summaryrefslogtreecommitdiff
path: root/vp8/common/arm/neon
diff options
context:
space:
mode:
Diffstat (limited to 'vp8/common/arm/neon')
-rw-r--r--vp8/common/arm/neon/bilinearpredict16x16_neon.asm357
-rw-r--r--vp8/common/arm/neon/bilinearpredict4x4_neon.asm130
-rw-r--r--vp8/common/arm/neon/bilinearpredict8x4_neon.asm135
-rw-r--r--vp8/common/arm/neon/bilinearpredict8x8_neon.asm183
-rw-r--r--vp8/common/arm/neon/bilinearpredict_neon_intrinsics.c696
5 files changed, 696 insertions, 805 deletions
diff --git a/vp8/common/arm/neon/bilinearpredict16x16_neon.asm b/vp8/common/arm/neon/bilinearpredict16x16_neon.asm
deleted file mode 100644
index e392786d4..000000000
--- a/vp8/common/arm/neon/bilinearpredict16x16_neon.asm
+++ /dev/null
@@ -1,357 +0,0 @@
-;
-; Copyright (c) 2010 The WebM 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_bilinear_predict16x16_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-; r0 unsigned char *src_ptr,
-; r1 int src_pixels_per_line,
-; r2 int xoffset,
-; r3 int yoffset,
-; r4 unsigned char *dst_ptr,
-; stack(r5) int dst_pitch
-
-|vp8_bilinear_predict16x16_neon| PROC
- push {r4-r5, lr}
-
- adr r12, bifilter16_coeff
- ldr r4, [sp, #12] ;load parameters from stack
- ldr r5, [sp, #16] ;load parameters from stack
-
- cmp r2, #0 ;skip first_pass filter if xoffset=0
- beq secondpass_bfilter16x16_only
-
- add r2, r12, r2, lsl #3 ;calculate filter location
-
- cmp r3, #0 ;skip second_pass filter if yoffset=0
-
- vld1.s32 {d31}, [r2] ;load first_pass filter
-
- beq firstpass_bfilter16x16_only
-
- sub sp, sp, #272 ;reserve space on stack for temporary storage
- vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
- mov lr, sp
- vld1.u8 {d5, d6, d7}, [r0], r1
-
- mov r2, #3 ;loop counter
- vld1.u8 {d8, d9, d10}, [r0], r1
-
- vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
- vld1.u8 {d11, d12, d13}, [r0], r1
-
- vdup.8 d1, d31[4]
-
-;First Pass: output_height lines x output_width columns (17x16)
-filt_blk2d_fp16x16_loop_neon
- pld [r0]
- pld [r0, r1]
- pld [r0, r1, lsl #1]
-
- vmull.u8 q7, d2, d0 ;(src_ptr[0] * vp8_filter[0])
- vmull.u8 q8, d3, d0
- vmull.u8 q9, d5, d0
- vmull.u8 q10, d6, d0
- vmull.u8 q11, d8, d0
- vmull.u8 q12, d9, d0
- vmull.u8 q13, d11, d0
- vmull.u8 q14, d12, d0
-
- vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
- vext.8 d5, d5, d6, #1
- vext.8 d8, d8, d9, #1
- vext.8 d11, d11, d12, #1
-
- vmlal.u8 q7, d2, d1 ;(src_ptr[0] * vp8_filter[1])
- vmlal.u8 q9, d5, d1
- vmlal.u8 q11, d8, d1
- vmlal.u8 q13, d11, d1
-
- vext.8 d3, d3, d4, #1
- vext.8 d6, d6, d7, #1
- vext.8 d9, d9, d10, #1
- vext.8 d12, d12, d13, #1
-
- vmlal.u8 q8, d3, d1 ;(src_ptr[0] * vp8_filter[1])
- vmlal.u8 q10, d6, d1
- vmlal.u8 q12, d9, d1
- vmlal.u8 q14, d12, d1
-
- subs r2, r2, #1
-
- vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d15, q8, #7
- vqrshrn.u16 d16, q9, #7
- vqrshrn.u16 d17, q10, #7
- vqrshrn.u16 d18, q11, #7
- vqrshrn.u16 d19, q12, #7
- vqrshrn.u16 d20, q13, #7
-
- vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
- vqrshrn.u16 d21, q14, #7
- vld1.u8 {d5, d6, d7}, [r0], r1
-
- vst1.u8 {d14, d15, d16, d17}, [lr]! ;store result
- vld1.u8 {d8, d9, d10}, [r0], r1
- vst1.u8 {d18, d19, d20, d21}, [lr]!
- vld1.u8 {d11, d12, d13}, [r0], r1
-
- bne filt_blk2d_fp16x16_loop_neon
-
-;First-pass filtering for rest 5 lines
- vld1.u8 {d14, d15, d16}, [r0], r1
-
- vmull.u8 q9, d2, d0 ;(src_ptr[0] * vp8_filter[0])
- vmull.u8 q10, d3, d0
- vmull.u8 q11, d5, d0
- vmull.u8 q12, d6, d0
- vmull.u8 q13, d8, d0
- vmull.u8 q14, d9, d0
-
- vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
- vext.8 d5, d5, d6, #1
- vext.8 d8, d8, d9, #1
-
- vmlal.u8 q9, d2, d1 ;(src_ptr[0] * vp8_filter[1])
- vmlal.u8 q11, d5, d1
- vmlal.u8 q13, d8, d1
-
- vext.8 d3, d3, d4, #1
- vext.8 d6, d6, d7, #1
- vext.8 d9, d9, d10, #1
-
- vmlal.u8 q10, d3, d1 ;(src_ptr[0] * vp8_filter[1])
- vmlal.u8 q12, d6, d1
- vmlal.u8 q14, d9, d1
-
- vmull.u8 q1, d11, d0
- vmull.u8 q2, d12, d0
- vmull.u8 q3, d14, d0
- vmull.u8 q4, d15, d0
-
- vext.8 d11, d11, d12, #1 ;construct src_ptr[1]
- vext.8 d14, d14, d15, #1
-
- vmlal.u8 q1, d11, d1 ;(src_ptr[0] * vp8_filter[1])
- vmlal.u8 q3, d14, d1
-
- vext.8 d12, d12, d13, #1
- vext.8 d15, d15, d16, #1
-
- vmlal.u8 q2, d12, d1 ;(src_ptr[0] * vp8_filter[1])
- vmlal.u8 q4, d15, d1
-
- vqrshrn.u16 d10, q9, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d11, q10, #7
- vqrshrn.u16 d12, q11, #7
- vqrshrn.u16 d13, q12, #7
- vqrshrn.u16 d14, q13, #7
- vqrshrn.u16 d15, q14, #7
- vqrshrn.u16 d16, q1, #7
- vqrshrn.u16 d17, q2, #7
- vqrshrn.u16 d18, q3, #7
- vqrshrn.u16 d19, q4, #7
-
- vst1.u8 {d10, d11, d12, d13}, [lr]! ;store result
- vst1.u8 {d14, d15, d16, d17}, [lr]!
- vst1.u8 {d18, d19}, [lr]!
-
-;Second pass: 16x16
-;secondpass_filter
- add r3, r12, r3, lsl #3
- sub lr, lr, #272
-
- vld1.u32 {d31}, [r3] ;load second_pass filter
-
- vld1.u8 {d22, d23}, [lr]! ;load src data
-
- vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
- vdup.8 d1, d31[4]
- mov r12, #4 ;loop counter
-
-filt_blk2d_sp16x16_loop_neon
- vld1.u8 {d24, d25}, [lr]!
- vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0])
- vld1.u8 {d26, d27}, [lr]!
- vmull.u8 q2, d23, d0
- vld1.u8 {d28, d29}, [lr]!
- vmull.u8 q3, d24, d0
- vld1.u8 {d30, d31}, [lr]!
-
- vmull.u8 q4, d25, d0
- vmull.u8 q5, d26, d0
- vmull.u8 q6, d27, d0
- vmull.u8 q7, d28, d0
- vmull.u8 q8, d29, d0
-
- vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * vp8_filter[1])
- vmlal.u8 q2, d25, d1
- vmlal.u8 q3, d26, d1
- vmlal.u8 q4, d27, d1
- vmlal.u8 q5, d28, d1
- vmlal.u8 q6, d29, d1
- vmlal.u8 q7, d30, d1
- vmlal.u8 q8, d31, d1
-
- subs r12, r12, #1
-
- vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d3, q2, #7
- vqrshrn.u16 d4, q3, #7
- vqrshrn.u16 d5, q4, #7
- vqrshrn.u16 d6, q5, #7
- vqrshrn.u16 d7, q6, #7
- vqrshrn.u16 d8, q7, #7
- vqrshrn.u16 d9, q8, #7
-
- vst1.u8 {d2, d3}, [r4], r5 ;store result
- vst1.u8 {d4, d5}, [r4], r5
- vst1.u8 {d6, d7}, [r4], r5
- vmov q11, q15
- vst1.u8 {d8, d9}, [r4], r5
-
- bne filt_blk2d_sp16x16_loop_neon
-
- add sp, sp, #272
-
- pop {r4-r5,pc}
-
-;--------------------
-firstpass_bfilter16x16_only
- mov r2, #4 ;loop counter
- vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
- vdup.8 d1, d31[4]
-
-;First Pass: output_height lines x output_width columns (16x16)
-filt_blk2d_fpo16x16_loop_neon
- vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
- vld1.u8 {d5, d6, d7}, [r0], r1
- vld1.u8 {d8, d9, d10}, [r0], r1
- vld1.u8 {d11, d12, d13}, [r0], r1
-
- pld [r0]
- pld [r0, r1]
- pld [r0, r1, lsl #1]
-
- vmull.u8 q7, d2, d0 ;(src_ptr[0] * vp8_filter[0])
- vmull.u8 q8, d3, d0
- vmull.u8 q9, d5, d0
- vmull.u8 q10, d6, d0
- vmull.u8 q11, d8, d0
- vmull.u8 q12, d9, d0
- vmull.u8 q13, d11, d0
- vmull.u8 q14, d12, d0
-
- vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
- vext.8 d5, d5, d6, #1
- vext.8 d8, d8, d9, #1
- vext.8 d11, d11, d12, #1
-
- vmlal.u8 q7, d2, d1 ;(src_ptr[0] * vp8_filter[1])
- vmlal.u8 q9, d5, d1
- vmlal.u8 q11, d8, d1
- vmlal.u8 q13, d11, d1
-
- vext.8 d3, d3, d4, #1
- vext.8 d6, d6, d7, #1
- vext.8 d9, d9, d10, #1
- vext.8 d12, d12, d13, #1
-
- vmlal.u8 q8, d3, d1 ;(src_ptr[0] * vp8_filter[1])
- vmlal.u8 q10, d6, d1
- vmlal.u8 q12, d9, d1
- vmlal.u8 q14, d12, d1
-
- subs r2, r2, #1
-
- vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d15, q8, #7
- vqrshrn.u16 d16, q9, #7
- vqrshrn.u16 d17, q10, #7
- vqrshrn.u16 d18, q11, #7
- vqrshrn.u16 d19, q12, #7
- vqrshrn.u16 d20, q13, #7
- vst1.u8 {d14, d15}, [r4], r5 ;store result
- vqrshrn.u16 d21, q14, #7
-
- vst1.u8 {d16, d17}, [r4], r5
- vst1.u8 {d18, d19}, [r4], r5
- vst1.u8 {d20, d21}, [r4], r5
-
- bne filt_blk2d_fpo16x16_loop_neon
- pop {r4-r5,pc}
-
-;---------------------
-secondpass_bfilter16x16_only
-;Second pass: 16x16
-;secondpass_filter
- add r3, r12, r3, lsl #3
- mov r12, #4 ;loop counter
- vld1.u32 {d31}, [r3] ;load second_pass filter
- vld1.u8 {d22, d23}, [r0], r1 ;load src data
-
- vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
- vdup.8 d1, d31[4]
-
-filt_blk2d_spo16x16_loop_neon
- vld1.u8 {d24, d25}, [r0], r1
- vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0])
- vld1.u8 {d26, d27}, [r0], r1
- vmull.u8 q2, d23, d0
- vld1.u8 {d28, d29}, [r0], r1
- vmull.u8 q3, d24, d0
- vld1.u8 {d30, d31}, [r0], r1
-
- vmull.u8 q4, d25, d0
- vmull.u8 q5, d26, d0
- vmull.u8 q6, d27, d0
- vmull.u8 q7, d28, d0
- vmull.u8 q8, d29, d0
-
- vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * vp8_filter[1])
- vmlal.u8 q2, d25, d1
- vmlal.u8 q3, d26, d1
- vmlal.u8 q4, d27, d1
- vmlal.u8 q5, d28, d1
- vmlal.u8 q6, d29, d1
- vmlal.u8 q7, d30, d1
- vmlal.u8 q8, d31, d1
-
- vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d3, q2, #7
- vqrshrn.u16 d4, q3, #7
- vqrshrn.u16 d5, q4, #7
- vqrshrn.u16 d6, q5, #7
- vqrshrn.u16 d7, q6, #7
- vqrshrn.u16 d8, q7, #7
- vqrshrn.u16 d9, q8, #7
-
- vst1.u8 {d2, d3}, [r4], r5 ;store result
- subs r12, r12, #1
- vst1.u8 {d4, d5}, [r4], r5
- vmov q11, q15
- vst1.u8 {d6, d7}, [r4], r5
- vst1.u8 {d8, d9}, [r4], r5
-
- bne filt_blk2d_spo16x16_loop_neon
- pop {r4-r5,pc}
-
- ENDP
-
-;-----------------
-
-bifilter16_coeff
- DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
-
- END
diff --git a/vp8/common/arm/neon/bilinearpredict4x4_neon.asm b/vp8/common/arm/neon/bilinearpredict4x4_neon.asm
deleted file mode 100644
index 0ac62436f..000000000
--- a/vp8/common/arm/neon/bilinearpredict4x4_neon.asm
+++ /dev/null
@@ -1,130 +0,0 @@
-;
-; Copyright (c) 2010 The WebM 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_bilinear_predict4x4_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-; r0 unsigned char *src_ptr,
-; r1 int src_pixels_per_line,
-; r2 int xoffset,
-; r3 int yoffset,
-; r4 unsigned char *dst_ptr,
-; stack(lr) int dst_pitch
-
-|vp8_bilinear_predict4x4_neon| PROC
- push {r4, lr}
-
- adr r12, bifilter4_coeff
- ldr r4, [sp, #8] ;load parameters from stack
- ldr lr, [sp, #12] ;load parameters from stack
-
- cmp r2, #0 ;skip first_pass filter if xoffset=0
- beq skip_firstpass_filter
-
-;First pass: output_height lines x output_width columns (5x4)
- vld1.u8 {d2}, [r0], r1 ;load src data
- add r2, r12, r2, lsl #3 ;calculate Hfilter location (2coeffsx4bytes=8bytes)
-
- vld1.u8 {d3}, [r0], r1
- vld1.u32 {d31}, [r2] ;first_pass filter
-
- vld1.u8 {d4}, [r0], r1
- vdup.8 d0, d31[0] ;first_pass filter (d0-d1)
- vld1.u8 {d5}, [r0], r1
- vdup.8 d1, d31[4]
- vld1.u8 {d6}, [r0], r1
-
- vshr.u64 q4, q1, #8 ;construct src_ptr[1]
- vshr.u64 q5, q2, #8
- vshr.u64 d12, d6, #8
-
- vzip.32 d2, d3 ;put 2-line data in 1 register (src_ptr[0])
- vzip.32 d4, d5
- vzip.32 d8, d9 ;put 2-line data in 1 register (src_ptr[1])
- vzip.32 d10, d11
-
- vmull.u8 q7, d2, d0 ;(src_ptr[0] * vp8_filter[0])
- vmull.u8 q8, d4, d0
- vmull.u8 q9, d6, d0
-
- vmlal.u8 q7, d8, d1 ;(src_ptr[1] * vp8_filter[1])
- vmlal.u8 q8, d10, d1
- vmlal.u8 q9, d12, d1
-
- vqrshrn.u16 d28, q7, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d29, q8, #7
- vqrshrn.u16 d30, q9, #7
-
-;Second pass: 4x4
-secondpass_filter
- cmp r3, #0 ;skip second_pass filter if yoffset=0
- beq skip_secondpass_filter
-
- add r3, r12, r3, lsl #3 ;calculate Vfilter location
- vld1.u32 {d31}, [r3] ;load second_pass filter
-
- vdup.8 d0, d31[0] ;second_pass filter parameters (d0-d5)
- vdup.8 d1, d31[4]
-
- vmull.u8 q1, d28, d0
- vmull.u8 q2, d29, d0
-
- vext.8 d26, d28, d29, #4 ;construct src_ptr[pixel_step]
- vext.8 d27, d29, d30, #4
-
- vmlal.u8 q1, d26, d1
- vmlal.u8 q2, d27, d1
-
- add r0, r4, lr
- add r1, r0, lr
- add r2, r1, lr
-
- vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d3, q2, #7
-
- vst1.32 {d2[0]}, [r4] ;store result
- vst1.32 {d2[1]}, [r0]
- vst1.32 {d3[0]}, [r1]
- vst1.32 {d3[1]}, [r2]
-
- pop {r4, pc}
-
-;--------------------
-skip_firstpass_filter
-
- vld1.32 {d28[0]}, [r0], r1 ;load src data
- vld1.32 {d28[1]}, [r0], r1
- vld1.32 {d29[0]}, [r0], r1
- vld1.32 {d29[1]}, [r0], r1
- vld1.32 {d30[0]}, [r0], r1
-
- b secondpass_filter
-
-;---------------------
-skip_secondpass_filter
- vst1.32 {d28[0]}, [r4], lr ;store result
- vst1.32 {d28[1]}, [r4], lr
- vst1.32 {d29[0]}, [r4], lr
- vst1.32 {d29[1]}, [r4], lr
-
- pop {r4, pc}
-
- ENDP
-
-;-----------------
-
-bifilter4_coeff
- DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
-
- END
diff --git a/vp8/common/arm/neon/bilinearpredict8x4_neon.asm b/vp8/common/arm/neon/bilinearpredict8x4_neon.asm
deleted file mode 100644
index 41f5c45ff..000000000
--- a/vp8/common/arm/neon/bilinearpredict8x4_neon.asm
+++ /dev/null
@@ -1,135 +0,0 @@
-;
-; Copyright (c) 2010 The WebM 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_bilinear_predict8x4_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-; r0 unsigned char *src_ptr,
-; r1 int src_pixels_per_line,
-; r2 int xoffset,
-; r3 int yoffset,
-; r4 unsigned char *dst_ptr,
-; stack(lr) int dst_pitch
-
-|vp8_bilinear_predict8x4_neon| PROC
- push {r4, lr}
-
- adr r12, bifilter8x4_coeff
- ldr r4, [sp, #8] ;load parameters from stack
- ldr lr, [sp, #12] ;load parameters from stack
-
- cmp r2, #0 ;skip first_pass filter if xoffset=0
- beq skip_firstpass_filter
-
-;First pass: output_height lines x output_width columns (5x8)
- add r2, r12, r2, lsl #3 ;calculate filter location
-
- vld1.u8 {q1}, [r0], r1 ;load src data
- vld1.u32 {d31}, [r2] ;load first_pass filter
- vld1.u8 {q2}, [r0], r1
- vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
- vld1.u8 {q3}, [r0], r1
- vdup.8 d1, d31[4]
- vld1.u8 {q4}, [r0], r1
-
- vmull.u8 q6, d2, d0 ;(src_ptr[0] * vp8_filter[0])
- vld1.u8 {q5}, [r0], r1
- vmull.u8 q7, d4, d0
- vmull.u8 q8, d6, d0
- vmull.u8 q9, d8, d0
- vmull.u8 q10, d10, d0
-
- vext.8 d3, d2, d3, #1 ;construct src_ptr[-1]
- vext.8 d5, d4, d5, #1
- vext.8 d7, d6, d7, #1
- vext.8 d9, d8, d9, #1
- vext.8 d11, d10, d11, #1
-
- vmlal.u8 q6, d3, d1 ;(src_ptr[1] * vp8_filter[1])
- vmlal.u8 q7, d5, d1
- vmlal.u8 q8, d7, d1
- vmlal.u8 q9, d9, d1
- vmlal.u8 q10, d11, d1
-
- vqrshrn.u16 d22, q6, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d23, q7, #7
- vqrshrn.u16 d24, q8, #7
- vqrshrn.u16 d25, q9, #7
- vqrshrn.u16 d26, q10, #7
-
-;Second pass: 4x8
-secondpass_filter
- cmp r3, #0 ;skip second_pass filter if yoffset=0
- beq skip_secondpass_filter
-
- add r3, r12, r3, lsl #3
- add r0, r4, lr
-
- vld1.u32 {d31}, [r3] ;load second_pass filter
- add r1, r0, lr
-
- vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
- vdup.8 d1, d31[4]
-
- vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0])
- vmull.u8 q2, d23, d0
- vmull.u8 q3, d24, d0
- vmull.u8 q4, d25, d0
-
- vmlal.u8 q1, d23, d1 ;(src_ptr[pixel_step] * vp8_filter[1])
- vmlal.u8 q2, d24, d1
- vmlal.u8 q3, d25, d1
- vmlal.u8 q4, d26, d1
-
- add r2, r1, lr
-
- vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d3, q2, #7
- vqrshrn.u16 d4, q3, #7
- vqrshrn.u16 d5, q4, #7
-
- vst1.u8 {d2}, [r4] ;store result
- vst1.u8 {d3}, [r0]
- vst1.u8 {d4}, [r1]
- vst1.u8 {d5}, [r2]
-
- pop {r4, pc}
-
-;--------------------
-skip_firstpass_filter
- vld1.u8 {d22}, [r0], r1 ;load src data
- vld1.u8 {d23}, [r0], r1
- vld1.u8 {d24}, [r0], r1
- vld1.u8 {d25}, [r0], r1
- vld1.u8 {d26}, [r0], r1
-
- b secondpass_filter
-
-;---------------------
-skip_secondpass_filter
- vst1.u8 {d22}, [r4], lr ;store result
- vst1.u8 {d23}, [r4], lr
- vst1.u8 {d24}, [r4], lr
- vst1.u8 {d25}, [r4], lr
-
- pop {r4, pc}
-
- ENDP
-
-;-----------------
-
-bifilter8x4_coeff
- DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
-
- END
diff --git a/vp8/common/arm/neon/bilinearpredict8x8_neon.asm b/vp8/common/arm/neon/bilinearpredict8x8_neon.asm
deleted file mode 100644
index c4711bc4d..000000000
--- a/vp8/common/arm/neon/bilinearpredict8x8_neon.asm
+++ /dev/null
@@ -1,183 +0,0 @@
-;
-; Copyright (c) 2010 The WebM 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_bilinear_predict8x8_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-; r0 unsigned char *src_ptr,
-; r1 int src_pixels_per_line,
-; r2 int xoffset,
-; r3 int yoffset,
-; r4 unsigned char *dst_ptr,
-; stack(lr) int dst_pitch
-
-|vp8_bilinear_predict8x8_neon| PROC
- push {r4, lr}
-
- adr r12, bifilter8_coeff
- ldr r4, [sp, #8] ;load parameters from stack
- ldr lr, [sp, #12] ;load parameters from stack
-
- cmp r2, #0 ;skip first_pass filter if xoffset=0
- beq skip_firstpass_filter
-
-;First pass: output_height lines x output_width columns (9x8)
- add r2, r12, r2, lsl #3 ;calculate filter location
-
- vld1.u8 {q1}, [r0], r1 ;load src data
- vld1.u32 {d31}, [r2] ;load first_pass filter
- vld1.u8 {q2}, [r0], r1
- vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
- vld1.u8 {q3}, [r0], r1
- vdup.8 d1, d31[4]
- vld1.u8 {q4}, [r0], r1
-
- vmull.u8 q6, d2, d0 ;(src_ptr[0] * vp8_filter[0])
- vmull.u8 q7, d4, d0
- vmull.u8 q8, d6, d0
- vmull.u8 q9, d8, d0
-
- vext.8 d3, d2, d3, #1 ;construct src_ptr[-1]
- vext.8 d5, d4, d5, #1
- vext.8 d7, d6, d7, #1
- vext.8 d9, d8, d9, #1
-
- vmlal.u8 q6, d3, d1 ;(src_ptr[1] * vp8_filter[1])
- vmlal.u8 q7, d5, d1
- vmlal.u8 q8, d7, d1
- vmlal.u8 q9, d9, d1
-
- vld1.u8 {q1}, [r0], r1 ;load src data
- vqrshrn.u16 d22, q6, #7 ;shift/round/saturate to u8
- vld1.u8 {q2}, [r0], r1
- vqrshrn.u16 d23, q7, #7
- vld1.u8 {q3}, [r0], r1
- vqrshrn.u16 d24, q8, #7
- vld1.u8 {q4}, [r0], r1
- vqrshrn.u16 d25, q9, #7
-
- ;first_pass filtering on the rest 5-line data
- vld1.u8 {q5}, [r0], r1
-
- vmull.u8 q6, d2, d0 ;(src_ptr[0] * vp8_filter[0])
- vmull.u8 q7, d4, d0
- vmull.u8 q8, d6, d0
- vmull.u8 q9, d8, d0
- vmull.u8 q10, d10, d0
-
- vext.8 d3, d2, d3, #1 ;construct src_ptr[-1]
- vext.8 d5, d4, d5, #1
- vext.8 d7, d6, d7, #1
- vext.8 d9, d8, d9, #1
- vext.8 d11, d10, d11, #1
-
- vmlal.u8 q6, d3, d1 ;(src_ptr[1] * vp8_filter[1])
- vmlal.u8 q7, d5, d1
- vmlal.u8 q8, d7, d1
- vmlal.u8 q9, d9, d1
- vmlal.u8 q10, d11, d1
-
- vqrshrn.u16 d26, q6, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d27, q7, #7
- vqrshrn.u16 d28, q8, #7
- vqrshrn.u16 d29, q9, #7
- vqrshrn.u16 d30, q10, #7
-
-;Second pass: 8x8
-secondpass_filter
- cmp r3, #0 ;skip second_pass filter if yoffset=0
- beq skip_secondpass_filter
-
- add r3, r12, r3, lsl #3
- add r0, r4, lr
-
- vld1.u32 {d31}, [r3] ;load second_pass filter
- add r1, r0, lr
-
- vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
- vdup.8 d1, d31[4]
-
- vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0])
- vmull.u8 q2, d23, d0
- vmull.u8 q3, d24, d0
- vmull.u8 q4, d25, d0
- vmull.u8 q5, d26, d0
- vmull.u8 q6, d27, d0
- vmull.u8 q7, d28, d0
- vmull.u8 q8, d29, d0
-
- vmlal.u8 q1, d23, d1 ;(src_ptr[pixel_step] * vp8_filter[1])
- vmlal.u8 q2, d24, d1
- vmlal.u8 q3, d25, d1
- vmlal.u8 q4, d26, d1
- vmlal.u8 q5, d27, d1
- vmlal.u8 q6, d28, d1
- vmlal.u8 q7, d29, d1
- vmlal.u8 q8, d30, d1
-
- vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
- vqrshrn.u16 d3, q2, #7
- vqrshrn.u16 d4, q3, #7
- vqrshrn.u16 d5, q4, #7
- vqrshrn.u16 d6, q5, #7
- vqrshrn.u16 d7, q6, #7
- vqrshrn.u16 d8, q7, #7
- vqrshrn.u16 d9, q8, #7
-
- vst1.u8 {d2}, [r4] ;store result
- vst1.u8 {d3}, [r0]
- vst1.u8 {d4}, [r1], lr
- vst1.u8 {d5}, [r1], lr
- vst1.u8 {d6}, [r1], lr
- vst1.u8 {d7}, [r1], lr
- vst1.u8 {d8}, [r1], lr
- vst1.u8 {d9}, [r1], lr
-
- pop {r4, pc}
-
-;--------------------
-skip_firstpass_filter
- vld1.u8 {d22}, [r0], r1 ;load src data
- vld1.u8 {d23}, [r0], r1
- vld1.u8 {d24}, [r0], r1
- vld1.u8 {d25}, [r0], r1
- vld1.u8 {d26}, [r0], r1
- vld1.u8 {d27}, [r0], r1
- vld1.u8 {d28}, [r0], r1
- vld1.u8 {d29}, [r0], r1
- vld1.u8 {d30}, [r0], r1
-
- b secondpass_filter
-
-;---------------------
-skip_secondpass_filter
- vst1.u8 {d22}, [r4], lr ;store result
- vst1.u8 {d23}, [r4], lr
- vst1.u8 {d24}, [r4], lr
- vst1.u8 {d25}, [r4], lr
- vst1.u8 {d26}, [r4], lr
- vst1.u8 {d27}, [r4], lr
- vst1.u8 {d28}, [r4], lr
- vst1.u8 {d29}, [r4], lr
-
- pop {r4, pc}
-
- ENDP
-
-;-----------------
-
-bifilter8_coeff
- DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
-
- END
diff --git a/vp8/common/arm/neon/bilinearpredict_neon_intrinsics.c b/vp8/common/arm/neon/bilinearpredict_neon_intrinsics.c
new file mode 100644
index 000000000..e1c3c2be7
--- /dev/null
+++ b/vp8/common/arm/neon/bilinearpredict_neon_intrinsics.c
@@ -0,0 +1,696 @@
+/*
+ * Copyright (c) 2014 The WebM 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.
+ */
+
+#include <arm_neon.h>
+
+static const uint16_t bifilter4_coeff[8][2] = {
+ {128, 0},
+ {112, 16},
+ { 96, 32},
+ { 80, 48},
+ { 64, 64},
+ { 48, 80},
+ { 32, 96},
+ { 16, 112}
+};
+
+void vp8_bilinear_predict4x4_neon(
+ unsigned char *src_ptr,
+ int src_pixels_per_line,
+ int xoffset,
+ int yoffset,
+ unsigned char *dst_ptr,
+ int dst_pitch) {
+ uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8;
+ uint8x8_t d26u8, d27u8, d28u8, d29u8, d30u8;
+ uint32x2_t d28u32, d29u32, d30u32;
+ uint8x16_t q1u8, q2u8;
+ uint16x8_t q1u16, q2u16;
+ uint16x8_t q7u16, q8u16, q9u16;
+ uint64x2_t q4u64, q5u64;
+ uint64x1_t d12u64;
+ uint32x2x2_t d0u32x2, d1u32x2, d2u32x2, d3u32x2;
+
+ if (xoffset == 0) { // skip_1stpass_filter
+ d28u32 = vld1_lane_u32((const uint32_t *)src_ptr, d28u32, 0);
+ src_ptr += src_pixels_per_line;
+ d28u32 = vld1_lane_u32((const uint32_t *)src_ptr, d28u32, 1);
+ src_ptr += src_pixels_per_line;
+ d29u32 = vld1_lane_u32((const uint32_t *)src_ptr, d29u32, 0);
+ src_ptr += src_pixels_per_line;
+ d29u32 = vld1_lane_u32((const uint32_t *)src_ptr, d29u32, 1);
+ src_ptr += src_pixels_per_line;
+ d30u32 = vld1_lane_u32((const uint32_t *)src_ptr, d30u32, 0);
+ d28u8 = vreinterpret_u8_u32(d28u32);
+ d29u8 = vreinterpret_u8_u32(d29u32);
+ d30u8 = vreinterpret_u8_u32(d30u32);
+ } else {
+ d2u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d3u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d4u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d5u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d6u8 = vld1_u8(src_ptr);
+
+ q1u8 = vcombine_u8(d2u8, d3u8);
+ q2u8 = vcombine_u8(d4u8, d5u8);
+
+ d0u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][0]);
+ d1u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][1]);
+
+ q4u64 = vshrq_n_u64(vreinterpretq_u64_u8(q1u8), 8);
+ q5u64 = vshrq_n_u64(vreinterpretq_u64_u8(q2u8), 8);
+ d12u64 = vshr_n_u64(vreinterpret_u64_u8(d6u8), 8);
+
+ d0u32x2 = vzip_u32(vreinterpret_u32_u8(vget_low_u8(q1u8)),
+ vreinterpret_u32_u8(vget_high_u8(q1u8)));
+ d1u32x2 = vzip_u32(vreinterpret_u32_u8(vget_low_u8(q2u8)),
+ vreinterpret_u32_u8(vget_high_u8(q2u8)));
+ d2u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q4u64)),
+ vreinterpret_u32_u64(vget_high_u64(q4u64)));
+ d3u32x2 = vzip_u32(vreinterpret_u32_u64(vget_low_u64(q5u64)),
+ vreinterpret_u32_u64(vget_high_u64(q5u64)));
+
+ q7u16 = vmull_u8(vreinterpret_u8_u32(d0u32x2.val[0]), d0u8);
+ q8u16 = vmull_u8(vreinterpret_u8_u32(d1u32x2.val[0]), d0u8);
+ q9u16 = vmull_u8(d6u8, d0u8);
+
+ q7u16 = vmlal_u8(q7u16, vreinterpret_u8_u32(d2u32x2.val[0]), d1u8);
+ q8u16 = vmlal_u8(q8u16, vreinterpret_u8_u32(d3u32x2.val[0]), d1u8);
+ q9u16 = vmlal_u8(q9u16, vreinterpret_u8_u64(d12u64), d1u8);
+
+ d28u8 = vqrshrn_n_u16(q7u16, 7);
+ d29u8 = vqrshrn_n_u16(q8u16, 7);
+ d30u8 = vqrshrn_n_u16(q9u16, 7);
+ }
+
+ // secondpass_filter
+ if (yoffset == 0) { // skip_2ndpass_filter
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d28u8), 0);
+ dst_ptr += dst_pitch;
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d28u8), 1);
+ dst_ptr += dst_pitch;
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d29u8), 0);
+ dst_ptr += dst_pitch;
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d29u8), 1);
+ } else {
+ d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]);
+ d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]);
+
+ q1u16 = vmull_u8(d28u8, d0u8);
+ q2u16 = vmull_u8(d29u8, d0u8);
+
+ d26u8 = vext_u8(d28u8, d29u8, 4);
+ d27u8 = vext_u8(d29u8, d30u8, 4);
+
+ q1u16 = vmlal_u8(q1u16, d26u8, d1u8);
+ q2u16 = vmlal_u8(q2u16, d27u8, d1u8);
+
+ d2u8 = vqrshrn_n_u16(q1u16, 7);
+ d3u8 = vqrshrn_n_u16(q2u16, 7);
+
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d2u8), 0);
+ dst_ptr += dst_pitch;
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d2u8), 1);
+ dst_ptr += dst_pitch;
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d3u8), 0);
+ dst_ptr += dst_pitch;
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d3u8), 1);
+ }
+ return;
+}
+
+void vp8_bilinear_predict8x4_neon(
+ unsigned char *src_ptr,
+ int src_pixels_per_line,
+ int xoffset,
+ int yoffset,
+ unsigned char *dst_ptr,
+ int dst_pitch) {
+ uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8;
+ uint8x8_t d7u8, d9u8, d11u8, d22u8, d23u8, d24u8, d25u8, d26u8;
+ uint8x16_t q1u8, q2u8, q3u8, q4u8, q5u8;
+ uint16x8_t q1u16, q2u16, q3u16, q4u16;
+ uint16x8_t q6u16, q7u16, q8u16, q9u16, q10u16;
+
+ if (xoffset == 0) { // skip_1stpass_filter
+ d22u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d23u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d24u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d25u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d26u8 = vld1_u8(src_ptr);
+ } else {
+ q1u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q2u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q3u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q4u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q5u8 = vld1q_u8(src_ptr);
+
+ d0u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][0]);
+ d1u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][1]);
+
+ q6u16 = vmull_u8(vget_low_u8(q1u8), d0u8);
+ q7u16 = vmull_u8(vget_low_u8(q2u8), d0u8);
+ q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8);
+ q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8);
+ q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8);
+
+ d3u8 = vext_u8(vget_low_u8(q1u8), vget_high_u8(q1u8), 1);
+ d5u8 = vext_u8(vget_low_u8(q2u8), vget_high_u8(q2u8), 1);
+ d7u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1);
+ d9u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1);
+ d11u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1);
+
+ q6u16 = vmlal_u8(q6u16, d3u8, d1u8);
+ q7u16 = vmlal_u8(q7u16, d5u8, d1u8);
+ q8u16 = vmlal_u8(q8u16, d7u8, d1u8);
+ q9u16 = vmlal_u8(q9u16, d9u8, d1u8);
+ q10u16 = vmlal_u8(q10u16, d11u8, d1u8);
+
+ d22u8 = vqrshrn_n_u16(q6u16, 7);
+ d23u8 = vqrshrn_n_u16(q7u16, 7);
+ d24u8 = vqrshrn_n_u16(q8u16, 7);
+ d25u8 = vqrshrn_n_u16(q9u16, 7);
+ d26u8 = vqrshrn_n_u16(q10u16, 7);
+ }
+
+ // secondpass_filter
+ if (yoffset == 0) { // skip_2ndpass_filter
+ vst1_u8((uint8_t *)dst_ptr, d22u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d23u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d24u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d25u8);
+ } else {
+ d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]);
+ d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]);
+
+ q1u16 = vmull_u8(d22u8, d0u8);
+ q2u16 = vmull_u8(d23u8, d0u8);
+ q3u16 = vmull_u8(d24u8, d0u8);
+ q4u16 = vmull_u8(d25u8, d0u8);
+
+ q1u16 = vmlal_u8(q1u16, d23u8, d1u8);
+ q2u16 = vmlal_u8(q2u16, d24u8, d1u8);
+ q3u16 = vmlal_u8(q3u16, d25u8, d1u8);
+ q4u16 = vmlal_u8(q4u16, d26u8, d1u8);
+
+ d2u8 = vqrshrn_n_u16(q1u16, 7);
+ d3u8 = vqrshrn_n_u16(q2u16, 7);
+ d4u8 = vqrshrn_n_u16(q3u16, 7);
+ d5u8 = vqrshrn_n_u16(q4u16, 7);
+
+ vst1_u8((uint8_t *)dst_ptr, d2u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d3u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d4u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d5u8);
+ }
+ return;
+}
+
+void vp8_bilinear_predict8x8_neon(
+ unsigned char *src_ptr,
+ int src_pixels_per_line,
+ int xoffset,
+ int yoffset,
+ unsigned char *dst_ptr,
+ int dst_pitch) {
+ uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8, d11u8;
+ uint8x8_t d22u8, d23u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8, d30u8;
+ uint8x16_t q1u8, q2u8, q3u8, q4u8, q5u8;
+ uint16x8_t q1u16, q2u16, q3u16, q4u16, q5u16;
+ uint16x8_t q6u16, q7u16, q8u16, q9u16, q10u16;
+
+ if (xoffset == 0) { // skip_1stpass_filter
+ d22u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d23u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d24u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d25u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d26u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d27u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d28u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d29u8 = vld1_u8(src_ptr); src_ptr += src_pixels_per_line;
+ d30u8 = vld1_u8(src_ptr);
+ } else {
+ q1u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q2u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q3u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q4u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+
+ d0u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][0]);
+ d1u8 = vdup_n_u8((uint8_t)bifilter4_coeff[xoffset][1]);
+
+ q6u16 = vmull_u8(vget_low_u8(q1u8), d0u8);
+ q7u16 = vmull_u8(vget_low_u8(q2u8), d0u8);
+ q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8);
+ q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8);
+
+ d3u8 = vext_u8(vget_low_u8(q1u8), vget_high_u8(q1u8), 1);
+ d5u8 = vext_u8(vget_low_u8(q2u8), vget_high_u8(q2u8), 1);
+ d7u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1);
+ d9u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1);
+
+ q6u16 = vmlal_u8(q6u16, d3u8, d1u8);
+ q7u16 = vmlal_u8(q7u16, d5u8, d1u8);
+ q8u16 = vmlal_u8(q8u16, d7u8, d1u8);
+ q9u16 = vmlal_u8(q9u16, d9u8, d1u8);
+
+ d22u8 = vqrshrn_n_u16(q6u16, 7);
+ d23u8 = vqrshrn_n_u16(q7u16, 7);
+ d24u8 = vqrshrn_n_u16(q8u16, 7);
+ d25u8 = vqrshrn_n_u16(q9u16, 7);
+
+ // first_pass filtering on the rest 5-line data
+ q1u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q2u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q3u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q4u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q5u8 = vld1q_u8(src_ptr);
+
+ q6u16 = vmull_u8(vget_low_u8(q1u8), d0u8);
+ q7u16 = vmull_u8(vget_low_u8(q2u8), d0u8);
+ q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8);
+ q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8);
+ q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8);
+
+ d3u8 = vext_u8(vget_low_u8(q1u8), vget_high_u8(q1u8), 1);
+ d5u8 = vext_u8(vget_low_u8(q2u8), vget_high_u8(q2u8), 1);
+ d7u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1);
+ d9u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1);
+ d11u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1);
+
+ q6u16 = vmlal_u8(q6u16, d3u8, d1u8);
+ q7u16 = vmlal_u8(q7u16, d5u8, d1u8);
+ q8u16 = vmlal_u8(q8u16, d7u8, d1u8);
+ q9u16 = vmlal_u8(q9u16, d9u8, d1u8);
+ q10u16 = vmlal_u8(q10u16, d11u8, d1u8);
+
+ d26u8 = vqrshrn_n_u16(q6u16, 7);
+ d27u8 = vqrshrn_n_u16(q7u16, 7);
+ d28u8 = vqrshrn_n_u16(q8u16, 7);
+ d29u8 = vqrshrn_n_u16(q9u16, 7);
+ d30u8 = vqrshrn_n_u16(q10u16, 7);
+ }
+
+ // secondpass_filter
+ if (yoffset == 0) { // skip_2ndpass_filter
+ vst1_u8((uint8_t *)dst_ptr, d22u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d23u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d24u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d25u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d26u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d27u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d28u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d29u8);
+ } else {
+ d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]);
+ d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]);
+
+ q1u16 = vmull_u8(d22u8, d0u8);
+ q2u16 = vmull_u8(d23u8, d0u8);
+ q3u16 = vmull_u8(d24u8, d0u8);
+ q4u16 = vmull_u8(d25u8, d0u8);
+ q5u16 = vmull_u8(d26u8, d0u8);
+ q6u16 = vmull_u8(d27u8, d0u8);
+ q7u16 = vmull_u8(d28u8, d0u8);
+ q8u16 = vmull_u8(d29u8, d0u8);
+
+ q1u16 = vmlal_u8(q1u16, d23u8, d1u8);
+ q2u16 = vmlal_u8(q2u16, d24u8, d1u8);
+ q3u16 = vmlal_u8(q3u16, d25u8, d1u8);
+ q4u16 = vmlal_u8(q4u16, d26u8, d1u8);
+ q5u16 = vmlal_u8(q5u16, d27u8, d1u8);
+ q6u16 = vmlal_u8(q6u16, d28u8, d1u8);
+ q7u16 = vmlal_u8(q7u16, d29u8, d1u8);
+ q8u16 = vmlal_u8(q8u16, d30u8, d1u8);
+
+ d2u8 = vqrshrn_n_u16(q1u16, 7);
+ d3u8 = vqrshrn_n_u16(q2u16, 7);
+ d4u8 = vqrshrn_n_u16(q3u16, 7);
+ d5u8 = vqrshrn_n_u16(q4u16, 7);
+ d6u8 = vqrshrn_n_u16(q5u16, 7);
+ d7u8 = vqrshrn_n_u16(q6u16, 7);
+ d8u8 = vqrshrn_n_u16(q7u16, 7);
+ d9u8 = vqrshrn_n_u16(q8u16, 7);
+
+ vst1_u8((uint8_t *)dst_ptr, d2u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d3u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d4u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d5u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d6u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d7u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d8u8); dst_ptr += dst_pitch;
+ vst1_u8((uint8_t *)dst_ptr, d9u8);
+ }
+ return;
+}
+
+void vp8_bilinear_predict16x16_neon(
+ unsigned char *src_ptr,
+ int src_pixels_per_line,
+ int xoffset,
+ int yoffset,
+ unsigned char *dst_ptr,
+ int dst_pitch) {
+ int i;
+ unsigned char tmp[272];
+ unsigned char *tmpp;
+ uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8;
+ uint8x8_t d10u8, d11u8, d12u8, d13u8, d14u8, d15u8, d16u8, d17u8, d18u8;
+ uint8x8_t d19u8, d20u8, d21u8;
+ uint8x16_t q1u8, q2u8, q3u8, q4u8, q5u8, q6u8, q7u8, q8u8, q9u8, q10u8;
+ uint8x16_t q11u8, q12u8, q13u8, q14u8, q15u8;
+ uint16x8_t q1u16, q2u16, q3u16, q4u16, q5u16, q6u16, q7u16, q8u16;
+ uint16x8_t q9u16, q10u16, q11u16, q12u16, q13u16, q14u16;
+
+ if (xoffset == 0) { // secondpass_bfilter16x16_only
+ d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]);
+ d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]);
+
+ q11u8 = vld1q_u8(src_ptr);
+ src_ptr += src_pixels_per_line;
+ for (i = 4; i > 0; i--) {
+ q12u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q13u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q14u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+ q15u8 = vld1q_u8(src_ptr); src_ptr += src_pixels_per_line;
+
+ q1u16 = vmull_u8(vget_low_u8(q11u8), d0u8);
+ q2u16 = vmull_u8(vget_high_u8(q11u8), d0u8);
+ q3u16 = vmull_u8(vget_low_u8(q12u8), d0u8);
+ q4u16 = vmull_u8(vget_high_u8(q12u8), d0u8);
+ q5u16 = vmull_u8(vget_low_u8(q13u8), d0u8);
+ q6u16 = vmull_u8(vget_high_u8(q13u8), d0u8);
+ q7u16 = vmull_u8(vget_low_u8(q14u8), d0u8);
+ q8u16 = vmull_u8(vget_high_u8(q14u8), d0u8);
+
+ q1u16 = vmlal_u8(q1u16, vget_low_u8(q12u8), d1u8);
+ q2u16 = vmlal_u8(q2u16, vget_high_u8(q12u8), d1u8);
+ q3u16 = vmlal_u8(q3u16, vget_low_u8(q13u8), d1u8);
+ q4u16 = vmlal_u8(q4u16, vget_high_u8(q13u8), d1u8);
+ q5u16 = vmlal_u8(q5u16, vget_low_u8(q14u8), d1u8);
+ q6u16 = vmlal_u8(q6u16, vget_high_u8(q14u8), d1u8);
+ q7u16 = vmlal_u8(q7u16, vget_low_u8(q15u8), d1u8);
+ q8u16 = vmlal_u8(q8u16, vget_high_u8(q15u8), d1u8);
+
+ d2u8 = vqrshrn_n_u16(q1u16, 7);
+ d3u8 = vqrshrn_n_u16(q2u16, 7);
+ d4u8 = vqrshrn_n_u16(q3u16, 7);
+ d5u8 = vqrshrn_n_u16(q4u16, 7);
+ d6u8 = vqrshrn_n_u16(q5u16, 7);
+ d7u8 = vqrshrn_n_u16(q6u16, 7);
+ d8u8 = vqrshrn_n_u16(q7u16, 7);
+ d9u8 = vqrshrn_n_u16(q8u16, 7);
+
+ q1u8 = vcombine_u8(d2u8, d3u8);
+ q2u8 = vcombine_u8(d4u8, d5u8);
+ q3u8 = vcombine_u8(d6u8, d7u8);
+ q4u8 = vcombine_u8(d8u8, d9u8);
+
+ q11u8 = q15u8;
+
+ vst1q_u8((uint8_t *)dst_ptr, q1u8); dst_ptr += dst_pitch;
+ vst1q_u8((uint8_t *)dst_ptr, q2u8); dst_ptr += dst_pitch;
+ vst1q_u8((uint8_t *)dst_ptr, q3u8); dst_ptr += dst_pitch;
+ vst1q_u8((uint8_t *)dst_ptr, q4u8); dst_ptr += dst_pitch;
+ }
+ return;
+ }
+
+ if (yoffset == 0) { // firstpass_bfilter16x16_only
+ d0u8 = vdup_n_u8(bifilter4_coeff[xoffset][0]);
+ d1u8 = vdup_n_u8(bifilter4_coeff[xoffset][1]);
+
+ for (i = 4; i > 0 ; i--) {
+ d2u8 = vld1_u8(src_ptr);
+ d3u8 = vld1_u8(src_ptr + 8);
+ d4u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+ d5u8 = vld1_u8(src_ptr);
+ d6u8 = vld1_u8(src_ptr + 8);
+ d7u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+ d8u8 = vld1_u8(src_ptr);
+ d9u8 = vld1_u8(src_ptr + 8);
+ d10u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+ d11u8 = vld1_u8(src_ptr);
+ d12u8 = vld1_u8(src_ptr + 8);
+ d13u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+
+ q7u16 = vmull_u8(d2u8, d0u8);
+ q8u16 = vmull_u8(d3u8, d0u8);
+ q9u16 = vmull_u8(d5u8, d0u8);
+ q10u16 = vmull_u8(d6u8, d0u8);
+ q11u16 = vmull_u8(d8u8, d0u8);
+ q12u16 = vmull_u8(d9u8, d0u8);
+ q13u16 = vmull_u8(d11u8, d0u8);
+ q14u16 = vmull_u8(d12u8, d0u8);
+
+ d2u8 = vext_u8(d2u8, d3u8, 1);
+ d5u8 = vext_u8(d5u8, d6u8, 1);
+ d8u8 = vext_u8(d8u8, d9u8, 1);
+ d11u8 = vext_u8(d11u8, d12u8, 1);
+
+ q7u16 = vmlal_u8(q7u16, d2u8, d1u8);
+ q9u16 = vmlal_u8(q9u16, d5u8, d1u8);
+ q11u16 = vmlal_u8(q11u16, d8u8, d1u8);
+ q13u16 = vmlal_u8(q13u16, d11u8, d1u8);
+
+ d3u8 = vext_u8(d3u8, d4u8, 1);
+ d6u8 = vext_u8(d6u8, d7u8, 1);
+ d9u8 = vext_u8(d9u8, d10u8, 1);
+ d12u8 = vext_u8(d12u8, d13u8, 1);
+
+ q8u16 = vmlal_u8(q8u16, d3u8, d1u8);
+ q10u16 = vmlal_u8(q10u16, d6u8, d1u8);
+ q12u16 = vmlal_u8(q12u16, d9u8, d1u8);
+ q14u16 = vmlal_u8(q14u16, d12u8, d1u8);
+
+ d14u8 = vqrshrn_n_u16(q7u16, 7);
+ d15u8 = vqrshrn_n_u16(q8u16, 7);
+ d16u8 = vqrshrn_n_u16(q9u16, 7);
+ d17u8 = vqrshrn_n_u16(q10u16, 7);
+ d18u8 = vqrshrn_n_u16(q11u16, 7);
+ d19u8 = vqrshrn_n_u16(q12u16, 7);
+ d20u8 = vqrshrn_n_u16(q13u16, 7);
+ d21u8 = vqrshrn_n_u16(q14u16, 7);
+
+ q7u8 = vcombine_u8(d14u8, d15u8);
+ q8u8 = vcombine_u8(d16u8, d17u8);
+ q9u8 = vcombine_u8(d18u8, d19u8);
+ q10u8 =vcombine_u8(d20u8, d21u8);
+
+ vst1q_u8((uint8_t *)dst_ptr, q7u8); dst_ptr += dst_pitch;
+ vst1q_u8((uint8_t *)dst_ptr, q8u8); dst_ptr += dst_pitch;
+ vst1q_u8((uint8_t *)dst_ptr, q9u8); dst_ptr += dst_pitch;
+ vst1q_u8((uint8_t *)dst_ptr, q10u8); dst_ptr += dst_pitch;
+ }
+ return;
+ }
+
+ d0u8 = vdup_n_u8(bifilter4_coeff[xoffset][0]);
+ d1u8 = vdup_n_u8(bifilter4_coeff[xoffset][1]);
+
+ d2u8 = vld1_u8(src_ptr);
+ d3u8 = vld1_u8(src_ptr + 8);
+ d4u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+ d5u8 = vld1_u8(src_ptr);
+ d6u8 = vld1_u8(src_ptr + 8);
+ d7u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+ d8u8 = vld1_u8(src_ptr);
+ d9u8 = vld1_u8(src_ptr + 8);
+ d10u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+ d11u8 = vld1_u8(src_ptr);
+ d12u8 = vld1_u8(src_ptr + 8);
+ d13u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+
+ // First Pass: output_height lines x output_width columns (17x16)
+ tmpp = tmp;
+ for (i = 3; i > 0; i--) {
+ q7u16 = vmull_u8(d2u8, d0u8);
+ q8u16 = vmull_u8(d3u8, d0u8);
+ q9u16 = vmull_u8(d5u8, d0u8);
+ q10u16 = vmull_u8(d6u8, d0u8);
+ q11u16 = vmull_u8(d8u8, d0u8);
+ q12u16 = vmull_u8(d9u8, d0u8);
+ q13u16 = vmull_u8(d11u8, d0u8);
+ q14u16 = vmull_u8(d12u8, d0u8);
+
+ d2u8 = vext_u8(d2u8, d3u8, 1);
+ d5u8 = vext_u8(d5u8, d6u8, 1);
+ d8u8 = vext_u8(d8u8, d9u8, 1);
+ d11u8 = vext_u8(d11u8, d12u8, 1);
+
+ q7u16 = vmlal_u8(q7u16, d2u8, d1u8);
+ q9u16 = vmlal_u8(q9u16, d5u8, d1u8);
+ q11u16 = vmlal_u8(q11u16, d8u8, d1u8);
+ q13u16 = vmlal_u8(q13u16, d11u8, d1u8);
+
+ d3u8 = vext_u8(d3u8, d4u8, 1);
+ d6u8 = vext_u8(d6u8, d7u8, 1);
+ d9u8 = vext_u8(d9u8, d10u8, 1);
+ d12u8 = vext_u8(d12u8, d13u8, 1);
+
+ q8u16 = vmlal_u8(q8u16, d3u8, d1u8);
+ q10u16 = vmlal_u8(q10u16, d6u8, d1u8);
+ q12u16 = vmlal_u8(q12u16, d9u8, d1u8);
+ q14u16 = vmlal_u8(q14u16, d12u8, d1u8);
+
+ d14u8 = vqrshrn_n_u16(q7u16, 7);
+ d15u8 = vqrshrn_n_u16(q8u16, 7);
+ d16u8 = vqrshrn_n_u16(q9u16, 7);
+ d17u8 = vqrshrn_n_u16(q10u16, 7);
+ d18u8 = vqrshrn_n_u16(q11u16, 7);
+ d19u8 = vqrshrn_n_u16(q12u16, 7);
+ d20u8 = vqrshrn_n_u16(q13u16, 7);
+ d21u8 = vqrshrn_n_u16(q14u16, 7);
+
+ d2u8 = vld1_u8(src_ptr);
+ d3u8 = vld1_u8(src_ptr + 8);
+ d4u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+ d5u8 = vld1_u8(src_ptr);
+ d6u8 = vld1_u8(src_ptr + 8);
+ d7u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+ d8u8 = vld1_u8(src_ptr);
+ d9u8 = vld1_u8(src_ptr + 8);
+ d10u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+ d11u8 = vld1_u8(src_ptr);
+ d12u8 = vld1_u8(src_ptr + 8);
+ d13u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+
+ q7u8 = vcombine_u8(d14u8, d15u8);
+ q8u8 = vcombine_u8(d16u8, d17u8);
+ q9u8 = vcombine_u8(d18u8, d19u8);
+ q10u8 = vcombine_u8(d20u8, d21u8);
+
+ vst1q_u8((uint8_t *)tmpp, q7u8); tmpp += 16;
+ vst1q_u8((uint8_t *)tmpp, q8u8); tmpp += 16;
+ vst1q_u8((uint8_t *)tmpp, q9u8); tmpp += 16;
+ vst1q_u8((uint8_t *)tmpp, q10u8); tmpp += 16;
+ }
+
+ // First-pass filtering for rest 5 lines
+ d14u8 = vld1_u8(src_ptr);
+ d15u8 = vld1_u8(src_ptr + 8);
+ d16u8 = vld1_u8(src_ptr + 16); src_ptr += src_pixels_per_line;
+
+ q9u16 = vmull_u8(d2u8, d0u8);
+ q10u16 = vmull_u8(d3u8, d0u8);
+ q11u16 = vmull_u8(d5u8, d0u8);
+ q12u16 = vmull_u8(d6u8, d0u8);
+ q13u16 = vmull_u8(d8u8, d0u8);
+ q14u16 = vmull_u8(d9u8, d0u8);
+
+ d2u8 = vext_u8(d2u8, d3u8, 1);
+ d5u8 = vext_u8(d5u8, d6u8, 1);
+ d8u8 = vext_u8(d8u8, d9u8, 1);
+
+ q9u16 = vmlal_u8(q9u16, d2u8, d1u8);
+ q11u16 = vmlal_u8(q11u16, d5u8, d1u8);
+ q13u16 = vmlal_u8(q13u16, d8u8, d1u8);
+
+ d3u8 = vext_u8(d3u8, d4u8, 1);
+ d6u8 = vext_u8(d6u8, d7u8, 1);
+ d9u8 = vext_u8(d9u8, d10u8, 1);
+
+ q10u16 = vmlal_u8(q10u16, d3u8, d1u8);
+ q12u16 = vmlal_u8(q12u16, d6u8, d1u8);
+ q14u16 = vmlal_u8(q14u16, d9u8, d1u8);
+
+ q1u16 = vmull_u8(d11u8, d0u8);
+ q2u16 = vmull_u8(d12u8, d0u8);
+ q3u16 = vmull_u8(d14u8, d0u8);
+ q4u16 = vmull_u8(d15u8, d0u8);
+
+ d11u8 = vext_u8(d11u8, d12u8, 1);
+ d14u8 = vext_u8(d14u8, d15u8, 1);
+
+ q1u16 = vmlal_u8(q1u16, d11u8, d1u8);
+ q3u16 = vmlal_u8(q3u16, d14u8, d1u8);
+
+ d12u8 = vext_u8(d12u8, d13u8, 1);
+ d15u8 = vext_u8(d15u8, d16u8, 1);
+
+ q2u16 = vmlal_u8(q2u16, d12u8, d1u8);
+ q4u16 = vmlal_u8(q4u16, d15u8, d1u8);
+
+ d10u8 = vqrshrn_n_u16(q9u16, 7);
+ d11u8 = vqrshrn_n_u16(q10u16, 7);
+ d12u8 = vqrshrn_n_u16(q11u16, 7);
+ d13u8 = vqrshrn_n_u16(q12u16, 7);
+ d14u8 = vqrshrn_n_u16(q13u16, 7);
+ d15u8 = vqrshrn_n_u16(q14u16, 7);
+ d16u8 = vqrshrn_n_u16(q1u16, 7);
+ d17u8 = vqrshrn_n_u16(q2u16, 7);
+ d18u8 = vqrshrn_n_u16(q3u16, 7);
+ d19u8 = vqrshrn_n_u16(q4u16, 7);
+
+ q5u8 = vcombine_u8(d10u8, d11u8);
+ q6u8 = vcombine_u8(d12u8, d13u8);
+ q7u8 = vcombine_u8(d14u8, d15u8);
+ q8u8 = vcombine_u8(d16u8, d17u8);
+ q9u8 = vcombine_u8(d18u8, d19u8);
+
+ vst1q_u8((uint8_t *)tmpp, q5u8); tmpp += 16;
+ vst1q_u8((uint8_t *)tmpp, q6u8); tmpp += 16;
+ vst1q_u8((uint8_t *)tmpp, q7u8); tmpp += 16;
+ vst1q_u8((uint8_t *)tmpp, q8u8); tmpp += 16;
+ vst1q_u8((uint8_t *)tmpp, q9u8);
+
+ // secondpass_filter
+ d0u8 = vdup_n_u8(bifilter4_coeff[yoffset][0]);
+ d1u8 = vdup_n_u8(bifilter4_coeff[yoffset][1]);
+
+ tmpp = tmp;
+ q11u8 = vld1q_u8(tmpp);
+ tmpp += 16;
+ for (i = 4; i > 0; i--) {
+ q12u8 = vld1q_u8(tmpp); tmpp += 16;
+ q13u8 = vld1q_u8(tmpp); tmpp += 16;
+ q14u8 = vld1q_u8(tmpp); tmpp += 16;
+ q15u8 = vld1q_u8(tmpp); tmpp += 16;
+
+ q1u16 = vmull_u8(vget_low_u8(q11u8), d0u8);
+ q2u16 = vmull_u8(vget_high_u8(q11u8), d0u8);
+ q3u16 = vmull_u8(vget_low_u8(q12u8), d0u8);
+ q4u16 = vmull_u8(vget_high_u8(q12u8), d0u8);
+ q5u16 = vmull_u8(vget_low_u8(q13u8), d0u8);
+ q6u16 = vmull_u8(vget_high_u8(q13u8), d0u8);
+ q7u16 = vmull_u8(vget_low_u8(q14u8), d0u8);
+ q8u16 = vmull_u8(vget_high_u8(q14u8), d0u8);
+
+ q1u16 = vmlal_u8(q1u16, vget_low_u8(q12u8), d1u8);
+ q2u16 = vmlal_u8(q2u16, vget_high_u8(q12u8), d1u8);
+ q3u16 = vmlal_u8(q3u16, vget_low_u8(q13u8), d1u8);
+ q4u16 = vmlal_u8(q4u16, vget_high_u8(q13u8), d1u8);
+ q5u16 = vmlal_u8(q5u16, vget_low_u8(q14u8), d1u8);
+ q6u16 = vmlal_u8(q6u16, vget_high_u8(q14u8), d1u8);
+ q7u16 = vmlal_u8(q7u16, vget_low_u8(q15u8), d1u8);
+ q8u16 = vmlal_u8(q8u16, vget_high_u8(q15u8), d1u8);
+
+ d2u8 = vqrshrn_n_u16(q1u16, 7);
+ d3u8 = vqrshrn_n_u16(q2u16, 7);
+ d4u8 = vqrshrn_n_u16(q3u16, 7);
+ d5u8 = vqrshrn_n_u16(q4u16, 7);
+ d6u8 = vqrshrn_n_u16(q5u16, 7);
+ d7u8 = vqrshrn_n_u16(q6u16, 7);
+ d8u8 = vqrshrn_n_u16(q7u16, 7);
+ d9u8 = vqrshrn_n_u16(q8u16, 7);
+
+ q1u8 = vcombine_u8(d2u8, d3u8);
+ q2u8 = vcombine_u8(d4u8, d5u8);
+ q3u8 = vcombine_u8(d6u8, d7u8);
+ q4u8 = vcombine_u8(d8u8, d9u8);
+
+ q11u8 = q15u8;
+
+ vst1q_u8((uint8_t *)dst_ptr, q1u8); dst_ptr += dst_pitch;
+ vst1q_u8((uint8_t *)dst_ptr, q2u8); dst_ptr += dst_pitch;
+ vst1q_u8((uint8_t *)dst_ptr, q3u8); dst_ptr += dst_pitch;
+ vst1q_u8((uint8_t *)dst_ptr, q4u8); dst_ptr += dst_pitch;
+ }
+ return;
+}
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-13 23:14:10 +0000