OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2009 The Android Open Source Project | 2 * Copyright 2009 The Android Open Source Project |
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 | 8 |
9 #include "SkBitmapScaler.h" | 9 #include "SkBitmapScaler.h" |
10 #include "SkBitmapProcState.h" | 10 #include "SkBitmapProcState.h" |
11 #include "SkColorPriv.h" | 11 #include "SkColorPriv.h" |
12 #include "SkPaint.h" | 12 #include "SkPaint.h" |
13 #include "SkTypes.h" | 13 #include "SkTypes.h" |
14 #include "SkUtils.h" | 14 #include "SkUtils.h" |
15 #include "SkUtilsArm.h" | 15 #include "SkUtilsArm.h" |
16 | 16 |
17 #include "SkConvolver.h" | 17 #include "SkConvolver.h" |
18 | 18 |
19 #if !defined(SK_CPU_ARM64) && SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN) | 19 void SkBitmapProcState::platformProcs() { } |
20 void SI8_D16_nofilter_DX_arm( | |
21 const SkBitmapProcState& s, | |
22 const uint32_t* SK_RESTRICT xy, | |
23 int count, | |
24 uint16_t* SK_RESTRICT colors) SK_ATTRIBUTE_OPTIMIZE_O1; | |
25 | |
26 void SI8_D16_nofilter_DX_arm(const SkBitmapProcState& s, | |
27 const uint32_t* SK_RESTRICT xy, | |
28 int count, uint16_t* SK_RESTRICT colors) { | |
29 SkASSERT(count > 0 && colors != NULL); | |
30 SkASSERT(s.fInvType <= (SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask)); | |
31 SkASSERT(kNone_SkFilterQuality == s.fFilterLevel); | |
32 | |
33 const uint16_t* SK_RESTRICT table = s.fPixmap.ctable()->read16BitCache(); | |
34 const uint8_t* SK_RESTRICT srcAddr = (const uint8_t*)s.fPixmap.addr(); | |
35 | |
36 // buffer is y32, x16, x16, x16, x16, x16 | |
37 // bump srcAddr to the proper row, since we're told Y never changes | |
38 SkASSERT((unsigned)xy[0] < (unsigned)s.fPixmap.height()); | |
39 srcAddr = (const uint8_t*)((const char*)srcAddr + xy[0] * s.fPixmap.rowBytes
()); | |
40 | |
41 uint8_t src; | |
42 | |
43 if (1 == s.fPixmap.width()) { | |
44 src = srcAddr[0]; | |
45 uint16_t dstValue = table[src]; | |
46 sk_memset16(colors, dstValue, count); | |
47 } else { | |
48 int i; | |
49 int count8 = count >> 3; | |
50 const uint16_t* SK_RESTRICT xx = (const uint16_t*)(xy + 1); | |
51 | |
52 asm volatile ( | |
53 "cmp %[count8], #0 \n\t" // compare loop co
unter with 0 | |
54 "beq 2f \n\t" // if loop counter
== 0, exit | |
55 "1: \n\t" | |
56 "ldmia %[xx]!, {r5, r7, r9, r11} \n\t" // load ptrs to pi
xels 0-7 | |
57 "subs %[count8], %[count8], #1 \n\t" // decrement loop
counter | |
58 "uxth r4, r5 \n\t" // extract ptr 0 | |
59 "mov r5, r5, lsr #16 \n\t" // extract ptr 1 | |
60 "uxth r6, r7 \n\t" // extract ptr 2 | |
61 "mov r7, r7, lsr #16 \n\t" // extract ptr 3 | |
62 "ldrb r4, [%[srcAddr], r4] \n\t" // load pixel 0 fr
om image | |
63 "uxth r8, r9 \n\t" // extract ptr 4 | |
64 "ldrb r5, [%[srcAddr], r5] \n\t" // load pixel 1 fr
om image | |
65 "mov r9, r9, lsr #16 \n\t" // extract ptr 5 | |
66 "ldrb r6, [%[srcAddr], r6] \n\t" // load pixel 2 fr
om image | |
67 "uxth r10, r11 \n\t" // extract ptr 6 | |
68 "ldrb r7, [%[srcAddr], r7] \n\t" // load pixel 3 fr
om image | |
69 "mov r11, r11, lsr #16 \n\t" // extract ptr 7 | |
70 "ldrb r8, [%[srcAddr], r8] \n\t" // load pixel 4 fr
om image | |
71 "add r4, r4, r4 \n\t" // double pixel 0
for RGB565 lookup | |
72 "ldrb r9, [%[srcAddr], r9] \n\t" // load pixel 5 fr
om image | |
73 "add r5, r5, r5 \n\t" // double pixel 1
for RGB565 lookup | |
74 "ldrb r10, [%[srcAddr], r10] \n\t" // load pixel 6 fr
om image | |
75 "add r6, r6, r6 \n\t" // double pixel 2
for RGB565 lookup | |
76 "ldrb r11, [%[srcAddr], r11] \n\t" // load pixel 7 fr
om image | |
77 "add r7, r7, r7 \n\t" // double pixel 3
for RGB565 lookup | |
78 "ldrh r4, [%[table], r4] \n\t" // load pixel 0 RG
B565 from colmap | |
79 "add r8, r8, r8 \n\t" // double pixel 4
for RGB565 lookup | |
80 "ldrh r5, [%[table], r5] \n\t" // load pixel 1 RG
B565 from colmap | |
81 "add r9, r9, r9 \n\t" // double pixel 5
for RGB565 lookup | |
82 "ldrh r6, [%[table], r6] \n\t" // load pixel 2 RG
B565 from colmap | |
83 "add r10, r10, r10 \n\t" // double pixel 6
for RGB565 lookup | |
84 "ldrh r7, [%[table], r7] \n\t" // load pixel 3 RG
B565 from colmap | |
85 "add r11, r11, r11 \n\t" // double pixel 7
for RGB565 lookup | |
86 "ldrh r8, [%[table], r8] \n\t" // load pixel 4 RG
B565 from colmap | |
87 "ldrh r9, [%[table], r9] \n\t" // load pixel 5 RG
B565 from colmap | |
88 "ldrh r10, [%[table], r10] \n\t" // load pixel 6 RG
B565 from colmap | |
89 "ldrh r11, [%[table], r11] \n\t" // load pixel 7 RG
B565 from colmap | |
90 "pkhbt r5, r4, r5, lsl #16 \n\t" // pack pixels 0 a
nd 1 | |
91 "pkhbt r6, r6, r7, lsl #16 \n\t" // pack pixels 2 a
nd 3 | |
92 "pkhbt r8, r8, r9, lsl #16 \n\t" // pack pixels 4 a
nd 5 | |
93 "pkhbt r10, r10, r11, lsl #16 \n\t" // pack pixels 6 a
nd 7 | |
94 "stmia %[colors]!, {r5, r6, r8, r10} \n\t" // store last 8 pi
xels | |
95 "bgt 1b \n\t" // loop if counter
> 0 | |
96 "2: \n\t" | |
97 : [xx] "+r" (xx), [count8] "+r" (count8), [colors] "+r" (colors) | |
98 : [table] "r" (table), [srcAddr] "r" (srcAddr) | |
99 : "memory", "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11" | |
100 ); | |
101 | |
102 for (i = (count & 7); i > 0; --i) { | |
103 src = srcAddr[*xx++]; *colors++ = table[src]; | |
104 } | |
105 } | |
106 } | |
107 | |
108 void SI8_opaque_D32_nofilter_DX_arm( | |
109 const SkBitmapProcState& s, | |
110 const uint32_t* SK_RESTRICT xy, | |
111 int count, | |
112 SkPMColor* SK_RESTRICT colors) SK_ATTRIBUTE_OPTIMIZE_O1; | |
113 | |
114 void SI8_opaque_D32_nofilter_DX_arm(const SkBitmapProcState& s, | |
115 const uint32_t* SK_RESTRICT xy, | |
116 int count, SkPMColor* SK_RESTRICT colors) { | |
117 SkASSERT(count > 0 && colors != NULL); | |
118 SkASSERT(s.fInvType <= (SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask)); | |
119 SkASSERT(kNone_SkFilterQuality == s.fFilterLevel); | |
120 | |
121 const SkPMColor* SK_RESTRICT table = s.fPixmap.ctable()->readColors(); | |
122 const uint8_t* SK_RESTRICT srcAddr = (const uint8_t*)s.fPixmap.addr(); | |
123 | |
124 // buffer is y32, x16, x16, x16, x16, x16 | |
125 // bump srcAddr to the proper row, since we're told Y never changes | |
126 SkASSERT((unsigned)xy[0] < (unsigned)s.fPixmap.height()); | |
127 srcAddr = (const uint8_t*)((const char*)srcAddr + xy[0] * s.fPixmap.rowBytes
()); | |
128 | |
129 if (1 == s.fPixmap.width()) { | |
130 uint8_t src = srcAddr[0]; | |
131 SkPMColor dstValue = table[src]; | |
132 sk_memset32(colors, dstValue, count); | |
133 } else { | |
134 const uint16_t* xx = (const uint16_t*)(xy + 1); | |
135 | |
136 asm volatile ( | |
137 "subs %[count], %[count], #8 \n\t" // decrement count
by 8, set flags | |
138 "blt 2f \n\t" // if count < 0, b
ranch to singles | |
139 "1: \n\t" // eights loop | |
140 "ldmia %[xx]!, {r5, r7, r9, r11} \n\t" // load ptrs to pi
xels 0-7 | |
141 "uxth r4, r5 \n\t" // extract ptr 0 | |
142 "mov r5, r5, lsr #16 \n\t" // extract ptr 1 | |
143 "uxth r6, r7 \n\t" // extract ptr 2 | |
144 "mov r7, r7, lsr #16 \n\t" // extract ptr 3 | |
145 "ldrb r4, [%[srcAddr], r4] \n\t" // load pixel 0 fr
om image | |
146 "uxth r8, r9 \n\t" // extract ptr 4 | |
147 "ldrb r5, [%[srcAddr], r5] \n\t" // load pixel 1 fr
om image | |
148 "mov r9, r9, lsr #16 \n\t" // extract ptr 5 | |
149 "ldrb r6, [%[srcAddr], r6] \n\t" // load pixel 2 fr
om image | |
150 "uxth r10, r11 \n\t" // extract ptr 6 | |
151 "ldrb r7, [%[srcAddr], r7] \n\t" // load pixel 3 fr
om image | |
152 "mov r11, r11, lsr #16 \n\t" // extract ptr 7 | |
153 "ldrb r8, [%[srcAddr], r8] \n\t" // load pixel 4 fr
om image | |
154 "ldrb r9, [%[srcAddr], r9] \n\t" // load pixel 5 fr
om image | |
155 "ldrb r10, [%[srcAddr], r10] \n\t" // load pixel 6 fr
om image | |
156 "ldrb r11, [%[srcAddr], r11] \n\t" // load pixel 7 fr
om image | |
157 "ldr r4, [%[table], r4, lsl #2] \n\t" // load pixel 0 Sk
PMColor from colmap | |
158 "ldr r5, [%[table], r5, lsl #2] \n\t" // load pixel 1 Sk
PMColor from colmap | |
159 "ldr r6, [%[table], r6, lsl #2] \n\t" // load pixel 2 Sk
PMColor from colmap | |
160 "ldr r7, [%[table], r7, lsl #2] \n\t" // load pixel 3 Sk
PMColor from colmap | |
161 "ldr r8, [%[table], r8, lsl #2] \n\t" // load pixel 4 Sk
PMColor from colmap | |
162 "ldr r9, [%[table], r9, lsl #2] \n\t" // load pixel 5 Sk
PMColor from colmap | |
163 "ldr r10, [%[table], r10, lsl #2] \n\t" // load pixel 6 Sk
PMColor from colmap | |
164 "ldr r11, [%[table], r11, lsl #2] \n\t" // load pixel 7 Sk
PMColor from colmap | |
165 "subs %[count], %[count], #8 \n\t" // decrement loop
counter | |
166 "stmia %[colors]!, {r4-r11} \n\t" // store 8 pixels | |
167 "bge 1b \n\t" // loop if counter
>= 0 | |
168 "2: \n\t" | |
169 "adds %[count], %[count], #8 \n\t" // fix up counter,
set flags | |
170 "beq 4f \n\t" // if count == 0,
branch to exit | |
171 "3: \n\t" // singles loop | |
172 "ldrh r4, [%[xx]], #2 \n\t" // load pixel ptr | |
173 "subs %[count], %[count], #1 \n\t" // decrement loop
counter | |
174 "ldrb r5, [%[srcAddr], r4] \n\t" // load pixel from
image | |
175 "ldr r6, [%[table], r5, lsl #2] \n\t" // load SkPMColor
from colmap | |
176 "str r6, [%[colors]], #4 \n\t" // store pixel, up
date ptr | |
177 "bne 3b \n\t" // loop if counter
!= 0 | |
178 "4: \n\t" // exit | |
179 : [xx] "+r" (xx), [count] "+r" (count), [colors] "+r" (colors) | |
180 : [table] "r" (table), [srcAddr] "r" (srcAddr) | |
181 : "memory", "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11" | |
182 ); | |
183 } | |
184 } | |
185 #endif // !defined(SK_CPU_ARM64) && SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN) | |
186 | 20 |
187 /////////////////////////////////////////////////////////////////////////////// | 21 /////////////////////////////////////////////////////////////////////////////// |
188 | 22 |
189 /* If we replace a sampleproc, then we null-out the associated shaderproc, | |
190 otherwise the shader won't even look at the matrix/sampler | |
191 */ | |
192 void SkBitmapProcState::platformProcs() { | |
193 #if !defined(SK_CPU_ARM64) && SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN) | |
194 bool isOpaque = 256 == fAlphaScale; | |
195 bool justDx = false; | |
196 | |
197 if (fInvType <= (SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask)) { | |
198 justDx = true; | |
199 } | |
200 | |
201 switch (fPixmap.colorType()) { | |
202 case kIndex_8_SkColorType: | |
203 if (justDx && kNone_SkFilterQuality == fFilterLevel) { | |
204 #if 0 /* crashing on android device */ | |
205 fSampleProc16 = SI8_D16_nofilter_DX_arm; | |
206 fShaderProc16 = NULL; | |
207 #endif | |
208 if (isOpaque) { | |
209 // this one is only very slighty faster than the C version | |
210 fSampleProc32 = SI8_opaque_D32_nofilter_DX_arm; | |
211 fShaderProc32 = NULL; | |
212 } | |
213 } | |
214 break; | |
215 default: | |
216 break; | |
217 } | |
218 #endif | |
219 } | |
220 | |
221 /////////////////////////////////////////////////////////////////////////////// | |
222 | |
223 extern void platformConvolutionProcs_arm_neon(SkConvolutionProcs* procs); | 23 extern void platformConvolutionProcs_arm_neon(SkConvolutionProcs* procs); |
224 | 24 |
225 void platformConvolutionProcs_arm(SkConvolutionProcs* procs) { | 25 void platformConvolutionProcs_arm(SkConvolutionProcs* procs) { |
226 } | 26 } |
227 | 27 |
228 void SkBitmapScaler::PlatformConvolutionProcs(SkConvolutionProcs* procs) { | 28 void SkBitmapScaler::PlatformConvolutionProcs(SkConvolutionProcs* procs) { |
229 SK_ARM_NEON_WRAP(platformConvolutionProcs_arm)(procs); | 29 SK_ARM_NEON_WRAP(platformConvolutionProcs_arm)(procs); |
230 } | 30 } |
OLD | NEW |