fix 32-bit parameters issue on arm64 assembly function

Karina committed 10 years ago

Commit d34e2091 parent 794376413 files · +322 −40Browse files

Showing the changes to codec/processing/src (4 of 13 files). Show all

Unified · Split

4 files

codec/processing/src/arm64/adaptive_quantization_aarch64_neon.S +2 −0
codec/processing/src/arm64/down_sample_aarch64_neon.S +13 −0
codec/processing/src/arm64/pixel_sad_aarch64_neon.S +3 −1
codec/processing/src/arm64/vaa_calc_aarch64_neon.S +6 −0

codec/processing/src/arm64/adaptive_quantization_aarch64_neon.S+2−0

@@ -33,6 +33,8 @@ · 32 unmodified lines
33	33	#ifdef HAVE_NEON_AARCH64
34	34	#include "arm_arch64_common_macro.S"
35	35	WELS_ASM_AARCH64_FUNC_BEGIN SampleVariance16x16_AArch64_neon
	36	+ SIGN_EXTENSION x1, w1
	37	+ SIGN_EXTENSION x3, w3
36	38	ld1 {v1.16b}, [x0], x1 //save the ref data (16bytes)
37	39	ld1 {v0.16b}, [x2], x3 //save the src data (16bytes)
38	40	uabd v2.16b, v0.16b, v1.16b

codec/processing/src/arm64/down_sample_aarch64_neon.S+13−0

@@ -174,6 +174,9 @@ comp_ds_bilinear_onethird_loop0: · 173 unmodified lines
174	174	//restore the tailer for the unasigned size
175	175	st1 {v16.16b}, [x0]
176	176	WELS_ASM_AARCH64_FUNC_END
	177	+;void DyadicBilinearQuarterDownsampler_AArch64_neon(uint8_t* pDst, const int32_t kiDstStride,
	178	+;uint8_t* pSrc, const int32_t kiSrcStride,
	179	+;const int32_t kiSrcWidth, const int32_t kiHeight);
177	180
178	181	WELS_ASM_AARCH64_FUNC_BEGIN DyadicBilinearQuarterDownsampler_AArch64_neon
179	182	//Initialize the register
@@ -229,6 +232,9 @@ comp_ds_bilinear_quarter_loop0: · 49 unmodified lines
229	232	st1 {v16.16b}, [x0]
230	233	WELS_ASM_AARCH64_FUNC_END
231	234
	235	+;void GeneralBilinearAccurateDownsampler_AArch64_neon (uint8_t* pDst, const int32_t kiDstStride,
	236	+; const int32_t kiDstWidth, const int32_t kiDstHeight,
	237	+; uint8_t* pSrc, const int32_t kiSrcStride, const uint32_t kuiScaleX, const uint32_t kuiScaleY);
232	238	WELS_ASM_AARCH64_FUNC_BEGIN GeneralBilinearAccurateDownsampler_AArch64_neon
233	239	mov w10, #32767
234	240	and w8, w6, w10
@@ -259,6 +265,13 @@ WELS_ASM_AARCH64_FUNC_BEGIN GeneralBilinearAccurateDownsampler_AArch64_neon · 24 unmodified lines
259	265
260	266	eor v26.16b, v26.16b, v26.16b
261	267	eor v27.16b, v27.16b, v27.16b
	268	+ SIGN_EXTENSION x1, w1
	269	+ SIGN_EXTENSION x2, w2
	270	+ SIGN_EXTENSION x3, w3
	271	+ SIGN_EXTENSION x5, w5
	272	+ SIGN_EXTENSION x6, w6
	273	+ SIGN_EXTENSION x7, w7
	274	+
262	275	sub x1, x1, x2
263	276	sub x3, x3, #1
264	277

codec/processing/src/arm64/pixel_sad_aarch64_neon.S+3−1

@@ -32,8 +32,10 @@ · 31 unmodified lines
32	32
33	33	#ifdef HAVE_NEON_AARCH64
34	34	#include "arm_arch64_common_macro.S"
35		−
	35	+;int32_t WelsProcessingSampleSad8x8_AArch64_neon (uint8_t, int32_t, uint8_t, int32_t);
36	36	WELS_ASM_AARCH64_FUNC_BEGIN WelsProcessingSampleSad8x8_AArch64_neon
	37	+ SIGN_EXTENSION x1, w1
	38	+ SIGN_EXTENSION x3, w3
37	39	ld1 {v0.8b}, [x0], x1
38	40	ld1 {v1.8b}, [x2], x3
39	41	uabdl v2.8h, v0.8b, v1.8b

codec/processing/src/arm64/vaa_calc_aarch64_neon.S+6−0

@@ -61,6 +61,8 @@ · 60 unmodified lines
61	61	*/
62	62	WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSad_AArch64_neon
63	63	eor v31.16b, v31.16b, v31.16b
	64	+
	65	+ SIGN_EXTENSION x4, w4
64	66	lsl x9, x4, #4
65	67	sub x10, x9, #16 //x10 keep the 16*pic_stride-16
66	68	sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width
@@ -119,6 +121,7 @@ WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSadBgd_AArch64_neon · 52 unmodified lines
119	121	ldr x15, [sp, #0]
120	122	eor v28.16b, v28.16b, v28.16b
121	123
	124	+ SIGN_EXTENSION x4, w4
122	125	lsl x9, x4, #4
123	126	sub x10, x9, #16 //x10 keep the 16*pic_stride-16
124	127	sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width
@@ -260,6 +263,7 @@ WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSadSsdBgd_AArch64_neon · 135 unmodified lines
260	263	ldr x15, [sp, #24] //p_mad8x8
261	264	eor v17.16b, v17.16b, v17.16b
262	265
	266	+ SIGN_EXTENSION x4, w4
263	267	lsl x9, x4, #4
264	268	sub x10, x9, #16 //x10 keep the 16*pic_stride-16
265	269	sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width
@@ -400,6 +404,7 @@ WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSadSsd_AArch64_neon · 134 unmodified lines
400	404	ldr x13, [sp, #8] //psqdiff16x16
401	405	eor v17.16b, v17.16b, v17.16b
402	406
	407	+ SIGN_EXTENSION x4, w4
403	408	lsl x9, x4, #4
404	409	sub x10, x9, #16 //x10 keep the 16*pic_stride-16
405	410	sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width
@@ -500,6 +505,7 @@ WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSadVar_AArch64_neon · 94 unmodified lines
500	505	ldr x12, [sp, #0] //psqsum16x16
501	506	eor v17.16b, v17.16b, v17.16b
502	507
	508	+ SIGN_EXTENSION x4, w4
503	509	lsl x9, x4, #4
504	510	sub x10, x9, #16 //x10 keep the 16*pic_stride-16
505	511	sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width

fix 32-bit parameters issue on arm64 assembly function

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code