct1
#define vp8_idct_idct1 vp8_short_idct4x4llm_1_v6
#undef vp8_idct_idct16
#define vp8_idct_idct16 vp8_short_idct4x4llm_v6_dual
+#undef vp8_idct_idct1_scalar_add
+#define vp8_idct_idct1_scalar_add vp8_dc_only_idct_add_v6
+
#undef vp8_idct_iwalsh1
-#define vp8_idct_iwalsh1 vp8_short_inv_walsh4x4_1_armv6
+#define vp8_idct_iwalsh1 vp8_short_inv_walsh4x4_1_v6
#undef vp8_idct_iwalsh16
-#define vp8_idct_iwalsh16 vp8_short_inv_walsh4x4_armv6
+#define vp8_idct_iwalsh16 vp8_short_inv_walsh4x4_v6
#endif
#if HAVE_ARMV7
extern prototype_idct(vp8_short_idct4x4llm_1_neon);
extern prototype_idct(vp8_short_idct4x4llm_neon);
+extern prototype_idct_scalar_add(vp8_dc_only_idct_add_neon);
extern prototype_second_order(vp8_short_inv_walsh4x4_1_neon);
extern prototype_second_order(vp8_short_inv_walsh4x4_neon);
#undef vp8_idct_idct16
#define vp8_idct_idct16 vp8_short_idct4x4llm_neon
+#undef vp8_idct_idct1_scalar_add
+#define vp8_idct_idct1_scalar_add vp8_dc_only_idct_add_neon
+
#undef vp8_idct_iwalsh1
#define vp8_idct_iwalsh1 vp8_short_inv_walsh4x4_1_neon
--- /dev/null
+;
+; Copyright (c) 2010 The VP8 project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license and patent
+; grant that can be found in the LICENSE file in the root of the source
+; tree. All contributing project authors may be found in the AUTHORS
+; file in the root of the source tree.
+;
+
+
+ EXPORT |vp8_dc_only_idct_add_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+;void vp8_dc_only_idct_add_neon(short input_dc, unsigned char *pred_ptr,
+; unsigned char *dst_ptr, int pitch, int stride)
+; r0 input_dc
+; r1 pred_ptr
+; r2 dst_ptr
+; r3 pitch
+; sp stride
+|vp8_dc_only_idct_add_neon| PROC
+ add r0, r0, #4
+ asr r0, r0, #3
+ ldr r12, [sp]
+ vdup.16 q0, r0
+
+ vld1.32 {d2[0]}, [r1], r3
+ vld1.32 {d2[1]}, [r1], r3
+ vld1.32 {d4[0]}, [r1], r3
+ vld1.32 {d4[1]}, [r1]
+
+ vaddw.u8 q1, q0, d2
+ vaddw.u8 q2, q0, d4
+
+ vqmovun.s16 d2, q1
+ vqmovun.s16 d4, q2
+
+ vst1.32 {d2[0]}, [r2], r12
+ vst1.32 {d2[1]}, [r2], r12
+ vst1.32 {d4[0]}, [r2], r12
+ vst1.32 {d4[1]}, [r2]
+
+ bx lr
+
+ ENDP
+ END
--- /dev/null
+;
+; Copyright (c) 2010 The VP8 project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license and patent
+; grant that can be found in the LICENSE file in the root of the source
+; tree. All contributing project authors may be found in the AUTHORS
+; file in the root of the source tree.
+;
+
+
+ EXPORT |vp8_dequant_dc_idct_add_v6|
+
+ AREA |.text|, CODE, READONLY
+
+;void vp8_dequant_dc_idct_v6(short *input, short *dq, unsigned char *pred,
+; unsigned char *dest, int pitch, int stride, int Dc)
+; r0 = input
+; r1 = dq
+; r2 = pred
+; r3 = dest
+; sp + 36 = pitch ; +4 = 40
+; sp + 40 = stride ; +4 = 44
+; sp + 44 = Dc ; +4 = 48
+
+
+|vp8_dequant_dc_idct_add_v6| PROC
+ stmdb sp!, {r4-r11, lr}
+
+ ldr r6, [sp, #44]
+
+ ldr r4, [r0] ;input
+ ldr r5, [r1], #4 ;dq
+
+ sub sp, sp, #4
+ str r3, [sp]
+
+ smultt r7, r4, r5
+
+ ldr r4, [r0, #4] ;input
+ ldr r5, [r1], #4 ;dq
+
+ strh r6, [r0], #2
+ strh r7, [r0], #2
+
+ smulbb r6, r4, r5
+ smultt r7, r4, r5
+
+ ldr r4, [r0, #4] ;input
+ ldr r5, [r1], #4 ;dq
+
+ strh r6, [r0], #2
+ strh r7, [r0], #2
+
+ mov r12, #3
+
+vp8_dequant_dc_add_loop
+ smulbb r6, r4, r5
+ smultt r7, r4, r5
+
+ ldr r4, [r0, #4] ;input
+ ldr r5, [r1], #4 ;dq
+
+ strh r6, [r0], #2
+ strh r7, [r0], #2
+
+ smulbb r6, r4, r5
+ smultt r7, r4, r5
+
+ subs r12, r12, #1
+
+ ldrne r4, [r0, #4]
+ ldrne r5, [r1], #4
+
+ strh r6, [r0], #2
+ strh r7, [r0], #2
+
+ bne vp8_dequant_dc_add_loop
+
+ sub r0, r0, #32
+ mov r1, r0
+
+; short_idct4x4llm_v6_dual
+ ldr r3, cospi8sqrt2minus1
+ ldr r4, sinpi8sqrt2
+ ldr r6, [r0, #8]
+ mov r5, #2
+vp8_dequant_dc_idct_loop1_v6
+ ldr r12, [r0, #24]
+ ldr r14, [r0, #16]
+ smulwt r9, r3, r6
+ smulwb r7, r3, r6
+ smulwt r10, r4, r6
+ smulwb r8, r4, r6
+ pkhbt r7, r7, r9, lsl #16
+ smulwt r11, r3, r12
+ pkhbt r8, r8, r10, lsl #16
+ uadd16 r6, r6, r7
+ smulwt r7, r4, r12
+ smulwb r9, r3, r12
+ smulwb r10, r4, r12
+ subs r5, r5, #1
+ pkhbt r9, r9, r11, lsl #16
+ ldr r11, [r0], #4
+ pkhbt r10, r10, r7, lsl #16
+ uadd16 r7, r12, r9
+ usub16 r7, r8, r7
+ uadd16 r6, r6, r10
+ uadd16 r10, r11, r14
+ usub16 r8, r11, r14
+ uadd16 r9, r10, r6
+ usub16 r10, r10, r6
+ uadd16 r6, r8, r7
+ usub16 r7, r8, r7
+ str r6, [r1, #8]
+ ldrne r6, [r0, #8]
+ str r7, [r1, #16]
+ str r10, [r1, #24]
+ str r9, [r1], #4
+ bne vp8_dequant_dc_idct_loop1_v6
+
+ mov r5, #2
+ sub r0, r1, #8
+vp8_dequant_dc_idct_loop2_v6
+ ldr r6, [r0], #4
+ ldr r7, [r0], #4
+ ldr r8, [r0], #4
+ ldr r9, [r0], #4
+ smulwt r1, r3, r6
+ smulwt r12, r4, r6
+ smulwt lr, r3, r8
+ smulwt r10, r4, r8
+ pkhbt r11, r8, r6, lsl #16
+ pkhbt r1, lr, r1, lsl #16
+ pkhbt r12, r10, r12, lsl #16
+ pkhtb r6, r6, r8, asr #16
+ uadd16 r6, r1, r6
+ pkhbt lr, r9, r7, lsl #16
+ uadd16 r10, r11, lr
+ usub16 lr, r11, lr
+ pkhtb r8, r7, r9, asr #16
+ subs r5, r5, #1
+ smulwt r1, r3, r8
+ smulwb r7, r3, r8
+ smulwt r11, r4, r8
+ smulwb r9, r4, r8
+ pkhbt r1, r7, r1, lsl #16
+ uadd16 r8, r1, r8
+ pkhbt r11, r9, r11, lsl #16
+ usub16 r1, r12, r8
+ uadd16 r8, r11, r6
+ ldr r9, c0x00040004
+ ldr r12, [sp, #40]
+ uadd16 r6, r10, r8
+ usub16 r7, r10, r8
+ uadd16 r7, r7, r9
+ uadd16 r6, r6, r9
+ uadd16 r10, r14, r1
+ usub16 r1, r14, r1
+ uadd16 r10, r10, r9
+ uadd16 r1, r1, r9
+ ldr r11, [r2], r12
+ mov r8, r7, asr #3
+ pkhtb r9, r8, r10, asr #19
+ mov r8, r1, asr #3
+ pkhtb r8, r8, r6, asr #19
+ uxtb16 lr, r11, ror #8
+ qadd16 r9, r9, lr
+ uxtb16 lr, r11
+ qadd16 r8, r8, lr
+ usat16 r9, #8, r9
+ usat16 r8, #8, r8
+ orr r9, r8, r9, lsl #8
+ ldr r11, [r2], r12
+ ldr lr, [sp]
+ ldr r12, [sp, #44]
+ mov r7, r7, lsl #16
+ mov r1, r1, lsl #16
+ mov r10, r10, lsl #16
+ mov r6, r6, lsl #16
+ mov r7, r7, asr #3
+ pkhtb r7, r7, r10, asr #19
+ mov r1, r1, asr #3
+ pkhtb r1, r1, r6, asr #19
+ uxtb16 r8, r11, ror #8
+ qadd16 r7, r7, r8
+ uxtb16 r8, r11
+ qadd16 r1, r1, r8
+ usat16 r7, #8, r7
+ usat16 r1, #8, r1
+ orr r1, r1, r7, lsl #8
+ str r9, [lr], r12
+ str r1, [lr], r12
+ str lr, [sp]
+ bne vp8_dequant_dc_idct_loop2_v6
+
+; vpx_memset
+ sub r0, r0, #32
+ add sp, sp, #4
+
+ mov r12, #0
+ str r12, [r0]
+ str r12, [r0, #4]
+ str r12, [r0, #8]
+ str r12, [r0, #12]
+ str r12, [r0, #16]
+ str r12, [r0, #20]
+ str r12, [r0, #24]
+ str r12, [r0, #28]
+
+ ldmia sp!, {r4 - r11, pc}
+ ENDP ; |vp8_dequant_dc_idct_add_v6|
+
+; Constant Pool
+cospi8sqrt2minus1 DCD 0x00004E7B
+sinpi8sqrt2 DCD 0x00008A8C
+c0x00040004 DCD 0x00040004
+
+ END
--- /dev/null
+;
+; Copyright (c) 2010 The VP8 project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license and patent
+; grant that can be found in the LICENSE file in the root of the source
+; tree. All contributing project authors may be found in the AUTHORS
+; file in the root of the source tree.
+;
+
+ EXPORT |vp8_dequant_idct_add_v6|
+
+ AREA |.text|, CODE, READONLY
+;void vp8_dequant_idct_v6(short *input, short *dq, unsigned char *pred,
+; unsigned char *dest, int pitch, int stride)
+; r0 = input
+; r1 = dq
+; r2 = pred
+; r3 = dest
+; sp + 36 = pitch ; +4 = 40
+; sp + 40 = stride ; +4 = 44
+
+
+|vp8_dequant_idct_add_v6| PROC
+ stmdb sp!, {r4-r11, lr}
+
+ ldr r4, [r0] ;input
+ ldr r5, [r1], #4 ;dq
+
+ sub sp, sp, #4
+ str r3, [sp]
+
+ mov r12, #4
+
+vp8_dequant_add_loop
+ smulbb r6, r4, r5
+ smultt r7, r4, r5
+
+ ldr r4, [r0, #4] ;input
+ ldr r5, [r1], #4 ;dq
+
+ strh r6, [r0], #2
+ strh r7, [r0], #2
+
+ smulbb r6, r4, r5
+ smultt r7, r4, r5
+
+ subs r12, r12, #1
+
+ ldrne r4, [r0, #4]
+ ldrne r5, [r1], #4
+
+ strh r6, [r0], #2
+ strh r7, [r0], #2
+
+ bne vp8_dequant_add_loop
+
+ sub r0, r0, #32
+ mov r1, r0
+
+; short_idct4x4llm_v6_dual
+ ldr r3, cospi8sqrt2minus1
+ ldr r4, sinpi8sqrt2
+ ldr r6, [r0, #8]
+ mov r5, #2
+vp8_dequant_idct_loop1_v6
+ ldr r12, [r0, #24]
+ ldr r14, [r0, #16]
+ smulwt r9, r3, r6
+ smulwb r7, r3, r6
+ smulwt r10, r4, r6
+ smulwb r8, r4, r6
+ pkhbt r7, r7, r9, lsl #16
+ smulwt r11, r3, r12
+ pkhbt r8, r8, r10, lsl #16
+ uadd16 r6, r6, r7
+ smulwt r7, r4, r12
+ smulwb r9, r3, r12
+ smulwb r10, r4, r12
+ subs r5, r5, #1
+ pkhbt r9, r9, r11, lsl #16
+ ldr r11, [r0], #4
+ pkhbt r10, r10, r7, lsl #16
+ uadd16 r7, r12, r9
+ usub16 r7, r8, r7
+ uadd16 r6, r6, r10
+ uadd16 r10, r11, r14
+ usub16 r8, r11, r14
+ uadd16 r9, r10, r6
+ usub16 r10, r10, r6
+ uadd16 r6, r8, r7
+ usub16 r7, r8, r7
+ str r6, [r1, #8]
+ ldrne r6, [r0, #8]
+ str r7, [r1, #16]
+ str r10, [r1, #24]
+ str r9, [r1], #4
+ bne vp8_dequant_idct_loop1_v6
+
+ mov r5, #2
+ sub r0, r1, #8
+vp8_dequant_idct_loop2_v6
+ ldr r6, [r0], #4
+ ldr r7, [r0], #4
+ ldr r8, [r0], #4
+ ldr r9, [r0], #4
+ smulwt r1, r3, r6
+ smulwt r12, r4, r6
+ smulwt lr, r3, r8
+ smulwt r10, r4, r8
+ pkhbt r11, r8, r6, lsl #16
+ pkhbt r1, lr, r1, lsl #16
+ pkhbt r12, r10, r12, lsl #16
+ pkhtb r6, r6, r8, asr #16
+ uadd16 r6, r1, r6
+ pkhbt lr, r9, r7, lsl #16
+ uadd16 r10, r11, lr
+ usub16 lr, r11, lr
+ pkhtb r8, r7, r9, asr #16
+ subs r5, r5, #1
+ smulwt r1, r3, r8
+ smulwb r7, r3, r8
+ smulwt r11, r4, r8
+ smulwb r9, r4, r8
+ pkhbt r1, r7, r1, lsl #16
+ uadd16 r8, r1, r8
+ pkhbt r11, r9, r11, lsl #16
+ usub16 r1, r12, r8
+ uadd16 r8, r11, r6
+ ldr r9, c0x00040004
+ ldr r12, [sp, #40]
+ uadd16 r6, r10, r8
+ usub16 r7, r10, r8
+ uadd16 r7, r7, r9
+ uadd16 r6, r6, r9
+ uadd16 r10, r14, r1
+ usub16 r1, r14, r1
+ uadd16 r10, r10, r9
+ uadd16 r1, r1, r9
+ ldr r11, [r2], r12
+ mov r8, r7, asr #3
+ pkhtb r9, r8, r10, asr #19
+ mov r8, r1, asr #3
+ pkhtb r8, r8, r6, asr #19
+ uxtb16 lr, r11, ror #8
+ qadd16 r9, r9, lr
+ uxtb16 lr, r11
+ qadd16 r8, r8, lr
+ usat16 r9, #8, r9
+ usat16 r8, #8, r8
+ orr r9, r8, r9, lsl #8
+ ldr r11, [r2], r12
+ ldr lr, [sp]
+ ldr r12, [sp, #44]
+ mov r7, r7, lsl #16
+ mov r1, r1, lsl #16
+ mov r10, r10, lsl #16
+ mov r6, r6, lsl #16
+ mov r7, r7, asr #3
+ pkhtb r7, r7, r10, asr #19
+ mov r1, r1, asr #3
+ pkhtb r1, r1, r6, asr #19
+ uxtb16 r8, r11, ror #8
+ qadd16 r7, r7, r8
+ uxtb16 r8, r11
+ qadd16 r1, r1, r8
+ usat16 r7, #8, r7
+ usat16 r1, #8, r1
+ orr r1, r1, r7, lsl #8
+ str r9, [lr], r12
+ str r1, [lr], r12
+ str lr, [sp]
+ bne vp8_dequant_idct_loop2_v6
+
+; vpx_memset
+ sub r0, r0, #32
+ add sp, sp, #4
+
+ mov r12, #0
+ str r12, [r0]
+ str r12, [r0, #4]
+ str r12, [r0, #8]
+ str r12, [r0, #12]
+ str r12, [r0, #16]
+ str r12, [r0, #20]
+ str r12, [r0, #24]
+ str r12, [r0, #28]
+
+ ldmia sp!, {r4 - r11, pc}
+ ENDP ; |vp8_dequant_idct_add_v6|
+
+; Constant Pool
+cospi8sqrt2minus1 DCD 0x00004E7B
+sinpi8sqrt2 DCD 0x00008A8C
+c0x00040004 DCD 0x00040004
+
+ END
+++ /dev/null
-;
-; Copyright (c) 2010 The VP8 project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
- EXPORT |vp8_dequant_dc_idct_v6|
- ; ARM
- ; REQUIRE8
- ; PRESERVE8
-
- AREA |.text|, CODE, READONLY ; name this block of code
-;void vp8_dequant_dc_idct_v6(short *input, short *dq, short *output, int pitch,int Dc)
-|vp8_dequant_dc_idct_v6| PROC
- stmdb sp!, {r4-r11, lr}
-
- ldr r6, [sp, #36] ;load Dc
-
- ldr r4, [r0] ;input
- ldr r5, [r1], #4 ;dq
-
- sub sp, sp, #4
- str r0, [sp]
-
- smultt r7, r4, r5
-
- ldr r4, [r0, #4] ;input
- ldr r5, [r1], #4 ;dq
-
- strh r6, [r0], #2
- strh r7, [r0], #2
-
- smulbb r6, r4, r5
- smultt r7, r4, r5
-
- ldr r4, [r0, #4] ;input
- ldr r5, [r1], #4 ;dq
-
- strh r6, [r0], #2
- strh r7, [r0], #2
-
- mov r12, #3
-
-dequant_dc_idct_loop
- smulbb r6, r4, r5
- smultt r7, r4, r5
-
- ldr r4, [r0, #4] ;input
- ldr r5, [r1], #4 ;dq
-
- strh r6, [r0], #2
- strh r7, [r0], #2
-
- smulbb r6, r4, r5
- smultt r7, r4, r5
-
- subs r12, r12, #1
-
- ldrne r4, [r0, #4]
- ldrne r5, [r1], #4
-
- strh r6, [r0], #2
- strh r7, [r0], #2
-
- bne dequant_dc_idct_loop
-
- sub r0, r0, #32
- mov r1, r2
- mov r2, r3
-
-; short_idct4x4llm_v6_dual
-
- mov r3, #0x00004E00 ; cos
- orr r3, r3, #0x0000007B ; cospi8sqrt2minus1
- mov r4, #0x00008A00 ; sin
- orr r4, r4, #0x0000008C ; sinpi8sqrt2
- mov r5, #0x2 ; i=2 i
-loop1_dual_11
- ldr r6, [r0, #(4*2)] ; i5 | i4 5|4
- ldr r12, [r0, #(12*2)] ; i13 | i12 13|12
- ldr r14, [r0, #(8*2)] ; i9 | i8 9|8
-
- smulwt r9, r3, r6 ; (ip[5] * cospi8sqrt2minus1) >> 16 5c
- smulwb r7, r3, r6 ; (ip[4] * cospi8sqrt2minus1) >> 16 4c
- smulwt r10, r4, r6 ; (ip[5] * sinpi8sqrt2) >> 16 5s
- smulwb r8, r4, r6 ; (ip[4] * sinpi8sqrt2) >> 16 4s
- pkhbt r7, r7, r9, lsl #16 ; 5c | 4c
- smulwt r11, r3, r12 ; (ip[13] * cospi8sqrt2minus1) >> 16 13c
- pkhbt r8, r8, r10, lsl #16 ; 5s | 4s
- uadd16 r6, r6, r7 ; 5c+5 | 4c+4
- smulwt r7, r4, r12 ; (ip[13] * sinpi8sqrt2) >> 16 13s
- smulwb r9, r3, r12 ; (ip[12] * cospi8sqrt2minus1) >> 16 12c
- smulwb r10, r4, r12 ; (ip[12] * sinpi8sqrt2) >> 16 12s
- subs r5, r5, #0x1 ; i-- --
- pkhbt r9, r9, r11, lsl #16 ; 13c | 12c
- ldr r11, [r0], #0x4 ; i1 | i0 ++ 1|0
- pkhbt r10, r10, r7, lsl #16 ; 13s | 12s
- uadd16 r7, r12, r9 ; 13c+13 | 12c+12
- usub16 r7, r8, r7 ; c c
- uadd16 r6, r6, r10 ; d d
- uadd16 r10, r11, r14 ; a a
- usub16 r8, r11, r14 ; b b
- uadd16 r9, r10, r6 ; a+d a+d
- usub16 r10, r10, r6 ; a-d a-d
- uadd16 r6, r8, r7 ; b+c b+c
- usub16 r7, r8, r7 ; b-c b-c
- str r6, [r1, r2] ; o5 | o4
- add r6, r2, r2 ; pitch * 2 p2
- str r7, [r1, r6] ; o9 | o8
- add r6, r6, r2 ; pitch * 3 p3
- str r10, [r1, r6] ; o13 | o12
- str r9, [r1], #0x4 ; o1 | o0 ++
- bne loop1_dual_11 ;
- mov r5, #0x2 ; i=2 i
- sub r0, r1, #8 ; reset input/output i/o
-loop2_dual_22
- ldr r6, [r0, r2] ; i5 | i4 5|4
- ldr r1, [r0] ; i1 | i0 1|0
- ldr r12, [r0, #0x4] ; i3 | i2 3|2
- add r14, r2, #0x4 ; pitch + 2 p+2
- ldr r14, [r0, r14] ; i7 | i6 7|6
- smulwt r9, r3, r6 ; (ip[5] * cospi8sqrt2minus1) >> 16 5c
- smulwt r7, r3, r1 ; (ip[1] * cospi8sqrt2minus1) >> 16 1c
- smulwt r10, r4, r6 ; (ip[5] * sinpi8sqrt2) >> 16 5s
- smulwt r8, r4, r1 ; (ip[1] * sinpi8sqrt2) >> 16 1s
- pkhbt r11, r6, r1, lsl #16 ; i0 | i4 0|4
- pkhbt r7, r9, r7, lsl #16 ; 1c | 5c
- pkhbt r8, r10, r8, lsl #16 ; 1s | 5s = temp1 © tc1
- pkhtb r1, r1, r6, asr #16 ; i1 | i5 1|5
- uadd16 r1, r7, r1 ; 1c+1 | 5c+5 = temp2 (d) td2
- pkhbt r9, r14, r12, lsl #16 ; i2 | i6 2|6
- uadd16 r10, r11, r9 ; a a
- usub16 r9, r11, r9 ; b b
- pkhtb r6, r12, r14, asr #16 ; i3 | i7 3|7
- subs r5, r5, #0x1 ; i-- --
- smulwt r7, r3, r6 ; (ip[3] * cospi8sqrt2minus1) >> 16 3c
- smulwt r11, r4, r6 ; (ip[3] * sinpi8sqrt2) >> 16 3s
- smulwb r12, r3, r6 ; (ip[7] * cospi8sqrt2minus1) >> 16 7c
- smulwb r14, r4, r6 ; (ip[7] * sinpi8sqrt2) >> 16 7s
-
- pkhbt r7, r12, r7, lsl #16 ; 3c | 7c
- pkhbt r11, r14, r11, lsl #16 ; 3s | 7s = temp1 (d) td1
- uadd16 r6, r7, r6 ; 3c+3 | 7c+7 = temp2 (c) tc2
- usub16 r12, r8, r6 ; c (o1 | o5) c
- uadd16 r6, r11, r1 ; d (o3 | o7) d
- uadd16 r7, r10, r6 ; a+d a+d
- mov r8, #0x4 ; set up 4's 4
- orr r8, r8, #0x40000 ; 4|4
- usub16 r6, r10, r6 ; a-d a-d
- uadd16 r6, r6, r8 ; a-d+4 3|7
- uadd16 r7, r7, r8 ; a+d+4 0|4
- uadd16 r10, r9, r12 ; b+c b+c
- usub16 r1, r9, r12 ; b-c b-c
- uadd16 r10, r10, r8 ; b+c+4 1|5
- uadd16 r1, r1, r8 ; b-c+4 2|6
- mov r8, r10, asr #19 ; o1 >> 3
- strh r8, [r0, #2] ; o1
- mov r8, r1, asr #19 ; o2 >> 3
- strh r8, [r0, #4] ; o2
- mov r8, r6, asr #19 ; o3 >> 3
- strh r8, [r0, #6] ; o3
- mov r8, r7, asr #19 ; o0 >> 3
- strh r8, [r0], r2 ; o0 +p
- sxth r10, r10 ;
- mov r8, r10, asr #3 ; o5 >> 3
- strh r8, [r0, #2] ; o5
- sxth r1, r1 ;
- mov r8, r1, asr #3 ; o6 >> 3
- strh r8, [r0, #4] ; o6
- sxth r6, r6 ;
- mov r8, r6, asr #3 ; o7 >> 3
- strh r8, [r0, #6] ; o7
- sxth r7, r7 ;
- mov r8, r7, asr #3 ; o4 >> 3
- strh r8, [r0], r2 ; o4 +p
-;;;;; subs r5, r5, #0x1 ; i-- --
- bne loop2_dual_22 ;
-
-
-;vpx_memset
- ldr r0, [sp]
- add sp, sp, #4
-
- mov r12, #0
- str r12, [r0]
- str r12, [r0, #4]
- str r12, [r0, #8]
- str r12, [r0, #12]
- str r12, [r0, #16]
- str r12, [r0, #20]
- str r12, [r0, #24]
- str r12, [r0, #28]
-
- ldmia sp!, {r4 - r11, pc} ; replace vars, return restore
-
- ENDP ;|vp8_dequant_dc_idct_v68|
-
- END
+++ /dev/null
-;
-; Copyright (c) 2010 The VP8 project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
- EXPORT |vp8_dequant_idct_v6|
- ; ARM
- ; REQUIRE8
- ; PRESERVE8
-
- AREA |.text|, CODE, READONLY ; name this block of code
-;void vp8_dequant_idct_v6(short *input, short *dq, short *output, int pitch)
-|vp8_dequant_idct_v6| PROC
- stmdb sp!, {r4-r11, lr}
-
- ldr r4, [r0] ;input
- ldr r5, [r1], #4 ;dq
-
- sub sp, sp, #4
- str r0, [sp]
-
- mov r12, #4
-
-dequant_idct_loop
- smulbb r6, r4, r5
- smultt r7, r4, r5
-
- ldr r4, [r0, #4] ;input
- ldr r5, [r1], #4 ;dq
-
- strh r6, [r0], #2
- strh r7, [r0], #2
-
- smulbb r6, r4, r5
- smultt r7, r4, r5
-
- subs r12, r12, #1
-
- ldrne r4, [r0, #4]
- ldrne r5, [r1], #4
-
- strh r6, [r0], #2
- strh r7, [r0], #2
-
- bne dequant_idct_loop
-
- sub r0, r0, #32
- mov r1, r2
- mov r2, r3
-
-; short_idct4x4llm_v6_dual
-
- mov r3, #0x00004E00 ; cos
- orr r3, r3, #0x0000007B ; cospi8sqrt2minus1
- mov r4, #0x00008A00 ; sin
- orr r4, r4, #0x0000008C ; sinpi8sqrt2
- mov r5, #0x2 ; i=2 i
-loop1_dual_1
- ldr r6, [r0, #(4*2)] ; i5 | i4 5|4
- ldr r12, [r0, #(12*2)] ; i13 | i12 13|12
- ldr r14, [r0, #(8*2)] ; i9 | i8 9|8
-
- smulwt r9, r3, r6 ; (ip[5] * cospi8sqrt2minus1) >> 16 5c
- smulwb r7, r3, r6 ; (ip[4] * cospi8sqrt2minus1) >> 16 4c
- smulwt r10, r4, r6 ; (ip[5] * sinpi8sqrt2) >> 16 5s
- smulwb r8, r4, r6 ; (ip[4] * sinpi8sqrt2) >> 16 4s
- pkhbt r7, r7, r9, lsl #16 ; 5c | 4c
- smulwt r11, r3, r12 ; (ip[13] * cospi8sqrt2minus1) >> 16 13c
- pkhbt r8, r8, r10, lsl #16 ; 5s | 4s
- uadd16 r6, r6, r7 ; 5c+5 | 4c+4
- smulwt r7, r4, r12 ; (ip[13] * sinpi8sqrt2) >> 16 13s
- smulwb r9, r3, r12 ; (ip[12] * cospi8sqrt2minus1) >> 16 12c
- smulwb r10, r4, r12 ; (ip[12] * sinpi8sqrt2) >> 16 12s
- subs r5, r5, #0x1 ; i-- --
- pkhbt r9, r9, r11, lsl #16 ; 13c | 12c
- ldr r11, [r0], #0x4 ; i1 | i0 ++ 1|0
- pkhbt r10, r10, r7, lsl #16 ; 13s | 12s
- uadd16 r7, r12, r9 ; 13c+13 | 12c+12
- usub16 r7, r8, r7 ; c c
- uadd16 r6, r6, r10 ; d d
- uadd16 r10, r11, r14 ; a a
- usub16 r8, r11, r14 ; b b
- uadd16 r9, r10, r6 ; a+d a+d
- usub16 r10, r10, r6 ; a-d a-d
- uadd16 r6, r8, r7 ; b+c b+c
- usub16 r7, r8, r7 ; b-c b-c
- str r6, [r1, r2] ; o5 | o4
- add r6, r2, r2 ; pitch * 2 p2
- str r7, [r1, r6] ; o9 | o8
- add r6, r6, r2 ; pitch * 3 p3
- str r10, [r1, r6] ; o13 | o12
- str r9, [r1], #0x4 ; o1 | o0 ++
- bne loop1_dual_1 ;
- mov r5, #0x2 ; i=2 i
- sub r0, r1, #8 ; reset input/output i/o
-loop2_dual_2
- ldr r6, [r0, r2] ; i5 | i4 5|4
- ldr r1, [r0] ; i1 | i0 1|0
- ldr r12, [r0, #0x4] ; i3 | i2 3|2
- add r14, r2, #0x4 ; pitch + 2 p+2
- ldr r14, [r0, r14] ; i7 | i6 7|6
- smulwt r9, r3, r6 ; (ip[5] * cospi8sqrt2minus1) >> 16 5c
- smulwt r7, r3, r1 ; (ip[1] * cospi8sqrt2minus1) >> 16 1c
- smulwt r10, r4, r6 ; (ip[5] * sinpi8sqrt2) >> 16 5s
- smulwt r8, r4, r1 ; (ip[1] * sinpi8sqrt2) >> 16 1s
- pkhbt r11, r6, r1, lsl #16 ; i0 | i4 0|4
- pkhbt r7, r9, r7, lsl #16 ; 1c | 5c
- pkhbt r8, r10, r8, lsl #16 ; 1s | 5s = temp1 © tc1
- pkhtb r1, r1, r6, asr #16 ; i1 | i5 1|5
- uadd16 r1, r7, r1 ; 1c+1 | 5c+5 = temp2 (d) td2
- pkhbt r9, r14, r12, lsl #16 ; i2 | i6 2|6
- uadd16 r10, r11, r9 ; a a
- usub16 r9, r11, r9 ; b b
- pkhtb r6, r12, r14, asr #16 ; i3 | i7 3|7
- subs r5, r5, #0x1 ; i-- --
- smulwt r7, r3, r6 ; (ip[3] * cospi8sqrt2minus1) >> 16 3c
- smulwt r11, r4, r6 ; (ip[3] * sinpi8sqrt2) >> 16 3s
- smulwb r12, r3, r6 ; (ip[7] * cospi8sqrt2minus1) >> 16 7c
- smulwb r14, r4, r6 ; (ip[7] * sinpi8sqrt2) >> 16 7s
-
- pkhbt r7, r12, r7, lsl #16 ; 3c | 7c
- pkhbt r11, r14, r11, lsl #16 ; 3s | 7s = temp1 (d) td1
- uadd16 r6, r7, r6 ; 3c+3 | 7c+7 = temp2 (c) tc2
- usub16 r12, r8, r6 ; c (o1 | o5) c
- uadd16 r6, r11, r1 ; d (o3 | o7) d
- uadd16 r7, r10, r6 ; a+d a+d
- mov r8, #0x4 ; set up 4's 4
- orr r8, r8, #0x40000 ; 4|4
- usub16 r6, r10, r6 ; a-d a-d
- uadd16 r6, r6, r8 ; a-d+4 3|7
- uadd16 r7, r7, r8 ; a+d+4 0|4
- uadd16 r10, r9, r12 ; b+c b+c
- usub16 r1, r9, r12 ; b-c b-c
- uadd16 r10, r10, r8 ; b+c+4 1|5
- uadd16 r1, r1, r8 ; b-c+4 2|6
- mov r8, r10, asr #19 ; o1 >> 3
- strh r8, [r0, #2] ; o1
- mov r8, r1, asr #19 ; o2 >> 3
- strh r8, [r0, #4] ; o2
- mov r8, r6, asr #19 ; o3 >> 3
- strh r8, [r0, #6] ; o3
- mov r8, r7, asr #19 ; o0 >> 3
- strh r8, [r0], r2 ; o0 +p
- sxth r10, r10 ;
- mov r8, r10, asr #3 ; o5 >> 3
- strh r8, [r0, #2] ; o5
- sxth r1, r1 ;
- mov r8, r1, asr #3 ; o6 >> 3
- strh r8, [r0, #4] ; o6
- sxth r6, r6 ;
- mov r8, r6, asr #3 ; o7 >> 3
- strh r8, [r0, #6] ; o7
- sxth r7, r7 ;
- mov r8, r7, asr #3 ; o4 >> 3
- strh r8, [r0], r2 ; o4 +p
-;;;;; subs r5, r5, #0x1 ; i-- --
- bne loop2_dual_2 ;
- ;
-
-;vpx_memset
- ldr r0, [sp]
- add sp, sp, #4
-
- mov r12, #0
- str r12, [r0]
- str r12, [r0, #4]
- str r12, [r0, #8]
- str r12, [r0, #12]
- str r12, [r0, #16]
- str r12, [r0, #20]
- str r12, [r0, #24]
- str r12, [r0, #28]
-
- ldmia sp!, {r4 - r11, pc} ; replace vars, return restore
-
- ENDP ;|vp8_dequant_idct_v6|
-
- END
#if HAVE_ARMV6
extern prototype_dequant_block(vp8_dequantize_b_v6);
+extern prototype_dequant_idct_add(vp8_dequant_idct_add_v6);
+extern prototype_dequant_dc_idct_add(vp8_dequant_dc_idct_add_v6);
#undef vp8_dequant_block
#define vp8_dequant_block vp8_dequantize_b_v6
+#undef vp8_dequant_idct_add
+#define vp8_dequant_idct_add vp8_dequant_idct_add_v6
+
+#undef vp8_dequant_dc_idct_add
+#define vp8_dequant_dc_idct_add vp8_dequant_dc_idct_add_v6
+#endif
+
#if HAVE_ARMV7
extern prototype_dequant_block(vp8_dequantize_b_neon);
+extern prototype_dequant_idct_add(vp8_dequant_idct_add_neon);
+extern prototype_dequant_dc_idct_add(vp8_dequant_dc_idct_add_neon);
#undef vp8_dequant_block
#define vp8_dequant_block vp8_dequantize_b_neon
+#undef vp8_dequant_idct_add
+#define vp8_dequant_idct_add vp8_dequant_idct_add_neon
+
+#undef vp8_dequant_dc_idct_add
+#define vp8_dequant_dc_idct_add vp8_dequant_dc_idct_add_neon
+#endif
+
#endif
;
- EXPORT |vp8_dequant_dc_idct_neon|
+ EXPORT |vp8_dequant_dc_idct_add_neon|
ARM
REQUIRE8
PRESERVE8
AREA ||.text||, CODE, READONLY, ALIGN=2
-;void vp8_dequant_dc_idct_c(short *input, short *dq, short *output, int pitch, int Dc);
+;void vp8_dequant_dc_idct_add_neon(short *input, short *dq, unsigned char *pred,
+; unsigned char *dest, int pitch, int stride,
+; int Dc);
; r0 short *input,
; r1 short *dq,
-; r2 short *output,
-; r3 int pitch,
-; (stack) int Dc
-|vp8_dequant_dc_idct_neon| PROC
+; r2 unsigned char *pred
+; r3 unsigned char *dest
+; sp int pitch
+; sp+4 int stride
+; sp+8 int Dc
+|vp8_dequant_dc_idct_add_neon| PROC
vld1.16 {q3, q4}, [r0]
vld1.16 {q5, q6}, [r1]
- ldr r1, [sp] ;load Dc from stack
+ ldr r1, [sp, #8] ;load Dc from stack
- ldr r12, _dcidct_coeff_
+ ldr r12, _CONSTANTS_
vmul.i16 q1, q3, q5 ;input for short_idct4x4llm_neon
vmul.i16 q2, q4, q6
vmov.16 d2[0], r1
+ ldr r1, [sp] ; pitch
+ vld1.32 {d14[0]}, [r2], r1
+ vld1.32 {d14[1]}, [r2], r1
+ vld1.32 {d15[0]}, [r2], r1
+ vld1.32 {d15[1]}, [r2]
+
+ ldr r1, [sp, #4] ; stride
+
;|short_idct4x4llm_neon| PROC
vld1.16 {d0}, [r12]
vswp d3, d4 ;q2(vp[4] vp[12])
vshr.s16 q3, q3, #1
vshr.s16 q4, q4, #1
- vqadd.s16 q3, q3, q2 ;modify since sinpi8sqrt2 > 65536/2 (negtive number)
+ vqadd.s16 q3, q3, q2
vqadd.s16 q4, q4, q2
- ;d6 - c1:temp1
- ;d7 - d1:temp2
- ;d8 - d1:temp1
- ;d9 - c1:temp2
-
vqsub.s16 d10, d6, d9 ;c1
vqadd.s16 d11, d7, d8 ;d1
vshr.s16 q3, q3, #1
vshr.s16 q4, q4, #1
- vqadd.s16 q3, q3, q2 ;modify since sinpi8sqrt2 > 65536/2 (negtive number)
+ vqadd.s16 q3, q3, q2
vqadd.s16 q4, q4, q2
vqsub.s16 d10, d6, d9 ;c1
vrshr.s16 d4, d4, #3
vrshr.s16 d5, d5, #3
- add r1, r2, r3
- add r12, r1, r3
- add r0, r12, r3
-
vtrn.32 d2, d4
vtrn.32 d3, d5
vtrn.16 d2, d3
vtrn.16 d4, d5
- vst1.16 {d2}, [r2]
- vst1.16 {d3}, [r1]
- vst1.16 {d4}, [r12]
- vst1.16 {d5}, [r0]
+ vaddw.u8 q1, q1, d14
+ vaddw.u8 q2, q2, d15
- bx lr
+ vqmovun.s16 d0, q1
+ vqmovun.s16 d1, q2
+
+ vst1.32 {d0[0]}, [r3], r1
+ vst1.32 {d0[1]}, [r3], r1
+ vst1.32 {d1[0]}, [r3], r1
+ vst1.32 {d1[1]}, [r3]
- ENDP
+ bx lr
-;-----------------
- AREA dcidct4x4_dat, DATA, READWRITE ;read/write by default
-;Data section with name data_area is specified. DCD reserves space in memory for 48 data.
-;One word each is reserved. Label filter_coeff can be used to access the data.
-;Data address: filter_coeff, filter_coeff+4, filter_coeff+8 ...
-_dcidct_coeff_
- DCD dcidct_coeff
-dcidct_coeff
- DCD 0x4e7b4e7b, 0x8a8c8a8c
+ ENDP ; |vp8_dequant_dc_idct_add_neon|
-;20091, 20091, 35468, 35468
+; Constant Pool
+_CONSTANTS_ DCD cospi8sqrt2minus1
+cospi8sqrt2minus1 DCD 0x4e7b4e7b
+sinpi8sqrt2 DCD 0x8a8c8a8c
END
;
- EXPORT |vp8_dequant_idct_neon|
+ EXPORT |vp8_dequant_idct_add_neon|
ARM
REQUIRE8
PRESERVE8
AREA ||.text||, CODE, READONLY, ALIGN=2
-;void vp8_dequant_idct_c(short *input, short *dq, short *output, int pitch);
+;void vp8_dequant_idct_neon(short *input, short *dq, unsigned char *pred,
+; unsigned char *dest, int pitch, int stride)
; r0 short *input,
; r1 short *dq,
-; r2 short *output,
-; r3 int pitch,
-|vp8_dequant_idct_neon| PROC
+; r2 unsigned char *pred
+; r3 unsigned char *dest
+; sp int pitch
+; sp+4 int stride
+
+|vp8_dequant_idct_add_neon| PROC
vld1.16 {q3, q4}, [r0]
vld1.16 {q5, q6}, [r1]
+ ldr r1, [sp] ; pitch
+ vld1.32 {d14[0]}, [r2], r1
+ vld1.32 {d14[1]}, [r2], r1
+ vld1.32 {d15[0]}, [r2], r1
+ vld1.32 {d15[1]}, [r2]
+
+ ldr r1, [sp, #4] ; stride
- ldr r12, _didct_coeff_
+ ldr r12, _CONSTANTS_
vmul.i16 q1, q3, q5 ;input for short_idct4x4llm_neon
vmul.i16 q2, q4, q6
vshr.s16 q3, q3, #1
vshr.s16 q4, q4, #1
- vqadd.s16 q3, q3, q2 ;modify since sinpi8sqrt2 > 65536/2 (negtive number)
+ vqadd.s16 q3, q3, q2
vqadd.s16 q4, q4, q2
- ;d6 - c1:temp1
- ;d7 - d1:temp2
- ;d8 - d1:temp1
- ;d9 - c1:temp2
-
vqsub.s16 d10, d6, d9 ;c1
vqadd.s16 d11, d7, d8 ;d1
vshr.s16 q3, q3, #1
vshr.s16 q4, q4, #1
- vqadd.s16 q3, q3, q2 ;modify since sinpi8sqrt2 > 65536/2 (negtive number)
+ vqadd.s16 q3, q3, q2
vqadd.s16 q4, q4, q2
vqsub.s16 d10, d6, d9 ;c1
vrshr.s16 d4, d4, #3
vrshr.s16 d5, d5, #3
- add r1, r2, r3
- add r12, r1, r3
- add r0, r12, r3
-
vtrn.32 d2, d4
vtrn.32 d3, d5
vtrn.16 d2, d3
vtrn.16 d4, d5
- vst1.16 {d2}, [r2]
- vst1.16 {d3}, [r1]
- vst1.16 {d4}, [r12]
- vst1.16 {d5}, [r0]
+ vaddw.u8 q1, q1, d14
+ vaddw.u8 q2, q2, d15
- bx lr
+ vqmovun.s16 d0, q1
+ vqmovun.s16 d1, q2
+
+ vst1.32 {d0[0]}, [r3], r1
+ vst1.32 {d0[1]}, [r3], r1
+ vst1.32 {d1[0]}, [r3], r1
+ vst1.32 {d1[1]}, [r3]
- ENDP
+ bx lr
-;-----------------
- AREA didct4x4_dat, DATA, READWRITE ;read/write by default
-;Data section with name data_area is specified. DCD reserves space in memory for 48 data.
-;One word each is reserved. Label filter_coeff can be used to access the data.
-;Data address: filter_coeff, filter_coeff+4, filter_coeff+8 ...
-_didct_coeff_
- DCD didct_coeff
-didct_coeff
- DCD 0x4e7b4e7b, 0x8a8c8a8c
+ ENDP ; |vp8_dequant_idct_add_neon|
-;20091, 20091, 35468, 35468
+; Constant Pool
+_CONSTANTS_ DCD cospi8sqrt2minus1
+cospi8sqrt2minus1 DCD 0x4e7b4e7b
+sinpi8sqrt2 DCD 0x8a8c8a8c
END
if (b->eob > 1)
{
- DEQUANT_INVOKE(&pbi->dequant, idct_dc_add)(b->qcoeff, &b->dequant[0][0], b->predictor, *(b->base_dst) + b->dst, 16, b->dst_stride,
- xd->block[24].diff[i]);
+ DEQUANT_INVOKE(&pbi->dequant, dc_idct_add)
+ (b->qcoeff, &b->dequant[0][0], b->predictor,
+ *(b->base_dst) + b->dst, 16, b->dst_stride,
+ xd->block[24].diff[i]);
}
else
{
}
}
-void vp8_dequant_idct_add_c(short *input, short *dq, unsigned char *pred, unsigned char *dest, int pitch, int stride)
+void vp8_dequant_idct_add_c(short *input, short *dq, unsigned char *pred,
+ unsigned char *dest, int pitch, int stride)
{
- // output needs to be at least pitch * 4 for vp8_short_idct4x4llm_c to work properly
- short output[16*4];
+ short output[16];
short *diff_ptr = output;
int r, c;
int i;
input[i] = dq[i] * input[i];
}
- vp8_short_idct4x4llm_c(input, output, pitch*2);
+ // the idct halves ( >> 1) the pitch
+ vp8_short_idct4x4llm_c(input, output, 4 << 1);
vpx_memset(input, 0, 32);
}
dest += stride;
- diff_ptr += pitch;
+ diff_ptr += 4;
pred += pitch;
}
}
-void vp8_dequant_dc_idct_add_c(short *input, short *dq, unsigned char *pred, unsigned char *dest, int pitch, int stride, int Dc)
+void vp8_dequant_dc_idct_add_c(short *input, short *dq, unsigned char *pred,
+ unsigned char *dest, int pitch, int stride,
+ int Dc)
{
int i;
- // output needs to be at least pitch * 4 for vp8_short_idct4x4llm_c to work properly
- short output[16*4];
+ short output[16];
short *diff_ptr = output;
int r, c;
input[i] = dq[i] * input[i];
}
- vp8_short_idct4x4llm_c(input, output, pitch*2);
+ // the idct halves ( >> 1) the pitch
+ vp8_short_idct4x4llm_c(input, output, 4 << 1);
vpx_memset(input, 0, 32);
}
dest += stride;
- diff_ptr += pitch;
+ diff_ptr += 4;
pred += pitch;
}
}
unsigned char *pred, unsigned char *output, \
int pitch, int stride)
-#define prototype_dequant_idct_dc_add(sym) \
+#define prototype_dequant_dc_idct_add(sym) \
void sym(short *input, short *dq, \
unsigned char *pred, unsigned char *output, \
int pitch, int stride, \
#endif
extern prototype_dequant_idct_add(vp8_dequant_idct_add);
-#ifndef vp8_dequant_idct_dc_add
-#define vp8_dequant_idct_dc_add vp8_dequant_dc_idct_add_c
+#ifndef vp8_dequant_dc_idct_add
+#define vp8_dequant_dc_idct_add vp8_dequant_dc_idct_add_c
#endif
-extern prototype_dequant_idct_dc_add(vp8_dequant_idct_dc_add);
+extern prototype_dequant_dc_idct_add(vp8_dequant_dc_idct_add);
typedef prototype_dequant_block((*vp8_dequant_block_fn_t));
typedef prototype_dequant_idct_add((*vp8_dequant_idct_add_fn_t));
-typedef prototype_dequant_idct_dc_add((*vp8_dequant_idct_dc_add_fn_t));
+typedef prototype_dequant_dc_idct_add((*vp8_dequant_dc_idct_add_fn_t));
typedef struct
{
vp8_dequant_block_fn_t block;
vp8_dequant_idct_add_fn_t idct_add;
- vp8_dequant_idct_dc_add_fn_t idct_dc_add;
+ vp8_dequant_dc_idct_add_fn_t dc_idct_add;
} vp8_dequant_rtcd_vtable_t;
#if CONFIG_RUNTIME_CPU_DETECT
pbi->mb.rtcd = &pbi->common.rtcd;
pbi->dequant.block = vp8_dequantize_b_c;
pbi->dequant.idct_add = vp8_dequant_idct_add_c;
- pbi->dequant.idct_dc_add = vp8_dequant_dc_idct_add_c;
+ pbi->dequant.dc_idct_add = vp8_dequant_dc_idct_add_c;
pbi->dboolhuff.start = vp8dx_start_decode_c;
pbi->dboolhuff.fill = vp8dx_bool_decoder_fill_c;
#if 0 //For use with RTCD, when implemented
#if HAVE_MMX
extern prototype_dequant_block(vp8_dequantize_b_mmx);
extern prototype_dequant_idct_add(vp8_dequant_idct_add_mmx);
-extern prototype_dequant_idct_dc_add(vp8_dequant_dc_idct_add_mmx);
+extern prototype_dequant_dc_idct_add(vp8_dequant_dc_idct_add_mmx);
#if !CONFIG_RUNTIME_CPU_DETECT
#define vp8_dequant_block vp8_dequantize_b_mmx
#undef vp8_dequant_idct_add
-#define vp8_dequant_idct_add vp8_dequant_idct_add_mmx
+#define vp8_dequant_idct_add vp8_dequant_idct_mmx
-#undef vp8_dequant_idct_dc
-#define vp8_dequant_idct_add_dc vp8_dequant_dc_idct_add_mmx
+#undef vp8_dequant_dc_idct_add
+#define vp8_dequant_dc_idct_add vp8_dequant_dc_idct_mmx
#endif
#endif
{
pbi->dequant.block = vp8_dequantize_b_mmx;
pbi->dequant.idct_add = vp8_dequant_idct_add_mmx;
- pbi->dequant.idct_dc_add = vp8_dequant_dc_idct_add_mmx;
+ pbi->dequant.dc_idct_add = vp8_dequant_dc_idct_add_mmx;
}
#endif
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/vpx_asm_offsets.c
VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/filter_c.c
+VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/idctllm.c
VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/recon.c
VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/reconintra4x4.c
VP8_COMMON_SRCS_REMOVE-$(HAVE_ARMV6) += common/generic/systemdependent.c
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/copymem8x4_v6$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/copymem8x8_v6$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/copymem16x16_v6$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/dc_only_idct_add_v6$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/iwalsh_v6$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/filter_v6$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV6) += common/arm/armv6/idct_v6$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/copymem8x4_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/copymem8x8_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/copymem16x16_neon$(ASM)
+VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/dc_only_idct_add_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/iwalsh_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/loopfiltersimplehorizontaledge_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_ARMV7) += common/arm/neon/loopfiltersimpleverticaledge_neon$(ASM)
#File list for armv6
# decoder
-VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/armv6/dequantdcidct_v6$(ASM)
-VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/armv6/dequantidct_v6$(ASM)
+VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/armv6/dequant_dc_idct_v6$(ASM)
+VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/armv6/dequant_idct_v6$(ASM)
VP8_DX_SRCS-$(HAVE_ARMV6) += decoder/arm/armv6/dequantize_v6$(ASM)
#File list for neon
# decoder
-VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/dequantdcidct_neon$(ASM)
-VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/dequantidct_neon$(ASM)
+VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/dequant_dc_idct_neon$(ASM)
+VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/dequant_idct_neon$(ASM)
VP8_DX_SRCS-$(HAVE_ARMV7) += decoder/arm/neon/dequantizeb_neon$(ASM)
projects / platform / upstream / libvpx.git / commitdiff