Alex Sakhartchouk | 7f126c7 | 2011-05-05 16:56:27 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2011 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 "rsdCore.h" |
| 19 | #include "rsdSampler.h" |
| 20 | |
| 21 | #include "rsContext.h" |
| 22 | #include "rsSampler.h" |
Jason Sams | 93eacc7 | 2012-12-18 14:26:57 -0800 | [diff] [blame] | 23 | |
| 24 | #ifndef RS_COMPATIBILITY_LIB |
Alex Sakhartchouk | 7f126c7 | 2011-05-05 16:56:27 -0700 | [diff] [blame] | 25 | #include "rsProgramVertex.h" |
| 26 | #include "rsProgramFragment.h" |
| 27 | |
| 28 | #include <GLES/gl.h> |
| 29 | #include <GLES/glext.h> |
Jason Sams | 93eacc7 | 2012-12-18 14:26:57 -0800 | [diff] [blame] | 30 | #endif |
Alex Sakhartchouk | 7f126c7 | 2011-05-05 16:56:27 -0700 | [diff] [blame] | 31 | |
Jason Sams | 06b0f7d | 2013-03-22 13:43:01 -0700 | [diff] [blame^] | 32 | typedef float float2 __attribute__((ext_vector_type(2))); |
| 33 | typedef float float3 __attribute__((ext_vector_type(3))); |
| 34 | typedef float float4 __attribute__((ext_vector_type(4))); |
| 35 | typedef uint8_t uchar4 __attribute__((ext_vector_type(4))); |
| 36 | |
Alex Sakhartchouk | 7f126c7 | 2011-05-05 16:56:27 -0700 | [diff] [blame] | 37 | using namespace android; |
| 38 | using namespace android::renderscript; |
| 39 | |
Jason Sams | 06b0f7d | 2013-03-22 13:43:01 -0700 | [diff] [blame^] | 40 | #if defined(ARCH_ARM_HAVE_VFP) |
| 41 | #define LOCAL_CALL __attribute__((pcs("aapcs-vfp"))) |
| 42 | #else |
| 43 | #define LOCAL_CALL |
| 44 | #endif |
| 45 | |
| 46 | // 565 Conversion bits taken from SkBitmap |
| 47 | #define SK_R16_BITS 5 |
| 48 | #define SK_G16_BITS 6 |
| 49 | #define SK_B16_BITS 5 |
| 50 | |
| 51 | #define SK_R16_SHIFT (SK_B16_BITS + SK_G16_BITS) |
| 52 | #define SK_G16_SHIFT (SK_B16_BITS) |
| 53 | #define SK_B16_SHIFT 0 |
| 54 | |
| 55 | #define SK_R16_MASK ((1 << SK_R16_BITS) - 1) |
| 56 | #define SK_G16_MASK ((1 << SK_G16_BITS) - 1) |
| 57 | #define SK_B16_MASK ((1 << SK_B16_BITS) - 1) |
| 58 | |
| 59 | static inline unsigned SkR16ToR32(unsigned r) { |
| 60 | return (r << (8 - SK_R16_BITS)) | (r >> (2 * SK_R16_BITS - 8)); |
| 61 | } |
| 62 | |
| 63 | static inline unsigned SkG16ToG32(unsigned g) { |
| 64 | return (g << (8 - SK_G16_BITS)) | (g >> (2 * SK_G16_BITS - 8)); |
| 65 | } |
| 66 | |
| 67 | static inline unsigned SkB16ToB32(unsigned b) { |
| 68 | return (b << (8 - SK_B16_BITS)) | (b >> (2 * SK_B16_BITS - 8)); |
| 69 | } |
| 70 | |
| 71 | #define SkPacked16ToR32(c) SkR16ToR32(SkGetPackedR16(c)) |
| 72 | #define SkPacked16ToG32(c) SkG16ToG32(SkGetPackedG16(c)) |
| 73 | #define SkPacked16ToB32(c) SkB16ToB32(SkGetPackedB16(c)) |
| 74 | |
| 75 | #define SkGetPackedR16(color) (((unsigned)(color) >> SK_R16_SHIFT) & SK_R16_MASK) |
| 76 | #define SkGetPackedG16(color) (((unsigned)(color) >> SK_G16_SHIFT) & SK_G16_MASK) |
| 77 | #define SkGetPackedB16(color) (((unsigned)(color) >> SK_B16_SHIFT) & SK_B16_MASK) |
| 78 | |
| 79 | static float3 getFrom565(uint16_t color) { |
| 80 | float3 result; |
| 81 | result.x = (float)SkPacked16ToR32(color); |
| 82 | result.y = (float)SkPacked16ToG32(color); |
| 83 | result.z = (float)SkPacked16ToB32(color); |
| 84 | return result; |
| 85 | } |
| 86 | |
| 87 | |
| 88 | |
| 89 | /** |
| 90 | * Allocation sampling |
| 91 | */ |
| 92 | static inline float getElementAt1(const uint8_t *p, int32_t x) { |
| 93 | float r = p[x]; |
| 94 | return r; |
| 95 | } |
| 96 | |
| 97 | static inline float2 getElementAt2(const uint8_t *p, int32_t x) { |
| 98 | x *= 2; |
| 99 | float2 r = {p[x], p[x+1]}; |
| 100 | return r; |
| 101 | } |
| 102 | |
| 103 | static inline float3 getElementAt3(const uint8_t *p, int32_t x) { |
| 104 | x *= 4; |
| 105 | float3 r = {p[x], p[x+1], p[x+2]}; |
| 106 | return r; |
| 107 | } |
| 108 | |
| 109 | static inline float4 getElementAt4(const uint8_t *p, int32_t x) { |
| 110 | x *= 4; |
| 111 | float4 r = {p[x], p[x+1], p[x+2], p[x+3]}; |
| 112 | return r; |
| 113 | } |
| 114 | |
| 115 | static inline float3 getElementAt565(const uint8_t *p, int32_t x) { |
| 116 | x *= 2; |
| 117 | float3 r = getFrom565(((const uint16_t *)p)[0]); |
| 118 | return r; |
| 119 | } |
| 120 | |
| 121 | static inline float getElementAt1(const uint8_t *p, size_t stride, int32_t x, int32_t y) { |
| 122 | p += y * stride; |
| 123 | float r = p[x]; |
| 124 | return r; |
| 125 | } |
| 126 | |
| 127 | static inline float2 getElementAt2(const uint8_t *p, size_t stride, int32_t x, int32_t y) { |
| 128 | p += y * stride; |
| 129 | x *= 2; |
| 130 | float2 r = {p[x], p[x+1]}; |
| 131 | return r; |
| 132 | } |
| 133 | |
| 134 | static inline float3 getElementAt3(const uint8_t *p, size_t stride, int32_t x, int32_t y) { |
| 135 | p += y * stride; |
| 136 | x *= 4; |
| 137 | float3 r = {p[x], p[x+1], p[x+2]}; |
| 138 | return r; |
| 139 | } |
| 140 | |
| 141 | static inline float4 getElementAt4(const uint8_t *p, size_t stride, int32_t x, int32_t y) { |
| 142 | p += y * stride; |
| 143 | x *= 4; |
| 144 | float4 r = {p[x], p[x+1], p[x+2], p[x+3]}; |
| 145 | return r; |
| 146 | } |
| 147 | |
| 148 | static inline float3 getElementAt565(const uint8_t *p, size_t stride, int32_t x, int32_t y) { |
| 149 | p += y * stride; |
| 150 | x *= 2; |
| 151 | float3 r = getFrom565(((const uint16_t *)p)[0]); |
| 152 | return r; |
| 153 | } |
| 154 | |
| 155 | |
| 156 | |
| 157 | |
| 158 | |
| 159 | static float4 LOCAL_CALL |
| 160 | getSample1D_A(const uint8_t *p, int32_t iPixel, |
| 161 | int32_t next, float w0, float w1) { |
| 162 | float p0 = getElementAt1(p, iPixel); |
| 163 | float p1 = getElementAt1(p, next); |
| 164 | float r = p0 * w0 + p1 * w1; |
| 165 | r *= (1.f / 255.f); |
| 166 | float4 ret = {0.f, 0.f, 0.f, r}; |
| 167 | return ret; |
| 168 | } |
| 169 | static float4 LOCAL_CALL |
| 170 | getSample1D_L(const uint8_t *p, int32_t iPixel, |
| 171 | int32_t next, float w0, float w1) { |
| 172 | float p0 = getElementAt1(p, iPixel); |
| 173 | float p1 = getElementAt1(p, next); |
| 174 | float r = p0 * w0 + p1 * w1; |
| 175 | r *= (1.f / 255.f); |
| 176 | float4 ret = {r, r, r, 1.f}; |
| 177 | return ret; |
| 178 | } |
| 179 | static float4 LOCAL_CALL |
| 180 | getSample1D_LA(const uint8_t *p, int32_t iPixel, |
| 181 | int32_t next, float w0, float w1) { |
| 182 | float2 p0 = getElementAt2(p, iPixel); |
| 183 | float2 p1 = getElementAt2(p, next); |
| 184 | float2 r = p0 * w0 + p1 * w1; |
| 185 | r *= (1.f / 255.f); |
| 186 | float4 ret = {r.x, r.x, r.x, r.y}; |
| 187 | return ret; |
| 188 | } |
| 189 | static float4 LOCAL_CALL |
| 190 | getSample1D_RGB(const uint8_t *p, int32_t iPixel, |
| 191 | int32_t next, float w0, float w1) { |
| 192 | float3 p0 = getElementAt3(p, iPixel); |
| 193 | float3 p1 = getElementAt3(p, next); |
| 194 | float3 r = p0 * w0 + p1 * w1; |
| 195 | r *= (1.f / 255.f); |
| 196 | float4 ret = {r.x, r.x, r.z, 1.f}; |
| 197 | return ret; |
| 198 | } |
| 199 | static float4 LOCAL_CALL |
| 200 | getSample1D_565(const uint8_t *p, int32_t iPixel, |
| 201 | int32_t next, float w0, float w1) { |
| 202 | float3 p0 = getElementAt565(p, iPixel); |
| 203 | float3 p1 = getElementAt565(p, next); |
| 204 | float3 r = p0 * w0 + p1 * w1; |
| 205 | r *= (1.f / 255.f); |
| 206 | float4 ret = {r.x, r.x, r.z, 1.f}; |
| 207 | return ret; |
| 208 | } |
| 209 | static float4 LOCAL_CALL |
| 210 | getSample1D_RGBA(const uint8_t *p, int32_t iPixel, |
| 211 | int32_t next, float w0, float w1) { |
| 212 | float4 p0 = getElementAt4(p, iPixel); |
| 213 | float4 p1 = getElementAt4(p, next); |
| 214 | float4 r = p0 * w0 + p1 * w1; |
| 215 | r *= (1.f / 255.f); |
| 216 | return r; |
| 217 | } |
| 218 | |
| 219 | |
| 220 | static float4 LOCAL_CALL |
| 221 | getSample2D_A(const uint8_t *p, size_t stride, |
| 222 | int locX, int locY, int nextX, int nextY, |
| 223 | float w0, float w1, float w2, float w3) { |
| 224 | float p0 = getElementAt1(p, stride, locX, locY); |
| 225 | float p1 = getElementAt1(p, stride, nextX, locY); |
| 226 | float p2 = getElementAt1(p, stride, locX, nextY); |
| 227 | float p3 = getElementAt1(p, stride, nextX, nextY); |
| 228 | float r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3; |
| 229 | r *= (1.f / 255.f); |
| 230 | float4 ret = {0.f, 0.f, 0.f, r}; |
| 231 | return ret; |
| 232 | } |
| 233 | static float4 LOCAL_CALL |
| 234 | getSample2D_L(const uint8_t *p, size_t stride, |
| 235 | int locX, int locY, int nextX, int nextY, |
| 236 | float w0, float w1, float w2, float w3) { |
| 237 | float p0 = getElementAt1(p, stride, locX, locY); |
| 238 | float p1 = getElementAt1(p, stride, nextX, locY); |
| 239 | float p2 = getElementAt1(p, stride, locX, nextY); |
| 240 | float p3 = getElementAt1(p, stride, nextX, nextY); |
| 241 | float r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3; |
| 242 | r *= (1.f / 255.f); |
| 243 | float4 ret = {r, r, r, 1.f}; |
| 244 | return ret; |
| 245 | } |
| 246 | static float4 LOCAL_CALL |
| 247 | getSample2D_LA(const uint8_t *p, size_t stride, |
| 248 | int locX, int locY, int nextX, int nextY, |
| 249 | float w0, float w1, float w2, float w3) { |
| 250 | float2 p0 = getElementAt2(p, stride, locX, locY); |
| 251 | float2 p1 = getElementAt2(p, stride, nextX, locY); |
| 252 | float2 p2 = getElementAt2(p, stride, locX, nextY); |
| 253 | float2 p3 = getElementAt2(p, stride, nextX, nextY); |
| 254 | float2 r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3; |
| 255 | r *= (1.f / 255.f); |
| 256 | float4 ret = {r.x, r.x, r.x, r.y}; |
| 257 | return ret; |
| 258 | } |
| 259 | static float4 LOCAL_CALL |
| 260 | getSample2D_RGB(const uint8_t *p, size_t stride, |
| 261 | int locX, int locY, int nextX, int nextY, |
| 262 | float w0, float w1, float w2, float w3) { |
| 263 | float4 p0 = getElementAt4(p, stride, locX, locY); |
| 264 | float4 p1 = getElementAt4(p, stride, nextX, locY); |
| 265 | float4 p2 = getElementAt4(p, stride, locX, nextY); |
| 266 | float4 p3 = getElementAt4(p, stride, nextX, nextY); |
| 267 | float4 r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3; |
| 268 | r *= (1.f / 255.f); |
| 269 | float4 ret = {r.x, r.y, r.z, 1.f}; |
| 270 | return ret; |
| 271 | } |
| 272 | static float4 LOCAL_CALL |
| 273 | getSample2D_RGBA(const uint8_t *p, size_t stride, |
| 274 | int locX, int locY, int nextX, int nextY, |
| 275 | float w0, float w1, float w2, float w3) { |
| 276 | float4 p0 = getElementAt4(p, stride, locX, locY); |
| 277 | float4 p1 = getElementAt4(p, stride, nextX, locY); |
| 278 | float4 p2 = getElementAt4(p, stride, locX, nextY); |
| 279 | float4 p3 = getElementAt4(p, stride, nextX, nextY); |
| 280 | float4 r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3; |
| 281 | r *= (1.f / 255.f); |
| 282 | return r; |
| 283 | } |
| 284 | static float4 getSample2D_565(const uint8_t *p, size_t stride, |
| 285 | int locX, int locY, int nextX, int nextY, |
| 286 | float w0, float w1, float w2, float w3) { |
| 287 | float3 p0 = getElementAt565(p, stride, locX, locY); |
| 288 | float3 p1 = getElementAt565(p, stride, nextX, locY); |
| 289 | float3 p2 = getElementAt565(p, stride, locX, nextY); |
| 290 | float3 p3 = getElementAt565(p, stride, nextX, nextY); |
| 291 | float3 r = p0 * w0 + p1 * w1 + p2 * w2 + p3 * w3; |
| 292 | r *= (1.f / 255.f); |
| 293 | float4 ret = {r.x, r.y, r.z, 1.f}; |
| 294 | return ret; |
| 295 | } |
| 296 | |
| 297 | |
| 298 | extern "C" { |
| 299 | typedef float4 Sampler2DFn(const uint8_t *p, size_t stride, |
| 300 | int lx, int ly, int nx, int ny, |
| 301 | float w0, float w1, float w2, float w3) LOCAL_CALL; |
| 302 | |
| 303 | Sampler2DFn rsdCpuGetSample2D_L_k; |
| 304 | Sampler2DFn rsdCpuGetSample2D_A_k; |
| 305 | Sampler2DFn rsdCpuGetSample2D_LA_k; |
| 306 | Sampler2DFn rsdCpuGetSample2D_RGB_k; |
| 307 | Sampler2DFn rsdCpuGetSample2D_RGBA_k; |
| 308 | } |
| 309 | |
| 310 | #if 0 |
| 311 | static Sampler2DFn* GetBilinearSampleTable2D[] = { |
| 312 | 0, 0, 0, 0, 0, 0, 0, |
| 313 | 0,//rsdCpuGetSample2D_L_k, |
| 314 | 0,//rsdCpuGetSample2D_A_k, |
| 315 | 0,//rsdCpuGetSample2D_LA_k, |
| 316 | 0,//rsdCpuGetSample2D_RGB_k, |
| 317 | rsdCpuGetSample2D_RGBA_k |
| 318 | }; |
| 319 | |
| 320 | #else |
| 321 | static Sampler2DFn* GetBilinearSampleTable2D[] = { |
| 322 | 0, 0, 0, 0, 0, 0, 0, |
| 323 | &getSample2D_L, |
| 324 | &getSample2D_A, |
| 325 | &getSample2D_LA, |
| 326 | &getSample2D_RGB, |
| 327 | &getSample2D_RGBA, |
| 328 | }; |
| 329 | #endif |
| 330 | |
| 331 | |
| 332 | static int applyWrapMode(RsSamplerValue mode, int coord, int size) { |
| 333 | switch (mode) { |
| 334 | case RS_SAMPLER_WRAP: |
| 335 | coord = coord % size; |
| 336 | if (coord < 0) { |
| 337 | coord += size; |
| 338 | } |
| 339 | break; |
| 340 | |
| 341 | case RS_SAMPLER_CLAMP: |
| 342 | coord = rsMax(0, rsMin(coord, size - 1)); |
| 343 | break; |
| 344 | |
| 345 | case RS_SAMPLER_MIRRORED_REPEAT: |
| 346 | coord = coord % (size * 2); |
| 347 | if (coord < 0) { |
| 348 | coord = (size * 2) + coord; |
| 349 | } |
| 350 | if (coord >= size) { |
| 351 | coord = (size * 2) - coord; |
| 352 | } |
| 353 | break; |
| 354 | |
| 355 | default: |
| 356 | coord = 0; |
| 357 | rsAssert(0); |
| 358 | } |
| 359 | return coord; |
| 360 | } |
| 361 | |
| 362 | static float4 |
| 363 | sample_LOD_LinearPixel(Allocation *a, const Type *type, |
| 364 | RsDataKind dk, RsDataType dt, |
| 365 | Sampler *s, |
| 366 | float uv, int32_t lod) { |
| 367 | RsSamplerValue wrapS = s->mHal.state.wrapS; |
| 368 | int32_t sourceW = type->mHal.state.lodDimX[lod]; |
| 369 | float pixelUV = uv * (float)(sourceW); |
| 370 | int32_t iPixel = (int32_t)(pixelUV); |
| 371 | float frac = pixelUV - (float)iPixel; |
| 372 | |
| 373 | if (frac < 0.5f) { |
| 374 | iPixel -= 1; |
| 375 | frac += 0.5f; |
| 376 | } else { |
| 377 | frac -= 0.5f; |
| 378 | } |
| 379 | |
| 380 | float oneMinusFrac = 1.0f - frac; |
| 381 | |
| 382 | int32_t next = applyWrapMode(wrapS, iPixel + 1, sourceW); |
| 383 | int32_t loc = applyWrapMode(wrapS, iPixel, sourceW); |
| 384 | |
| 385 | const uint8_t *ptr = (const uint8_t *)a->mHal.drvState.lod[lod].mallocPtr; |
| 386 | |
| 387 | if (dt == RS_TYPE_UNSIGNED_5_6_5) { |
| 388 | return getSample1D_565(ptr, loc, next, next, frac); |
| 389 | } |
| 390 | |
| 391 | switch(dk) { |
| 392 | case RS_KIND_PIXEL_L: |
| 393 | return getSample1D_L(ptr, loc, next, next, frac); |
| 394 | case RS_KIND_PIXEL_A: |
| 395 | return getSample1D_A(ptr, loc, next, next, frac); |
| 396 | case RS_KIND_PIXEL_LA: |
| 397 | return getSample1D_LA(ptr, loc, next, next, frac); |
| 398 | case RS_KIND_PIXEL_RGB: |
| 399 | return getSample1D_RGB(ptr, loc, next, next, frac); |
| 400 | case RS_KIND_PIXEL_RGBA: |
| 401 | return getSample1D_RGBA(ptr, loc, next, next, frac); |
| 402 | |
| 403 | case RS_KIND_PIXEL_YUV: |
| 404 | case RS_KIND_USER: |
| 405 | case RS_KIND_INVALID: |
| 406 | case RS_KIND_PIXEL_DEPTH: |
| 407 | rsAssert(0); |
| 408 | break; |
| 409 | } |
| 410 | |
| 411 | return 0.f; |
| 412 | } |
| 413 | |
| 414 | static float4 |
| 415 | sample_LOD_NearestPixel(Allocation *a, const Type *type, |
| 416 | RsDataKind dk, RsDataType dt, |
| 417 | Sampler *s, float uv, int32_t lod) { |
| 418 | RsSamplerValue wrapS = s->mHal.state.wrapS; |
| 419 | int32_t sourceW = type->mHal.state.lodDimX[lod]; |
| 420 | int32_t iPixel = (int32_t)(uv * (float)(sourceW)); |
| 421 | int32_t location = applyWrapMode(wrapS, iPixel, sourceW); |
| 422 | |
| 423 | |
| 424 | const uint8_t *ptr = (const uint8_t *)a->mHal.drvState.lod[lod].mallocPtr; |
| 425 | |
| 426 | float4 result = {0.f, 0.f, 0.f, 1.f}; |
| 427 | if (dt == RS_TYPE_UNSIGNED_5_6_5) { |
| 428 | result.xyz = getElementAt565(ptr, iPixel); |
| 429 | return result; |
| 430 | } |
| 431 | |
| 432 | switch(dk) { |
| 433 | case RS_KIND_PIXEL_L: |
| 434 | { |
| 435 | float t = getElementAt1(ptr, iPixel); |
| 436 | result.xyz = t; |
| 437 | } |
| 438 | break; |
| 439 | case RS_KIND_PIXEL_A: |
| 440 | result.w = getElementAt1(ptr, iPixel); |
| 441 | break; |
| 442 | case RS_KIND_PIXEL_LA: |
| 443 | { |
| 444 | float2 t = getElementAt2(ptr, iPixel); |
| 445 | result.xyz = t.x; |
| 446 | result.w = t.y; |
| 447 | } |
| 448 | break; |
| 449 | case RS_KIND_PIXEL_RGB: |
| 450 | result.xyz = getElementAt3(ptr, iPixel); |
| 451 | break; |
| 452 | case RS_KIND_PIXEL_RGBA: |
| 453 | result = getElementAt4(ptr, iPixel); |
| 454 | break; |
| 455 | |
| 456 | case RS_KIND_PIXEL_YUV: |
| 457 | case RS_KIND_USER: |
| 458 | case RS_KIND_INVALID: |
| 459 | case RS_KIND_PIXEL_DEPTH: |
| 460 | rsAssert(0); |
| 461 | break; |
| 462 | } |
| 463 | |
| 464 | return result * (1.f / 255.f); |
| 465 | } |
| 466 | |
| 467 | |
| 468 | static float4 |
| 469 | sample_LOD_LinearPixel(Allocation *a, const Type *type, |
| 470 | RsDataKind dk, RsDataType dt, |
| 471 | Sampler *s, float u, float v, int32_t lod) { |
| 472 | const RsSamplerValue wrapS = s->mHal.state.wrapS; |
| 473 | const RsSamplerValue wrapT = s->mHal.state.wrapT; |
| 474 | const int sourceW = type->mHal.state.lodDimX[lod]; |
| 475 | const int sourceH = type->mHal.state.lodDimY[lod]; |
| 476 | |
| 477 | float pixelU = u * (float)sourceW; |
| 478 | float pixelV = v * (float)sourceH; |
| 479 | int iPixelU = (int)pixelU; |
| 480 | int iPixelV = (int)pixelV; |
| 481 | |
| 482 | float fracU = pixelU - iPixelU; |
| 483 | float fracV = pixelV - iPixelV; |
| 484 | |
| 485 | if (fracU < 0.5f) { |
| 486 | iPixelU -= 1; |
| 487 | fracU += 0.5f; |
| 488 | } else { |
| 489 | fracU -= 0.5f; |
| 490 | } |
| 491 | if (fracV < 0.5f) { |
| 492 | iPixelV -= 1; |
| 493 | fracV += 0.5f; |
| 494 | } else { |
| 495 | fracV -= 0.5f; |
| 496 | } |
| 497 | float oneMinusFracU = 1.0f - fracU; |
| 498 | float oneMinusFracV = 1.0f - fracV; |
| 499 | |
| 500 | float w1 = oneMinusFracU * oneMinusFracV; |
| 501 | float w2 = fracU * oneMinusFracV; |
| 502 | float w3 = oneMinusFracU * fracV; |
| 503 | float w4 = fracU * fracV; |
| 504 | |
| 505 | int nextX = applyWrapMode(wrapS, iPixelU + 1, sourceW); |
| 506 | int nextY = applyWrapMode(wrapT, iPixelV + 1, sourceH); |
| 507 | int locX = applyWrapMode(wrapS, iPixelU, sourceW); |
| 508 | int locY = applyWrapMode(wrapT, iPixelV, sourceH); |
| 509 | |
| 510 | const uint8_t *ptr = (const uint8_t *)a->mHal.drvState.lod[lod].mallocPtr; |
| 511 | size_t stride = a->mHal.drvState.lod[lod].stride; |
| 512 | |
| 513 | if (dt == RS_TYPE_UNSIGNED_5_6_5) { |
| 514 | return getSample2D_565(ptr, stride, locX, locY, nextX, nextY, w1, w2, w3, w4); |
| 515 | } |
| 516 | |
| 517 | return GetBilinearSampleTable2D[dk](ptr, stride, locX, locY, nextX, nextY, w1, w2, w3, w4); |
| 518 | } |
| 519 | |
| 520 | static float4 |
| 521 | sample_LOD_NearestPixel(Allocation *a, const Type *type, |
| 522 | RsDataKind dk, RsDataType dt, |
| 523 | Sampler *s, |
| 524 | float u, float v, int32_t lod) { |
| 525 | RsSamplerValue wrapS = s->mHal.state.wrapS; |
| 526 | RsSamplerValue wrapT = s->mHal.state.wrapT; |
| 527 | |
| 528 | int32_t sourceW = type->mHal.state.lodDimX[lod]; |
| 529 | int32_t sourceH = type->mHal.state.lodDimY[lod]; |
| 530 | |
| 531 | int locX = applyWrapMode(wrapS, u * sourceW, sourceW); |
| 532 | int locY = applyWrapMode(wrapT, v * sourceH, sourceH); |
| 533 | |
| 534 | |
| 535 | const uint8_t *ptr = (const uint8_t *)a->mHal.drvState.lod[lod].mallocPtr; |
| 536 | size_t stride = a->mHal.drvState.lod[lod].stride; |
| 537 | |
| 538 | float4 result = {0.f, 0.f, 0.f, 1.f}; |
| 539 | if (dt == RS_TYPE_UNSIGNED_5_6_5) { |
| 540 | result.xyz = getElementAt565(ptr, stride, locX, locY); |
| 541 | return result; |
| 542 | } |
| 543 | |
| 544 | switch(dk) { |
| 545 | case RS_KIND_PIXEL_L: |
| 546 | { |
| 547 | float t = getElementAt1(ptr, stride, locX, locY); |
| 548 | result.xyz = t; |
| 549 | } |
| 550 | break; |
| 551 | case RS_KIND_PIXEL_A: |
| 552 | result.w = getElementAt1(ptr, stride, locX, locY); |
| 553 | break; |
| 554 | case RS_KIND_PIXEL_LA: |
| 555 | { |
| 556 | float2 t = getElementAt2(ptr, stride, locX, locY); |
| 557 | result.xyz = t.x; |
| 558 | result.w = t.y; |
| 559 | } |
| 560 | break; |
| 561 | case RS_KIND_PIXEL_RGB: |
| 562 | result.xyz = getElementAt3(ptr, stride, locX, locY); |
| 563 | break; |
| 564 | case RS_KIND_PIXEL_RGBA: |
| 565 | result = getElementAt4(ptr, stride, locX, locY); |
| 566 | break; |
| 567 | |
| 568 | |
| 569 | case RS_KIND_PIXEL_YUV: |
| 570 | case RS_KIND_USER: |
| 571 | case RS_KIND_INVALID: |
| 572 | case RS_KIND_PIXEL_DEPTH: |
| 573 | rsAssert(0); |
| 574 | break; |
| 575 | } |
| 576 | |
| 577 | return result * (1.f / 255.f); |
| 578 | } |
| 579 | |
| 580 | |
| 581 | |
| 582 | static float4 GenericSample1D(Allocation *a, Sampler *s, float u, float lod) { |
| 583 | const Type *type = a->getType(); |
| 584 | const Element *elem = type->getElement(); |
| 585 | const RsDataKind dk = elem->getKind(); |
| 586 | const RsDataType dt = elem->getType(); |
| 587 | |
| 588 | if (dk == RS_KIND_USER || (dt != RS_TYPE_UNSIGNED_8 && dt != RS_TYPE_UNSIGNED_5_6_5)) { |
| 589 | return 0.f; |
| 590 | } |
| 591 | |
| 592 | if (!(a->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE)) { |
| 593 | const Context *rsc = RsdCpuReference::getTlsContext(); |
| 594 | rsc->setError(RS_ERROR_BAD_VALUE, "Sampling from texture witout USAGE_GRAPHICS_TEXTURE."); |
| 595 | return 0.f; |
| 596 | } |
| 597 | |
| 598 | if (lod <= 0.0f) { |
| 599 | if (s->mHal.state.magFilter == RS_SAMPLER_NEAREST) { |
| 600 | return sample_LOD_NearestPixel(a, type, dk, dt, s, u, 0); |
| 601 | } |
| 602 | return sample_LOD_LinearPixel(a, type, dk, dt, s, u, 0); |
| 603 | } |
| 604 | |
| 605 | if (s->mHal.state.minFilter == RS_SAMPLER_LINEAR_MIP_NEAREST) { |
| 606 | int32_t maxLOD = type->mHal.state.lodCount - 1; |
| 607 | lod = rsMin(lod, (float)maxLOD); |
| 608 | int32_t nearestLOD = (int32_t)round(lod); |
| 609 | return sample_LOD_LinearPixel(a, type, dk, dt, s, u, nearestLOD); |
| 610 | } |
| 611 | |
| 612 | if (s->mHal.state.minFilter == RS_SAMPLER_LINEAR_MIP_LINEAR) { |
| 613 | int32_t lod0 = (int32_t)floor(lod); |
| 614 | int32_t lod1 = (int32_t)ceil(lod); |
| 615 | int32_t maxLOD = type->mHal.state.lodCount - 1; |
| 616 | lod0 = rsMin(lod0, maxLOD); |
| 617 | lod1 = rsMin(lod1, maxLOD); |
| 618 | float4 sample0 = sample_LOD_LinearPixel(a, type, dk, dt, s, u, lod0); |
| 619 | float4 sample1 = sample_LOD_LinearPixel(a, type, dk, dt, s, u, lod1); |
| 620 | float frac = lod - (float)lod0; |
| 621 | return sample0 * (1.0f - frac) + sample1 * frac; |
| 622 | } |
| 623 | |
| 624 | return sample_LOD_NearestPixel(a, type, dk, dt, s, u, 0); |
| 625 | } |
| 626 | |
| 627 | static float4 GenericSample2D(Allocation *a, Sampler *s, float u, float v, float lod) { |
| 628 | const Type *type = a->getType(); |
| 629 | const Element *elem = type->getElement(); |
| 630 | const RsDataKind dk = elem->getKind(); |
| 631 | const RsDataType dt = elem->getType(); |
| 632 | |
| 633 | if (dk == RS_KIND_USER || (dt != RS_TYPE_UNSIGNED_8 && dt != RS_TYPE_UNSIGNED_5_6_5)) { |
| 634 | return 0.f; |
| 635 | } |
| 636 | |
| 637 | if (!(a->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE)) { |
| 638 | const Context *rsc = RsdCpuReference::getTlsContext(); |
| 639 | rsc->setError(RS_ERROR_BAD_VALUE, "Sampling from texture witout USAGE_GRAPHICS_TEXTURE."); |
| 640 | return 0.f; |
| 641 | } |
| 642 | |
| 643 | if (lod <= 0.0f) { |
| 644 | if (s->mHal.state.magFilter == RS_SAMPLER_NEAREST) { |
| 645 | return sample_LOD_NearestPixel(a, type, dk, dt, s, u, v, 0); |
| 646 | } |
| 647 | return sample_LOD_LinearPixel(a, type, dk, dt, s, u, v, 0); |
| 648 | } |
| 649 | |
| 650 | if (s->mHal.state.minFilter == RS_SAMPLER_LINEAR_MIP_NEAREST) { |
| 651 | int32_t maxLOD = type->mHal.state.lodCount - 1; |
| 652 | lod = rsMin(lod, (float)maxLOD); |
| 653 | int32_t nearestLOD = (int32_t)round(lod); |
| 654 | return sample_LOD_LinearPixel(a, type, dk, dt, s, u, v, nearestLOD); |
| 655 | } |
| 656 | |
| 657 | if (s->mHal.state.minFilter == RS_SAMPLER_LINEAR_MIP_LINEAR) { |
| 658 | int32_t lod0 = (int32_t)floor(lod); |
| 659 | int32_t lod1 = (int32_t)ceil(lod); |
| 660 | int32_t maxLOD = type->mHal.state.lodCount - 1; |
| 661 | lod0 = rsMin(lod0, maxLOD); |
| 662 | lod1 = rsMin(lod1, maxLOD); |
| 663 | float4 sample0 = sample_LOD_LinearPixel(a, type, dk, dt, s, u, v, lod0); |
| 664 | float4 sample1 = sample_LOD_LinearPixel(a, type, dk, dt, s, u, v, lod1); |
| 665 | float frac = lod - (float)lod0; |
| 666 | return sample0 * (1.0f - frac) + sample1 * frac; |
| 667 | } |
| 668 | |
| 669 | return sample_LOD_NearestPixel(a, type, dk, dt, s, u, v, 0); |
| 670 | } |
| 671 | |
| 672 | |
| 673 | |
| 674 | |
| 675 | // Must match pixel kind in rsDefines.h |
| 676 | static void * NearestWrap[] = { |
| 677 | (void *) GenericSample1D, // L, |
| 678 | (void *) GenericSample1D, // A, |
| 679 | (void *) GenericSample1D, // LA, |
| 680 | (void *) GenericSample1D, // RGB, |
| 681 | (void *) GenericSample1D, // RGBA, |
| 682 | 0, |
| 683 | (void *) GenericSample1D, // YUV |
| 684 | |
| 685 | (void *) GenericSample2D, // L, |
| 686 | (void *) GenericSample2D, // A, |
| 687 | (void *) GenericSample2D, // LA, |
| 688 | (void *) GenericSample2D, // RGB, |
| 689 | (void *) GenericSample2D, // RGBA, |
| 690 | 0, |
| 691 | (void *) GenericSample2D, // YUV |
| 692 | }; |
| 693 | |
| 694 | static void * NearestClamp[] = { |
| 695 | (void *) GenericSample1D, // L, |
| 696 | (void *) GenericSample1D, // A, |
| 697 | (void *) GenericSample1D, // LA, |
| 698 | (void *) GenericSample1D, // RGB, |
| 699 | (void *) GenericSample1D, // RGBA, |
| 700 | 0, |
| 701 | (void *) GenericSample1D, // YUV |
| 702 | |
| 703 | (void *) GenericSample2D, // L, |
| 704 | (void *) GenericSample2D, // A, |
| 705 | (void *) GenericSample2D, // LA, |
| 706 | (void *) GenericSample2D, // RGB, |
| 707 | (void *) GenericSample2D, // RGBA, |
| 708 | 0, |
| 709 | (void *) GenericSample2D, // YUV |
| 710 | }; |
| 711 | |
| 712 | static void * NearestMirroredRepeat[] = { |
| 713 | (void *) GenericSample1D, // L, |
| 714 | (void *) GenericSample1D, // A, |
| 715 | (void *) GenericSample1D, // LA, |
| 716 | (void *) GenericSample1D, // RGB, |
| 717 | (void *) GenericSample1D, // RGBA, |
| 718 | 0, |
| 719 | (void *) GenericSample1D, // YUV |
| 720 | |
| 721 | (void *) GenericSample2D, // L, |
| 722 | (void *) GenericSample2D, // A, |
| 723 | (void *) GenericSample2D, // LA, |
| 724 | (void *) GenericSample2D, // RGB, |
| 725 | (void *) GenericSample2D, // RGBA, |
| 726 | 0, |
| 727 | (void *) GenericSample2D, // YUV |
| 728 | }; |
| 729 | |
| 730 | // Must match pixel kind in rsDefines.h |
| 731 | static void * LinearWrap[] = { |
| 732 | (void *) GenericSample1D, // L, |
| 733 | (void *) GenericSample1D, // A, |
| 734 | (void *) GenericSample1D, // LA, |
| 735 | (void *) GenericSample1D, // RGB, |
| 736 | (void *) GenericSample1D, // RGBA, |
| 737 | 0, |
| 738 | (void *) GenericSample1D, // YUV |
| 739 | |
| 740 | (void *) GenericSample2D, // L, |
| 741 | (void *) GenericSample2D, // A, |
| 742 | (void *) GenericSample2D, // LA, |
| 743 | (void *) GenericSample2D, // RGB, |
| 744 | (void *) GenericSample2D, // RGBA, |
| 745 | 0, |
| 746 | (void *) GenericSample2D, // YUV |
| 747 | }; |
| 748 | |
| 749 | // Must match pixel kind in rsDefines.h |
| 750 | static void * LinearClamp[] = { |
| 751 | (void *) GenericSample1D, // L, |
| 752 | (void *) GenericSample1D, // A, |
| 753 | (void *) GenericSample1D, // LA, |
| 754 | (void *) GenericSample1D, // RGB, |
| 755 | (void *) GenericSample1D, // RGBA, |
| 756 | 0, |
| 757 | (void *) GenericSample1D, // YUV |
| 758 | |
| 759 | (void *) GenericSample2D, // L, |
| 760 | (void *) GenericSample2D, // A, |
| 761 | (void *) GenericSample2D, // LA, |
| 762 | (void *) GenericSample2D, // RGB, |
| 763 | (void *) GenericSample2D, // RGBA, |
| 764 | 0, |
| 765 | (void *) GenericSample2D, // YUV |
| 766 | }; |
| 767 | |
| 768 | // Must match pixel kind in rsDefines.h |
| 769 | static void * LinearMirroredRepeat[] = { |
| 770 | (void *) GenericSample1D, // L, |
| 771 | (void *) GenericSample1D, // A, |
| 772 | (void *) GenericSample1D, // LA, |
| 773 | (void *) GenericSample1D, // RGB, |
| 774 | (void *) GenericSample1D, // RGBA, |
| 775 | 0, |
| 776 | (void *) GenericSample1D, // YUV |
| 777 | |
| 778 | (void *) GenericSample2D, // L, |
| 779 | (void *) GenericSample2D, // A, |
| 780 | (void *) GenericSample2D, // LA, |
| 781 | (void *) GenericSample2D, // RGB, |
| 782 | (void *) GenericSample2D, // RGBA, |
| 783 | 0, |
| 784 | (void *) GenericSample2D, // YUV |
| 785 | }; |
| 786 | |
| 787 | // Must match pixel kind in rsDefines.h |
| 788 | static void * Generic[] = { |
| 789 | (void *) GenericSample1D, // L, |
| 790 | (void *) GenericSample1D, // A, |
| 791 | (void *) GenericSample1D, // LA, |
| 792 | (void *) GenericSample1D, // RGB, |
| 793 | (void *) GenericSample1D, // RGBA, |
| 794 | 0, |
| 795 | (void *) GenericSample1D, // YUV |
| 796 | |
| 797 | (void *) GenericSample2D, // L, |
| 798 | (void *) GenericSample2D, // A, |
| 799 | (void *) GenericSample2D, // LA, |
| 800 | (void *) GenericSample2D, // RGB, |
| 801 | (void *) GenericSample2D, // RGBA, |
| 802 | 0, |
| 803 | (void *) GenericSample2D, // YUV |
| 804 | }; |
| 805 | |
| 806 | bool rsdSamplerInit(const Context *, const Sampler *s) { |
| 807 | s->mHal.drv = Generic; |
| 808 | |
| 809 | if ((s->mHal.state.minFilter == s->mHal.state.magFilter) && |
| 810 | (s->mHal.state.wrapS == s->mHal.state.wrapT)) { |
| 811 | // We have fast paths for these. |
| 812 | |
| 813 | switch(s->mHal.state.minFilter) { |
| 814 | case RS_SAMPLER_NEAREST: |
| 815 | switch(s->mHal.state.wrapS) { |
| 816 | case RS_SAMPLER_WRAP: |
| 817 | s->mHal.drv = NearestWrap; |
| 818 | break; |
| 819 | case RS_SAMPLER_CLAMP: |
| 820 | s->mHal.drv = NearestClamp; |
| 821 | break; |
| 822 | case RS_SAMPLER_MIRRORED_REPEAT: |
| 823 | s->mHal.drv = NearestMirroredRepeat; |
| 824 | break; |
| 825 | default: |
| 826 | break; |
| 827 | } |
| 828 | break; |
| 829 | case RS_SAMPLER_LINEAR: |
| 830 | switch(s->mHal.state.wrapS) { |
| 831 | case RS_SAMPLER_WRAP: |
| 832 | s->mHal.drv = LinearWrap; |
| 833 | break; |
| 834 | case RS_SAMPLER_CLAMP: |
| 835 | s->mHal.drv = LinearClamp; |
| 836 | break; |
| 837 | case RS_SAMPLER_MIRRORED_REPEAT: |
| 838 | s->mHal.drv = LinearMirroredRepeat; |
| 839 | break; |
| 840 | default: |
| 841 | break; |
| 842 | } |
| 843 | break; |
| 844 | case RS_SAMPLER_LINEAR_MIP_LINEAR: |
| 845 | switch(s->mHal.state.wrapS) { |
| 846 | case RS_SAMPLER_WRAP: |
| 847 | s->mHal.drv = LinearWrap; |
| 848 | break; |
| 849 | case RS_SAMPLER_CLAMP: |
| 850 | s->mHal.drv = LinearClamp; |
| 851 | break; |
| 852 | case RS_SAMPLER_MIRRORED_REPEAT: |
| 853 | s->mHal.drv = LinearMirroredRepeat; |
| 854 | break; |
| 855 | default: |
| 856 | break; |
| 857 | } |
| 858 | break; |
| 859 | default: |
| 860 | rsAssert(0); |
| 861 | break; |
| 862 | } |
| 863 | |
| 864 | } |
| 865 | |
Alex Sakhartchouk | 7f126c7 | 2011-05-05 16:56:27 -0700 | [diff] [blame] | 866 | return true; |
| 867 | } |
| 868 | |
| 869 | void rsdSamplerDestroy(const android::renderscript::Context *rsc, |
| 870 | const android::renderscript::Sampler *s) { |
| 871 | } |