generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
e0309e1509
vpx/vp8/decoder/arm/neon/idct_dequant_0_2x_neon.asm

80 lines
2.3 KiB
NASM
Raw Normal View History

Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
;
; Copyright (c) 2010 The WebM 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 |idct_dequant_0_2x_neon|
ARM
REQUIRE8
PRESERVE8
AREA ||.text||, CODE, READONLY, ALIGN=2
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
;void idct_dequant_0_2x_neon(short *q, short dq,
; unsigned char *dst, int stride);
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
; r0 *q
; r1 dq
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
; r2 *dst
; r3 stride
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
|idct_dequant_0_2x_neon| PROC
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
push {r4, r5}
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
add r12, r2, #4
vld1.32 {d2[0]}, [r2], r3
vld1.32 {d8[0]}, [r12], r3
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
vld1.32 {d2[1]}, [r2], r3
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
vld1.32 {d8[1]}, [r12], r3
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
vld1.32 {d4[0]}, [r2], r3
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
vld1.32 {d10[0]}, [r12], r3
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
vld1.32 {d4[1]}, [r2], r3
vld1.32 {d10[1]}, [r12], r3
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
ldrh r12, [r0] ; lo q
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
ldrh r4, [r0, #32] ; hi q
mov r5, #0
strh r5, [r0]
strh r5, [r0, #32]
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
sxth r12, r12 ; lo
mul r0, r12, r1
add r0, r0, #4
asr r0, r0, #3
vdup.16 q0, r0
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
sxth r4, r4 ; hi
mul r0, r4, r1
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
add r0, r0, #4
asr r0, r0, #3
vdup.16 q3, r0
vaddw.u8 q1, q0, d2 ; lo
vaddw.u8 q2, q0, d4
vaddw.u8 q4, q3, d8 ; hi
vaddw.u8 q5, q3, d10
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
sub r2, r2, r3, lsl #2 ; dst - 4*stride
add r0, r2, #4
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
vqmovun.s16 d2, q1 ; lo
vqmovun.s16 d4, q2
vqmovun.s16 d8, q4 ; hi
vqmovun.s16 d10, q5
vst1.32 {d2[0]}, [r2], r3 ; lo
vst1.32 {d8[0]}, [r0], r3 ; hi
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
vst1.32 {d2[1]}, [r2], r3
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
vst1.32 {d8[1]}, [r0], r3
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
vst1.32 {d4[0]}, [r2], r3
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
vst1.32 {d10[0]}, [r0], r3
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
vst1.32 {d4[1]}, [r2]
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
vst1.32 {d10[1]}, [r0]
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
pop {r4, r5}
bx lr
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
Remove usage of predict buffer for decode Instead of using the predict buffer, the decoder now writes the predictor into the recon buffer. For blocks with eob=0, unnecessary idcts can be eliminated. This gave a performance boost of ~1.8% for the HD clips used. Tero: Added needed changes to ARM side and scheduled some assembly code to prevent interlocks. Patch Set 6: Merged (I1bcdca7a95aacc3a181b9faa6b10e3a71ee24df3) into this commit because of similarities in the idct functions. Patch Set 7: EC bug fix. Change-Id: Ie31d90b5d3522e1108163f2ac491e455e3f955e6
2011-10-18 18:06:50 +02:00
ENDP ; |idct_dequant_0_2x_neon|
Update NEON wide idcts Expand 93c32a55 which used SSE2 instructions to do two idct/dequant/recons at a time to NEON. Initial working commit. More work needs to be put into rearranging and interlacing the data to take advantage of quadword operations, which is when we'll hopefully see a much better boost Change-Id: I86d59d96f15e0d0f9710253e2c098ac2ff2865d1
2010-09-07 20:21:27 +02:00
END
Reference in New Issue Copy Permalink