From 62c64118209fe4210e48cb5b37a59e97c61c9441 Mon Sep 17 00:00:00 2001 From: Johann Date: Thu, 24 Jul 2014 09:32:01 -0700 Subject: [PATCH] Remove neon version of vp8 extend borders The code fails the unit test. Speed comparisons to the C are invalid because the code frequently didn't correctly extend the right and bottom portions of the frame. Reduce maximum frame size on ARM devices to avoid OOM Change-Id: Ia664c86406f0bb8120fd7ad401f32d0bd44994fb --- test/scale_border_test.cc | 6 - .../neon/vp8_vpxyv12_extendframeborders_neon.asm | 308 --------------------- vpx_scale/arm/neon/yv12extend_arm.c | 2 +- vpx_scale/vpx_scale.mk | 1 - vpx_scale/vpx_scale_rtcd.pl | 2 - 5 files changed, 1 insertion(+), 318 deletions(-) delete mode 100644 vpx_scale/arm/neon/vp8_vpxyv12_extendframeborders_neon.asm diff --git a/test/scale_border_test.cc b/test/scale_border_test.cc index f31b831..cc9a69a 100644 --- a/test/scale_border_test.cc +++ b/test/scale_border_test.cc @@ -179,10 +179,4 @@ TEST_P(ExtendBorderTest, ExtendBorder) { INSTANTIATE_TEST_CASE_P(C, ExtendBorderTest, ::testing::Values(vp8_yv12_extend_frame_borders_c)); - -#if HAVE_NEON -INSTANTIATE_TEST_CASE_P(DISABLED_NEON, ExtendBorderTest, - ::testing::Values(vp8_yv12_extend_frame_borders_neon)); -#endif - } // namespace diff --git a/vpx_scale/arm/neon/vp8_vpxyv12_extendframeborders_neon.asm b/vpx_scale/arm/neon/vp8_vpxyv12_extendframeborders_neon.asm deleted file mode 100644 index b2eb9eb..0000000 --- a/vpx_scale/arm/neon/vp8_vpxyv12_extendframeborders_neon.asm +++ /dev/null @@ -1,308 +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_yv12_extend_frame_borders_neon| - ARM - REQUIRE8 - PRESERVE8 - - INCLUDE vpx_scale_asm_offsets.asm - - AREA ||.text||, CODE, READONLY, ALIGN=2 -;void vp8_yv12_extend_frame_borders_neon (YV12_BUFFER_CONFIG *ybf); -; we depend on VP8BORDERINPIXELS being 32 - -|vp8_yv12_extend_frame_borders_neon| PROC - push {r4 - r10, lr} - vpush {d8 - d15} - - ; Border = 32 - ldr r3, [r0, #yv12_buffer_config_y_width] ; plane_width - ldr r1, [r0, #yv12_buffer_config_y_buffer] ; src_ptr1 - ldr r4, [r0, #yv12_buffer_config_y_height] ; plane_height - ldr lr, [r0, #yv12_buffer_config_y_stride] ; plane_stride - -; Border copy for Y plane -; copy the left and right most columns out - add r6, r1, r3 ; dest_ptr2 = src_ptr2 + 1 (src_ptr1 + plane_width) - sub r2, r6, #1 ; src_ptr2 = src_ptr1 + plane_width - 1 - sub r5, r1, #32 ; dest_ptr1 = src_ptr1 - Border - - mov r12, r4, lsr #2 ; plane_height / 4 - -copy_left_right_y - vld1.8 {d0[], d1[]}, [r1], lr - vld1.8 {d4[], d5[]}, [r2], lr - vld1.8 {d8[], d9[]}, [r1], lr - vld1.8 {d12[], d13[]}, [r2], lr - vld1.8 {d16[], d17[]}, [r1], lr - vld1.8 {d20[], d21[]}, [r2], lr - vld1.8 {d24[], d25[]}, [r1], lr - vld1.8 {d28[], d29[]}, [r2], lr - - vmov q1, q0 - vmov q3, q2 - vmov q5, q4 - vmov q7, q6 - vmov q9, q8 - vmov q11, q10 - vmov q13, q12 - vmov q15, q14 - - subs r12, r12, #1 - - vst1.8 {q0, q1}, [r5], lr - vst1.8 {q2, q3}, [r6], lr - vst1.8 {q4, q5}, [r5], lr - vst1.8 {q6, q7}, [r6], lr - vst1.8 {q8, q9}, [r5], lr - vst1.8 {q10, q11}, [r6], lr - vst1.8 {q12, q13}, [r5], lr - vst1.8 {q14, q15}, [r6], lr - - bne copy_left_right_y - -;Now copy the top and bottom source lines into each line of the respective borders - ldr r1, [r0, #yv12_buffer_config_y_buffer] ; y_buffer - mul r8, r4, lr ; plane_height * plane_stride - - ; copy width is plane_stride - movs r12, lr, lsr #7 ; plane_stride / 128 - - sub r1, r1, #32 ; src_ptr1 = y_buffer - Border - add r6, r1, r8 ; dest_ptr2 = src_ptr2 - plane_stride (src_ptr1 + (plane_height * plane_stride)) - sub r2, r6, lr ; src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride - sub r5, r1, lr, asl #5 ; dest_ptr1 = src_ptr1 - (Border * plane_stride) - ble extra_y_copy_needed ; plane stride < 128 - -copy_top_bottom_y - vld1.8 {q0, q1}, [r1]! - vld1.8 {q8, q9}, [r2]! - vld1.8 {q2, q3}, [r1]! - vld1.8 {q10, q11}, [r2]! - vld1.8 {q4, q5}, [r1]! - vld1.8 {q12, q13}, [r2]! - vld1.8 {q6, q7}, [r1]! - vld1.8 {q14, q15}, [r2]! - - mov r7, #32 ; Border - -top_bottom_32 - subs r7, r7, #1 - - vst1.8 {q0, q1}, [r5]! - vst1.8 {q8, q9}, [r6]! - vst1.8 {q2, q3}, [r5]! - vst1.8 {q10, q11}, [r6]! - vst1.8 {q4, q5}, [r5]! - vst1.8 {q12, q13}, [r6]! - vst1.8 {q6, q7}, [r5]! - vst1.8 {q14, q15}, [r6]! - - add r5, r5, lr ; dest_ptr1 += plane_stride - sub r5, r5, #128 ; dest_ptr1 -= 128 - add r6, r6, lr ; dest_ptr2 += plane_stride - sub r6, r6, #128 ; dest_ptr2 -= 128 - - bne top_bottom_32 - - sub r5, r1, lr, asl #5 ; src_ptr1 - (Border* plane_stride) - add r6, r2, lr ; src_ptr2 + plane_stride - - subs r12, r12, #1 - bne copy_top_bottom_y - -extra_y_copy_needed - mov r7, lr, lsr #4 ; check to see if extra copy is needed - ands r7, r7, #0x7 - bne extra_top_bottom_y -end_of_border_copy_y - -;Border copy for U, V planes -; Border = 16 - ldr r7, [r0, #yv12_buffer_config_u_buffer] ; src_ptr1 - ldr lr, [r0, #yv12_buffer_config_uv_stride] ; plane_stride - ldr r3, [r0, #yv12_buffer_config_uv_width] ; plane_width - ldr r4, [r0, #yv12_buffer_config_uv_height] ; plane_height - - mov r10, #2 - -;copy the left and right most columns out -border_copy_uv - mov r1, r7 ; src_ptr1 needs to be saved for second half of loop - sub r5, r1, #16 ; dest_ptr1 = src_ptr1 - Border - add r6, r1, r3 ; dest_ptr2 = src_ptr2 + 1 (src_ptr1 + plane_width) - sub r2, r6, #1 ; src_ptr2 = src_ptr1 + plane_width - 1 - - mov r12, r4, lsr #3 ; plane_height / 8 - -copy_left_right_uv - vld1.8 {d0[], d1[]}, [r1], lr - vld1.8 {d2[], d3[]}, [r2], lr - vld1.8 {d4[], d5[]}, [r1], lr - vld1.8 {d6[], d7[]}, [r2], lr - vld1.8 {d8[], d9[]}, [r1], lr - vld1.8 {d10[], d11[]}, [r2], lr - vld1.8 {d12[], d13[]}, [r1], lr - vld1.8 {d14[], d15[]}, [r2], lr - vld1.8 {d16[], d17[]}, [r1], lr - vld1.8 {d18[], d19[]}, [r2], lr - vld1.8 {d20[], d21[]}, [r1], lr - vld1.8 {d22[], d23[]}, [r2], lr - vld1.8 {d24[], d25[]}, [r1], lr - vld1.8 {d26[], d27[]}, [r2], lr - vld1.8 {d28[], d29[]}, [r1], lr - vld1.8 {d30[], d31[]}, [r2], lr - - subs r12, r12, #1 - - vst1.8 {q0}, [r5], lr - vst1.8 {q1}, [r6], lr - vst1.8 {q2}, [r5], lr - vst1.8 {q3}, [r6], lr - vst1.8 {q4}, [r5], lr - vst1.8 {q5}, [r6], lr - vst1.8 {q6}, [r5], lr - vst1.8 {q7}, [r6], lr - vst1.8 {q8}, [r5], lr - vst1.8 {q9}, [r6], lr - vst1.8 {q10}, [r5], lr - vst1.8 {q11}, [r6], lr - vst1.8 {q12}, [r5], lr - vst1.8 {q13}, [r6], lr - vst1.8 {q14}, [r5], lr - vst1.8 {q15}, [r6], lr - - bne copy_left_right_uv - -;Now copy the top and bottom source lines into each line of the respective borders - mov r1, r7 - mul r8, r4, lr ; plane_height * plane_stride - movs r12, lr, lsr #6 ; plane_stride / 64 - - sub r1, r1, #16 ; src_ptr1 = u_buffer - Border - add r6, r1, r8 ; dest_ptr2 = src_ptr2 + plane_stride (src_ptr1 + (plane_height * plane_stride) - sub r2, r6, lr ; src_ptr2 = src_ptr1 + (plane_height * plane_stride) - plane_stride - sub r5, r1, lr, asl #4 ; dest_ptr1 = src_ptr1 - (Border * plane_stride) - ble extra_uv_copy_needed ; plane_stride < 64 - -copy_top_bottom_uv - vld1.8 {q0, q1}, [r1]! - vld1.8 {q8, q9}, [r2]! - vld1.8 {q2, q3}, [r1]! - vld1.8 {q10, q11}, [r2]! - - mov r7, #16 ; Border - -top_bottom_16 - subs r7, r7, #1 - - vst1.8 {q0, q1}, [r5]! - vst1.8 {q8, q9}, [r6]! - vst1.8 {q2, q3}, [r5]! - vst1.8 {q10, q11}, [r6]! - - add r5, r5, lr ; dest_ptr1 += plane_stride - sub r5, r5, #64 - add r6, r6, lr ; dest_ptr2 += plane_stride - sub r6, r6, #64 - - bne top_bottom_16 - - sub r5, r1, lr, asl #4 ; dest_ptr1 = src_ptr1 - (Border * plane_stride) - add r6, r2, lr ; dest_ptr2 = src_ptr2 + plane_stride - - subs r12, r12, #1 - bne copy_top_bottom_uv -extra_uv_copy_needed - mov r7, lr, lsr #3 ; check to see if extra copy is needed - ands r7, r7, #0x7 - bne extra_top_bottom_uv - -end_of_border_copy_uv - subs r10, r10, #1 - ldrne r7, [r0, #yv12_buffer_config_v_buffer] ; src_ptr1 - bne border_copy_uv - - vpop {d8 - d15} - pop {r4 - r10, pc} - -;;;;;;;;;;;;;;;;;;;;;; -extra_top_bottom_y - vld1.8 {q0}, [r1]! - vld1.8 {q2}, [r2]! - - mov r9, #4 ; 32 >> 3 - -extra_top_bottom_32 - subs r9, r9, #1 - - vst1.8 {q0}, [r5], lr - vst1.8 {q2}, [r6], lr - vst1.8 {q0}, [r5], lr - vst1.8 {q2}, [r6], lr - vst1.8 {q0}, [r5], lr - vst1.8 {q2}, [r6], lr - vst1.8 {q0}, [r5], lr - vst1.8 {q2}, [r6], lr - vst1.8 {q0}, [r5], lr - vst1.8 {q2}, [r6], lr - vst1.8 {q0}, [r5], lr - vst1.8 {q2}, [r6], lr - vst1.8 {q0}, [r5], lr - vst1.8 {q2}, [r6], lr - vst1.8 {q0}, [r5], lr - vst1.8 {q2}, [r6], lr - bne extra_top_bottom_32 - - sub r5, r1, lr, asl #5 ; src_ptr1 - (Border * plane_stride) - add r6, r2, lr ; src_ptr2 + plane_stride - subs r7, r7, #1 - bne extra_top_bottom_y - - b end_of_border_copy_y - -extra_top_bottom_uv - vld1.8 {d0}, [r1]! - vld1.8 {d8}, [r2]! - - mov r9, #2 ; 16 >> 3 - -extra_top_bottom_16 - subs r9, r9, #1 - - vst1.8 {d0}, [r5], lr - vst1.8 {d8}, [r6], lr - vst1.8 {d0}, [r5], lr - vst1.8 {d8}, [r6], lr - vst1.8 {d0}, [r5], lr - vst1.8 {d8}, [r6], lr - vst1.8 {d0}, [r5], lr - vst1.8 {d8}, [r6], lr - vst1.8 {d0}, [r5], lr - vst1.8 {d8}, [r6], lr - vst1.8 {d0}, [r5], lr - vst1.8 {d8}, [r6], lr - vst1.8 {d0}, [r5], lr - vst1.8 {d8}, [r6], lr - vst1.8 {d0}, [r5], lr - vst1.8 {d8}, [r6], lr - bne extra_top_bottom_16 - - sub r5, r1, lr, asl #4 ; src_ptr1 - (Border * plane_stride) - add r6, r2, lr ; src_ptr2 + plane_stride - subs r7, r7, #1 - bne extra_top_bottom_uv - - b end_of_border_copy_uv - - ENDP - END diff --git a/vpx_scale/arm/neon/yv12extend_arm.c b/vpx_scale/arm/neon/yv12extend_arm.c index fac7bbc..d408eb3 100644 --- a/vpx_scale/arm/neon/yv12extend_arm.c +++ b/vpx_scale/arm/neon/yv12extend_arm.c @@ -17,5 +17,5 @@ extern void vp8_yv12_copy_frame_func_neon( void vp8_yv12_copy_frame_neon(const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc) { vp8_yv12_copy_frame_func_neon(src_ybc, dst_ybc); - vp8_yv12_extend_frame_borders_neon(dst_ybc); + vp8_yv12_extend_frame_borders_c(dst_ybc); } diff --git a/vpx_scale/vpx_scale.mk b/vpx_scale/vpx_scale.mk index 95e7483..1fa41af 100644 --- a/vpx_scale/vpx_scale.mk +++ b/vpx_scale/vpx_scale.mk @@ -12,7 +12,6 @@ SCALE_SRCS-yes += vpx_scale_rtcd.pl #neon SCALE_SRCS-$(HAVE_NEON_ASM) += arm/neon/vp8_vpxyv12_copyframe_func_neon$(ASM) SCALE_SRCS-$(HAVE_NEON_ASM) += arm/neon/vp8_vpxyv12_copysrcframe_func_neon$(ASM) -SCALE_SRCS-$(HAVE_NEON_ASM) += arm/neon/vp8_vpxyv12_extendframeborders_neon$(ASM) SCALE_SRCS-$(HAVE_NEON_ASM) += arm/neon/yv12extend_arm.c #mips(dspr2) diff --git a/vpx_scale/vpx_scale_rtcd.pl b/vpx_scale/vpx_scale_rtcd.pl index 2e3f1ff..5a7f973 100644 --- a/vpx_scale/vpx_scale_rtcd.pl +++ b/vpx_scale/vpx_scale_rtcd.pl @@ -17,8 +17,6 @@ if (vpx_config("CONFIG_SPATIAL_RESAMPLING") eq "yes") { } add_proto qw/void vp8_yv12_extend_frame_borders/, "struct yv12_buffer_config *ybf"; -specialize qw/vp8_yv12_extend_frame_borders neon_asm/; -$vp8_yv12_extend_frame_borders_neon_asm=vp8_yv12_extend_frame_borders_neon; add_proto qw/void vp8_yv12_copy_frame/, "const struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc"; specialize qw/vp8_yv12_copy_frame neon_asm/; -- 2.7.4