1 /*
2 * Copyright (C) 2013 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17
18 #include "rsCpuIntrinsic.h"
19 #include "rsCpuIntrinsicInlines.h"
20
21 #ifdef RS_COMPATIBILITY_LIB
22 #include "rsCompatibilityLib.h"
23 #endif
24
25 #ifndef RS_COMPATIBILITY_LIB
26 #include "hardware/gralloc.h"
27 #endif
28
29 using namespace android;
30 using namespace android::renderscript;
31
32 namespace android {
33 namespace renderscript {
34
35
36 class RsdCpuScriptIntrinsicYuvToRGB : public RsdCpuScriptIntrinsic {
37 public:
38 virtual void populateScript(Script *);
39 virtual void invokeFreeChildren();
40
41 virtual void setGlobalObj(uint32_t slot, ObjectBase *data);
42
43 virtual ~RsdCpuScriptIntrinsicYuvToRGB();
44 RsdCpuScriptIntrinsicYuvToRGB(RsdCpuReferenceImpl *ctx, const Script *s, const Element *e);
45
46 protected:
47 ObjectBaseRef<Allocation> alloc;
48
49 static void kernel(const RsForEachStubParamStruct *p,
50 uint32_t xstart, uint32_t xend,
51 uint32_t instep, uint32_t outstep);
52 };
53
54 }
55 }
56
57
setGlobalObj(uint32_t slot,ObjectBase * data)58 void RsdCpuScriptIntrinsicYuvToRGB::setGlobalObj(uint32_t slot, ObjectBase *data) {
59 rsAssert(slot == 0);
60 alloc.set(static_cast<Allocation *>(data));
61 }
62
63
64
65
rsYuvToRGBA_uchar4(uchar y,uchar u,uchar v)66 static uchar4 rsYuvToRGBA_uchar4(uchar y, uchar u, uchar v) {
67 short Y = ((short)y) - 16;
68 short U = ((short)u) - 128;
69 short V = ((short)v) - 128;
70
71 short4 p;
72 p.x = (Y * 298 + V * 409 + 128) >> 8;
73 p.y = (Y * 298 - U * 100 - V * 208 + 128) >> 8;
74 p.z = (Y * 298 + U * 516 + 128) >> 8;
75 p.w = 255;
76 if(p.x < 0) {
77 p.x = 0;
78 }
79 if(p.x > 255) {
80 p.x = 255;
81 }
82 if(p.y < 0) {
83 p.y = 0;
84 }
85 if(p.y > 255) {
86 p.y = 255;
87 }
88 if(p.z < 0) {
89 p.z = 0;
90 }
91 if(p.z > 255) {
92 p.z = 255;
93 }
94
95 return (uchar4){static_cast<uchar>(p.x), static_cast<uchar>(p.y),
96 static_cast<uchar>(p.z), static_cast<uchar>(p.w)};
97 }
98
99
100 extern "C" void rsdIntrinsicYuv_K(void *dst, const uchar *Y, const uchar *uv, uint32_t xstart, size_t xend);
101 extern "C" void rsdIntrinsicYuvR_K(void *dst, const uchar *Y, const uchar *uv, uint32_t xstart, size_t xend);
102 extern "C" void rsdIntrinsicYuv2_K(void *dst, const uchar *Y, const uchar *u, const uchar *v, size_t xstart, size_t xend);
103
kernel(const RsForEachStubParamStruct * p,uint32_t xstart,uint32_t xend,uint32_t instep,uint32_t outstep)104 void RsdCpuScriptIntrinsicYuvToRGB::kernel(const RsForEachStubParamStruct *p,
105 uint32_t xstart, uint32_t xend,
106 uint32_t instep, uint32_t outstep) {
107 RsdCpuScriptIntrinsicYuvToRGB *cp = (RsdCpuScriptIntrinsicYuvToRGB *)p->usr;
108 if (!cp->alloc.get()) {
109 ALOGE("YuvToRGB executed without input, skipping");
110 return;
111 }
112 const uchar *pinY = (const uchar *)cp->alloc->mHal.drvState.lod[0].mallocPtr;
113 if (pinY == NULL) {
114 ALOGE("YuvToRGB executed without data, skipping");
115 return;
116 }
117
118 size_t strideY = cp->alloc->mHal.drvState.lod[0].stride;
119
120 // calculate correct stride in legacy case
121 if (cp->alloc->mHal.drvState.lod[0].dimY == 0) {
122 strideY = p->dimX;
123 }
124 const uchar *Y = pinY + (p->y * strideY);
125
126 uchar4 *out = (uchar4 *)p->out + xstart;
127 uint32_t x1 = xstart;
128 uint32_t x2 = xend;
129
130 size_t cstep = cp->alloc->mHal.drvState.yuv.step;
131
132 const uchar *pinU = (const uchar *)cp->alloc->mHal.drvState.lod[1].mallocPtr;
133 const size_t strideU = cp->alloc->mHal.drvState.lod[1].stride;
134 const uchar *u = pinU + ((p->y >> 1) * strideU);
135
136 const uchar *pinV = (const uchar *)cp->alloc->mHal.drvState.lod[2].mallocPtr;
137 const size_t strideV = cp->alloc->mHal.drvState.lod[2].stride;
138 const uchar *v = pinV + ((p->y >> 1) * strideV);
139
140 //ALOGE("pinY, %p, Y, %p, p->y, %d, strideY, %d", pinY, Y, p->y, strideY);
141 //ALOGE("pinU, %p, U, %p, p->y, %d, strideU, %d", pinU, u, p->y, strideU);
142 //ALOGE("pinV, %p, V, %p, p->y, %d, strideV, %d", pinV, v, p->y, strideV);
143 //ALOGE("dimX, %d, dimY, %d", cp->alloc->mHal.drvState.lod[0].dimX, cp->alloc->mHal.drvState.lod[0].dimY);
144 //ALOGE("p->dimX, %d, p->dimY, %d", p->dimX, p->dimY);
145
146 if (pinU == NULL) {
147 // Legacy yuv support didn't fill in uv
148 v = ((uint8_t *)cp->alloc->mHal.drvState.lod[0].mallocPtr) +
149 (strideY * p->dimY) +
150 ((p->y >> 1) * strideY);
151 u = v + 1;
152 cstep = 2;
153 }
154
155 /* If we start on an odd pixel then deal with it here and bump things along
156 * so that subsequent code can carry on with even-odd pairing assumptions.
157 */
158 if((x1 & 1) && (x2 > x1)) {
159 int cx = (x1 >> 1) * cstep;
160 *out = rsYuvToRGBA_uchar4(Y[x1], u[cx], v[cx]);
161 out++;
162 x1++;
163 }
164
165 #if defined(ARCH_ARM_USE_INTRINSICS)
166 if((x2 > x1) && gArchUseSIMD) {
167 int32_t len = x2 - x1;
168 if (cstep == 1) {
169 rsdIntrinsicYuv2_K(p->out, Y, u, v, x1, x2);
170 x1 += len;
171 out += len;
172 } else if (cstep == 2) {
173 // Check for proper interleave
174 intptr_t ipu = (intptr_t)u;
175 intptr_t ipv = (intptr_t)v;
176
177 if (ipu == (ipv + 1)) {
178 rsdIntrinsicYuv_K(p->out, Y, v, x1, x2);
179 x1 += len;
180 out += len;
181 } else if (ipu == (ipv - 1)) {
182 rsdIntrinsicYuvR_K(p->out, Y, u, x1, x2);
183 x1 += len;
184 out += len;
185 }
186 }
187 }
188 #endif
189
190 if(x2 > x1) {
191 // ALOGE("y %i %i %i", p->y, x1, x2);
192 while(x1 < x2) {
193 int cx = (x1 >> 1) * cstep;
194 *out = rsYuvToRGBA_uchar4(Y[x1], u[cx], v[cx]);
195 out++;
196 x1++;
197 *out = rsYuvToRGBA_uchar4(Y[x1], u[cx], v[cx]);
198 out++;
199 x1++;
200 }
201 }
202
203 }
204
RsdCpuScriptIntrinsicYuvToRGB(RsdCpuReferenceImpl * ctx,const Script * s,const Element * e)205 RsdCpuScriptIntrinsicYuvToRGB::RsdCpuScriptIntrinsicYuvToRGB(
206 RsdCpuReferenceImpl *ctx, const Script *s, const Element *e)
207 : RsdCpuScriptIntrinsic(ctx, s, e, RS_SCRIPT_INTRINSIC_ID_YUV_TO_RGB) {
208
209 mRootPtr = &kernel;
210 }
211
~RsdCpuScriptIntrinsicYuvToRGB()212 RsdCpuScriptIntrinsicYuvToRGB::~RsdCpuScriptIntrinsicYuvToRGB() {
213 }
214
populateScript(Script * s)215 void RsdCpuScriptIntrinsicYuvToRGB::populateScript(Script *s) {
216 s->mHal.info.exportedVariableCount = 1;
217 }
218
invokeFreeChildren()219 void RsdCpuScriptIntrinsicYuvToRGB::invokeFreeChildren() {
220 alloc.clear();
221 }
222
223
rsdIntrinsic_YuvToRGB(RsdCpuReferenceImpl * ctx,const Script * s,const Element * e)224 RsdCpuScriptImpl * rsdIntrinsic_YuvToRGB(RsdCpuReferenceImpl *ctx,
225 const Script *s, const Element *e) {
226 return new RsdCpuScriptIntrinsicYuvToRGB(ctx, s, e);
227 }
228