OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2015 Google Inc. | 2 * Copyright 2015 Google Inc. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license that can be | 4 * Use of this source code is governed by a BSD-style license that can be |
5 * found in the LICENSE file. | 5 * found in the LICENSE file. |
6 */ | 6 */ |
7 | 7 |
8 #include "SkTypes.h" | 8 #include "SkTypes.h" |
9 #include <arm_neon.h> | 9 #include <arm_neon.h> |
10 | 10 |
11 void sk_memset32_neon(uint32_t dst[], uint32_t value, int count) { | 11 void sk_memset32_neon(uint32_t dst[], uint32_t value, int count) { |
12 uint32x4_t v4 = vdupq_n_u32(value); | 12 uint32x4_t v4 = vdupq_n_u32(value); |
13 uint32x4x4_t v16 = { v4, v4, v4, v4 }; | 13 uint32x4x4_t v16 = {{ v4, v4, v4, v4 }}; |
14 | 14 |
15 while (count >= 16) { | 15 while (count >= 16) { |
16 vst4q_u32(dst, v16); // This swizzles, but we don't care: all lanes are
the same, value. | 16 vst4q_u32(dst, v16); // This swizzles, but we don't care: all lanes are
the same, value. |
17 dst += 16; | 17 dst += 16; |
18 count -= 16; | 18 count -= 16; |
19 } | 19 } |
20 SkASSERT(count < 16); | 20 SkASSERT(count < 16); |
21 switch (count / 4) { | 21 switch (count / 4) { |
22 case 3: vst1q_u32(dst, v4); dst += 4; count -= 4; | 22 case 3: vst1q_u32(dst, v4); dst += 4; count -= 4; |
23 case 2: vst1q_u32(dst, v4); dst += 4; count -= 4; | 23 case 2: vst1q_u32(dst, v4); dst += 4; count -= 4; |
24 case 1: vst1q_u32(dst, v4); dst += 4; count -= 4; | 24 case 1: vst1q_u32(dst, v4); dst += 4; count -= 4; |
25 } | 25 } |
26 SkASSERT(count < 4); | 26 SkASSERT(count < 4); |
27 if (count >= 2) { | 27 if (count >= 2) { |
28 vst1_u32(dst, vget_low_u32(v4)); | 28 vst1_u32(dst, vget_low_u32(v4)); |
29 dst += 2; | 29 dst += 2; |
30 count -= 2; | 30 count -= 2; |
31 } | 31 } |
32 SkASSERT(count < 2); | 32 SkASSERT(count < 2); |
33 if (count > 0) { | 33 if (count > 0) { |
34 *dst = value; | 34 *dst = value; |
35 } | 35 } |
36 } | 36 } |
37 | 37 |
38 void sk_memset16_neon(uint16_t dst[], uint16_t value, int count) { | 38 void sk_memset16_neon(uint16_t dst[], uint16_t value, int count) { |
39 uint16x8_t v8 = vdupq_n_u16(value); | 39 uint16x8_t v8 = vdupq_n_u16(value); |
40 uint16x8x4_t v32 = { v8, v8, v8, v8 }; | 40 uint16x8x4_t v32 = {{ v8, v8, v8, v8 }}; |
41 | 41 |
42 while (count >= 32) { | 42 while (count >= 32) { |
43 vst4q_u16(dst, v32); // This swizzles, but we don't care: all lanes are
the same, value. | 43 vst4q_u16(dst, v32); // This swizzles, but we don't care: all lanes are
the same, value. |
44 dst += 32; | 44 dst += 32; |
45 count -= 32; | 45 count -= 32; |
46 } | 46 } |
47 SkASSERT(count < 32); | 47 SkASSERT(count < 32); |
48 switch (count / 8) { | 48 switch (count / 8) { |
49 case 3: vst1q_u16(dst, v8); dst += 8; count -= 8; | 49 case 3: vst1q_u16(dst, v8); dst += 8; count -= 8; |
50 case 2: vst1q_u16(dst, v8); dst += 8; count -= 8; | 50 case 2: vst1q_u16(dst, v8); dst += 8; count -= 8; |
51 case 1: vst1q_u16(dst, v8); dst += 8; count -= 8; | 51 case 1: vst1q_u16(dst, v8); dst += 8; count -= 8; |
52 } | 52 } |
53 SkASSERT(count < 8); | 53 SkASSERT(count < 8); |
54 if (count >= 4) { | 54 if (count >= 4) { |
55 vst1_u16(dst, vget_low_u16(v8)); | 55 vst1_u16(dst, vget_low_u16(v8)); |
56 dst += 4; | 56 dst += 4; |
57 count -= 4; | 57 count -= 4; |
58 } | 58 } |
59 SkASSERT(count < 4); | 59 SkASSERT(count < 4); |
60 switch (count) { | 60 switch (count) { |
61 case 3: *dst++ = value; | 61 case 3: *dst++ = value; |
62 case 2: *dst++ = value; | 62 case 2: *dst++ = value; |
63 case 1: *dst = value; | 63 case 1: *dst = value; |
64 } | 64 } |
65 } | 65 } |
66 | 66 |
OLD | NEW |