| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #pragma once | ||
| 2 | |||
| 3 | #include "lib8static.h" | ||
| 4 | #include "../crgb.h" | ||
| 5 | #include "../fl/namespace.h" | ||
| 6 | |||
| 7 | FASTLED_NAMESPACE_BEGIN | ||
| 8 | |||
| 9 | /// @file scale8.h | ||
| 10 | /// Fast, efficient 8-bit scaling functions specifically | ||
| 11 | /// designed for high-performance LED programming. | ||
| 12 | |||
| 13 | /// @addtogroup lib8tion | ||
| 14 | /// @{ | ||
| 15 | |||
| 16 | /// @defgroup Scaling Scaling Functions | ||
| 17 | /// Fast, efficient 8-bit scaling functions specifically | ||
| 18 | /// designed for high-performance LED programming. | ||
| 19 | /// | ||
| 20 | /// Because of the AVR(Arduino) and ARM assembly language | ||
| 21 | /// implementations provided, using these functions often | ||
| 22 | /// results in smaller and faster code than the equivalent | ||
| 23 | /// program using plain "C" arithmetic and logic. | ||
| 24 | /// @{ | ||
| 25 | |||
| 26 | /// Scale one byte by a second one, which is treated as | ||
| 27 | /// the numerator of a fraction whose denominator is 256. | ||
| 28 | /// | ||
| 29 | /// In other words, it computes i * (scale / 256) | ||
| 30 | /// @param i input value to scale | ||
| 31 | /// @param scale scale factor, in n/256 units | ||
| 32 | /// @returns scaled value | ||
| 33 | /// @note Takes 4 clocks on AVR with MUL, 2 clocks on ARM | ||
| 34 | LIB8STATIC_ALWAYS_INLINE uint8_t scale8(uint8_t i, fract8 scale) { | ||
| 35 | #if SCALE8_C == 1 | ||
| 36 | #if (FASTLED_SCALE8_FIXED == 1) | ||
| 37 | return (((uint16_t)i) * (1 + (uint16_t)(scale))) >> 8; | ||
| 38 | #else | ||
| 39 | 745200 | return ((uint16_t)i * (uint16_t)(scale)) >> 8; | |
| 40 | #endif | ||
| 41 | #elif SCALE8_AVRASM == 1 | ||
| 42 | #if defined(LIB8_ATTINY) | ||
| 43 | #if (FASTLED_SCALE8_FIXED == 1) | ||
| 44 | uint8_t work = i; | ||
| 45 | #else | ||
| 46 | uint8_t work = 0; | ||
| 47 | #endif | ||
| 48 | uint8_t cnt = 0x80; | ||
| 49 | asm volatile( | ||
| 50 | #if (FASTLED_SCALE8_FIXED == 1) | ||
| 51 | " inc %[scale] \n\t" | ||
| 52 | " breq DONE_%= \n\t" | ||
| 53 | " clr %[work] \n\t" | ||
| 54 | #endif | ||
| 55 | "LOOP_%=: \n\t" | ||
| 56 | /*" sbrc %[scale], 0 \n\t" | ||
| 57 | " add %[work], %[i] \n\t" | ||
| 58 | " ror %[work] \n\t" | ||
| 59 | " lsr %[scale] \n\t" | ||
| 60 | " clc \n\t"*/ | ||
| 61 | " sbrc %[scale], 0 \n\t" | ||
| 62 | " add %[work], %[i] \n\t" | ||
| 63 | " ror %[work] \n\t" | ||
| 64 | " lsr %[scale] \n\t" | ||
| 65 | " lsr %[cnt] \n\t" | ||
| 66 | "brcc LOOP_%= \n\t" | ||
| 67 | "DONE_%=: \n\t" | ||
| 68 | : [work] "+r"(work), [cnt] "+r"(cnt) | ||
| 69 | : [scale] "r"(scale), [i] "r"(i) | ||
| 70 | :); | ||
| 71 | return work; | ||
| 72 | #else | ||
| 73 | asm volatile( | ||
| 74 | #if (FASTLED_SCALE8_FIXED == 1) | ||
| 75 | // Multiply 8-bit i * 8-bit scale, giving 16-bit r1,r0 | ||
| 76 | "mul %0, %1 \n\t" | ||
| 77 | // Add i to r0, possibly setting the carry flag | ||
| 78 | "add r0, %0 \n\t" | ||
| 79 | // load the immediate 0 into i (note, this does _not_ touch any flags) | ||
| 80 | "ldi %0, 0x00 \n\t" | ||
| 81 | // walk and chew gum at the same time | ||
| 82 | "adc %0, r1 \n\t" | ||
| 83 | #else | ||
| 84 | /* Multiply 8-bit i * 8-bit scale, giving 16-bit r1,r0 */ | ||
| 85 | "mul %0, %1 \n\t" | ||
| 86 | /* Move the high 8-bits of the product (r1) back to i */ | ||
| 87 | "mov %0, r1 \n\t" | ||
| 88 | /* Restore r1 to "0"; it's expected to always be that */ | ||
| 89 | #endif | ||
| 90 | "clr __zero_reg__ \n\t" | ||
| 91 | |||
| 92 | : "+d"(i) /* writes to i; r16-r31, restricted by ldi */ | ||
| 93 | : "r"(scale) /* uses scale */ | ||
| 94 | : "r0", "r1" /* clobbers r0, r1 */ | ||
| 95 | ); | ||
| 96 | /* Return the result */ | ||
| 97 | return i; | ||
| 98 | #endif | ||
| 99 | #else | ||
| 100 | #error "No implementation for scale8 available." | ||
| 101 | #endif | ||
| 102 | } | ||
| 103 | |||
| 104 | constexpr uint8_t scale8_constexpr(uint8_t i, fract8 scale) { | ||
| 105 | return (((uint16_t)i) * (1 + (uint16_t)(scale))) >> 8; | ||
| 106 | } | ||
| 107 | |||
| 108 | /// The "video" version of scale8() guarantees that the output will | ||
| 109 | /// be only be zero if one or both of the inputs are zero. | ||
| 110 | /// If both inputs are non-zero, the output is guaranteed to be non-zero. | ||
| 111 | /// This makes for better "video"/LED dimming, at the cost of | ||
| 112 | /// several additional cycles. | ||
| 113 | /// @param i input value to scale | ||
| 114 | /// @param scale scale factor, in n/256 units | ||
| 115 | /// @returns scaled value | ||
| 116 | /// @see scale8() | ||
| 117 | LIB8STATIC_ALWAYS_INLINE uint8_t scale8_video(uint8_t i, fract8 scale) { | ||
| 118 | #if SCALE8_C == 1 || defined(LIB8_ATTINY) | ||
| 119 |
5/16✓ Branch 0 taken 94181 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 94181 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
✓ Branch 12 taken 1782 times.
✓ Branch 13 taken 4979 times.
✓ Branch 14 taken 1782 times.
✗ Branch 15 not taken.
|
100942 | uint8_t j = (((int)i * (int)scale) >> 8) + ((i && scale) ? 1 : 0); |
| 120 | // uint8_t nonzeroscale = (scale != 0) ? 1 : 0; | ||
| 121 | // uint8_t j = (i == 0) ? 0 : (((int)i * (int)(scale) ) >> 8) + | ||
| 122 | // nonzeroscale; | ||
| 123 | 100942 | return j; | |
| 124 | #elif SCALE8_AVRASM == 1 | ||
| 125 | uint8_t j = 0; | ||
| 126 | asm volatile(" tst %[i]\n\t" | ||
| 127 | " breq L_%=\n\t" | ||
| 128 | " mul %[i], %[scale]\n\t" | ||
| 129 | " mov %[j], r1\n\t" | ||
| 130 | " clr __zero_reg__\n\t" | ||
| 131 | " cpse %[scale], r1\n\t" | ||
| 132 | " subi %[j], 0xFF\n\t" | ||
| 133 | "L_%=: \n\t" | ||
| 134 | : [j] "+d"(j) // r16-r31, restricted by subi | ||
| 135 | : [i] "r"(i), [scale] "r"(scale) | ||
| 136 | : "r0", "r1"); | ||
| 137 | return j; | ||
| 138 | // uint8_t nonzeroscale = (scale != 0) ? 1 : 0; | ||
| 139 | // asm volatile( | ||
| 140 | // " tst %0 \n" | ||
| 141 | // " breq L_%= \n" | ||
| 142 | // " mul %0, %1 \n" | ||
| 143 | // " mov %0, r1 \n" | ||
| 144 | // " add %0, %2 \n" | ||
| 145 | // " clr __zero_reg__ \n" | ||
| 146 | // "L_%=: \n" | ||
| 147 | // : "+a" (i) | ||
| 148 | // : "a" (scale), "a" (nonzeroscale) | ||
| 149 | // : "r0", "r1"); | ||
| 150 | // // Return the result | ||
| 151 | // return i; | ||
| 152 | #else | ||
| 153 | #error "No implementation for scale8_video available." | ||
| 154 | #endif | ||
| 155 | } | ||
| 156 | |||
| 157 | /// @defgroup ScalingDirty Scaling Functions that Leave R1 Dirty | ||
| 158 | /// These functions are more efficient for scaling multiple | ||
| 159 | /// bytes at once, but require calling cleanup_R1() afterwards. | ||
| 160 | /// @{ | ||
| 161 | |||
| 162 | /// This version of scale8() does not clean up the R1 register on AVR. | ||
| 163 | /// If you are doing several "scale8()'s" in a row, use this, and | ||
| 164 | /// then explicitly call cleanup_R1(). | ||
| 165 | /// @warning You **MUST** call cleanup_R1() after using this function! | ||
| 166 | /// @param i input value to scale | ||
| 167 | /// @param scale scale factor, in n/256 units | ||
| 168 | /// @returns scaled value | ||
| 169 | /// @see scale8() | ||
| 170 | LIB8STATIC_ALWAYS_INLINE uint8_t scale8_LEAVING_R1_DIRTY(uint8_t i, | ||
| 171 | fract8 scale) { | ||
| 172 | #if SCALE8_C == 1 | ||
| 173 | #if (FASTLED_SCALE8_FIXED == 1) | ||
| 174 | return (((uint16_t)i) * ((uint16_t)(scale) + 1)) >> 8; | ||
| 175 | #else | ||
| 176 | 160409 | return ((int)i * (int)(scale)) >> 8; | |
| 177 | #endif | ||
| 178 | #elif SCALE8_AVRASM == 1 | ||
| 179 | asm volatile( | ||
| 180 | #if (FASTLED_SCALE8_FIXED == 1) | ||
| 181 | // Multiply 8-bit i * 8-bit scale, giving 16-bit r1,r0 | ||
| 182 | "mul %0, %1 \n\t" | ||
| 183 | // Add i to r0, possibly setting the carry flag | ||
| 184 | "add r0, %0 \n\t" | ||
| 185 | // load the immediate 0 into i (note, this does _not_ touch any flags) | ||
| 186 | "ldi %0, 0x00 \n\t" | ||
| 187 | // walk and chew gum at the same time | ||
| 188 | "adc %0, r1 \n\t" | ||
| 189 | #else | ||
| 190 | /* Multiply 8-bit i * 8-bit scale, giving 16-bit r1,r0 */ | ||
| 191 | "mul %0, %1 \n\t" | ||
| 192 | /* Move the high 8-bits of the product (r1) back to i */ | ||
| 193 | "mov %0, r1 \n\t" | ||
| 194 | #endif | ||
| 195 | /* R1 IS LEFT DIRTY HERE; YOU MUST ZERO IT OUT YOURSELF */ | ||
| 196 | /* "clr __zero_reg__ \n\t" */ | ||
| 197 | : "+d"(i) /* writes to i; r16-r31, restricted by ldi */ | ||
| 198 | : "r"(scale) /* uses scale */ | ||
| 199 | : "r0", "r1" /* clobbers r0, r1 */ | ||
| 200 | ); | ||
| 201 | // Return the result | ||
| 202 | return i; | ||
| 203 | #else | ||
| 204 | #error "No implementation for scale8_LEAVING_R1_DIRTY available." | ||
| 205 | #endif | ||
| 206 | } | ||
| 207 | |||
| 208 | /// In place modifying version of scale8() that does not clean up the R1 | ||
| 209 | /// register on AVR. If you are doing several "scale8()'s" in a row, use this, | ||
| 210 | /// and then explicitly call cleanup_R1(). | ||
| 211 | /// @warning You **MUST** call cleanup_R1() after using this function! | ||
| 212 | /// @par | ||
| 213 | /// @warning This function always modifies its arguments in place! | ||
| 214 | /// @param i input value to scale | ||
| 215 | /// @param scale scale factor, in n/256 units | ||
| 216 | /// @see scale8() | ||
| 217 | LIB8STATIC_ALWAYS_INLINE void nscale8_LEAVING_R1_DIRTY(uint8_t &i, | ||
| 218 | fract8 scale) { | ||
| 219 | #if SCALE8_C == 1 | ||
| 220 | #if (FASTLED_SCALE8_FIXED == 1) | ||
| 221 | i = (((uint16_t)i) * ((uint16_t)(scale) + 1)) >> 8; | ||
| 222 | #else | ||
| 223 | i = ((int)i * (int)(scale)) >> 8; | ||
| 224 | #endif | ||
| 225 | #elif SCALE8_AVRASM == 1 | ||
| 226 | asm volatile( | ||
| 227 | #if (FASTLED_SCALE8_FIXED == 1) | ||
| 228 | // Multiply 8-bit i * 8-bit scale, giving 16-bit r1,r0 | ||
| 229 | "mul %0, %1 \n\t" | ||
| 230 | // Add i to r0, possibly setting the carry flag | ||
| 231 | "add r0, %0 \n\t" | ||
| 232 | // load the immediate 0 into i (note, this does _not_ touch any flags) | ||
| 233 | "ldi %0, 0x00 \n\t" | ||
| 234 | // walk and chew gum at the same time | ||
| 235 | "adc %0, r1 \n\t" | ||
| 236 | #else | ||
| 237 | /* Multiply 8-bit i * 8-bit scale, giving 16-bit r1,r0 */ | ||
| 238 | "mul %0, %1 \n\t" | ||
| 239 | /* Move the high 8-bits of the product (r1) back to i */ | ||
| 240 | "mov %0, r1 \n\t" | ||
| 241 | #endif | ||
| 242 | /* R1 IS LEFT DIRTY HERE; YOU MUST ZERO IT OUT YOURSELF */ | ||
| 243 | /* "clr __zero_reg__ \n\t" */ | ||
| 244 | |||
| 245 | : "+d"(i) /* writes to i; r16-r31, restricted by ldi */ | ||
| 246 | : "r"(scale) /* uses scale */ | ||
| 247 | : "r0", "r1" /* clobbers r0, r1 */ | ||
| 248 | ); | ||
| 249 | #else | ||
| 250 | #error "No implementation for nscale8_LEAVING_R1_DIRTY available." | ||
| 251 | #endif | ||
| 252 | } | ||
| 253 | |||
| 254 | /// This version of scale8_video() does not clean up the R1 register on AVR. | ||
| 255 | /// If you are doing several "scale8_video()'s" in a row, use this, and | ||
| 256 | /// then explicitly call cleanup_R1(). | ||
| 257 | /// @warning You **MUST** call cleanup_R1() after using this function! | ||
| 258 | /// @param i input value to scale | ||
| 259 | /// @param scale scale factor, in n/256 units | ||
| 260 | /// @returns scaled value | ||
| 261 | /// @see scale8_video() | ||
| 262 | LIB8STATIC_ALWAYS_INLINE uint8_t scale8_video_LEAVING_R1_DIRTY(uint8_t i, | ||
| 263 | fract8 scale) { | ||
| 264 | #if SCALE8_C == 1 || defined(LIB8_ATTINY) | ||
| 265 |
3/12✓ Branch 0 taken 104632 times.
✓ Branch 1 taken 2875 times.
✓ Branch 2 taken 104632 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
|
107507 | uint8_t j = (((int)i * (int)scale) >> 8) + ((i && scale) ? 1 : 0); |
| 266 | // uint8_t nonzeroscale = (scale != 0) ? 1 : 0; | ||
| 267 | // uint8_t j = (i == 0) ? 0 : (((int)i * (int)(scale) ) >> 8) + | ||
| 268 | // nonzeroscale; | ||
| 269 | 107507 | return j; | |
| 270 | #elif SCALE8_AVRASM == 1 | ||
| 271 | uint8_t j = 0; | ||
| 272 | asm volatile(" tst %[i]\n\t" | ||
| 273 | " breq L_%=\n\t" | ||
| 274 | " mul %[i], %[scale]\n\t" | ||
| 275 | " mov %[j], r1\n\t" | ||
| 276 | " breq L_%=\n\t" | ||
| 277 | " subi %[j], 0xFF\n\t" | ||
| 278 | "L_%=: \n\t" | ||
| 279 | : [j] "+d"(j) // r16-r31, restricted by subi | ||
| 280 | : [i] "r"(i), [scale] "r"(scale) | ||
| 281 | : "r0", "r1"); | ||
| 282 | return j; | ||
| 283 | // uint8_t nonzeroscale = (scale != 0) ? 1 : 0; | ||
| 284 | // asm volatile( | ||
| 285 | // " tst %0 \n" | ||
| 286 | // " breq L_%= \n" | ||
| 287 | // " mul %0, %1 \n" | ||
| 288 | // " mov %0, r1 \n" | ||
| 289 | // " add %0, %2 \n" | ||
| 290 | // " clr __zero_reg__ \n" | ||
| 291 | // "L_%=: \n" | ||
| 292 | // : "+a" (i) | ||
| 293 | // : "a" (scale), "a" (nonzeroscale) | ||
| 294 | // : "r0", "r1"); | ||
| 295 | // // Return the result | ||
| 296 | // return i; | ||
| 297 | #else | ||
| 298 | #error "No implementation for scale8_video_LEAVING_R1_DIRTY available." | ||
| 299 | #endif | ||
| 300 | } | ||
| 301 | |||
| 302 | /// In place modifying version of scale8_video() that does not clean up the R1 | ||
| 303 | /// register on AVR. If you are doing several "scale8_video()'s" in a row, use | ||
| 304 | /// this, and then explicitly call cleanup_R1(). | ||
| 305 | /// @warning You **MUST** call cleanup_R1() after using this function! | ||
| 306 | /// @par | ||
| 307 | /// @warning This function always modifies its arguments in place! | ||
| 308 | /// @param i input value to scale | ||
| 309 | /// @param scale scale factor, in n/256 units | ||
| 310 | /// @see scale8_video() | ||
| 311 | LIB8STATIC_ALWAYS_INLINE void nscale8_video_LEAVING_R1_DIRTY(uint8_t &i, | ||
| 312 | fract8 scale) { | ||
| 313 | #if SCALE8_C == 1 || defined(LIB8_ATTINY) | ||
| 314 | i = (((int)i * (int)scale) >> 8) + ((i && scale) ? 1 : 0); | ||
| 315 | #elif SCALE8_AVRASM == 1 | ||
| 316 | asm volatile(" tst %[i]\n\t" | ||
| 317 | " breq L_%=\n\t" | ||
| 318 | " mul %[i], %[scale]\n\t" | ||
| 319 | " mov %[i], r1\n\t" | ||
| 320 | " breq L_%=\n\t" | ||
| 321 | " subi %[i], 0xFF\n\t" | ||
| 322 | "L_%=: \n\t" | ||
| 323 | : [i] "+d"(i) // r16-r31, restricted by subi | ||
| 324 | : [scale] "r"(scale) | ||
| 325 | : "r0", "r1"); | ||
| 326 | #else | ||
| 327 | #error "No implementation for scale8_video_LEAVING_R1_DIRTY available." | ||
| 328 | #endif | ||
| 329 | } | ||
| 330 | |||
| 331 | /// Clean up the r1 register after a series of *LEAVING_R1_DIRTY calls | ||
| 332 | /// @ingroup ScalingDirty | ||
| 333 | LIB8STATIC_ALWAYS_INLINE void cleanup_R1() { | ||
| 334 | #if CLEANUP_R1_AVRASM == 1 | ||
| 335 | // Restore r1 to "0"; it's expected to always be that | ||
| 336 | asm volatile("clr __zero_reg__ \n\t" : : : "r1"); | ||
| 337 | #endif | ||
| 338 | 22625 | } | |
| 339 | |||
| 340 | constexpr CRGB nscale8x3_constexpr(uint8_t r, uint8_t g, uint8_t b, fract8 scale) { | ||
| 341 | return CRGB(((int)r * (int)(scale)) >> 8, ((int)g * (int)(scale)) >> 8, | ||
| 342 | ((int)b * (int)(scale)) >> 8); | ||
| 343 | } | ||
| 344 | |||
| 345 | /// @} ScalingDirty | ||
| 346 | |||
| 347 | /// Scale three one-byte values by a fourth one, which is treated as | ||
| 348 | /// the numerator of a fraction whose demominator is 256. | ||
| 349 | /// | ||
| 350 | /// In other words, it computes r,g,b * (scale / 256) | ||
| 351 | /// | ||
| 352 | /// @warning This function always modifies its arguments in place! | ||
| 353 | /// @param r first value to scale | ||
| 354 | /// @param g second value to scale | ||
| 355 | /// @param b third value to scale | ||
| 356 | /// @param scale scale factor, in n/256 units | ||
| 357 | 27063 | LIB8STATIC void nscale8x3(uint8_t &r, uint8_t &g, uint8_t &b, fract8 scale) { | |
| 358 | #if SCALE8_C == 1 | ||
| 359 | #if (FASTLED_SCALE8_FIXED == 1) | ||
| 360 | uint16_t scale_fixed = scale + 1; | ||
| 361 | r = (((uint16_t)r) * scale_fixed) >> 8; | ||
| 362 | g = (((uint16_t)g) * scale_fixed) >> 8; | ||
| 363 | b = (((uint16_t)b) * scale_fixed) >> 8; | ||
| 364 | #else | ||
| 365 | 27063 | r = ((int)r * (int)(scale)) >> 8; | |
| 366 | 27063 | g = ((int)g * (int)(scale)) >> 8; | |
| 367 | 27063 | b = ((int)b * (int)(scale)) >> 8; | |
| 368 | #endif | ||
| 369 | #elif SCALE8_AVRASM == 1 | ||
| 370 | r = scale8_LEAVING_R1_DIRTY(r, scale); | ||
| 371 | g = scale8_LEAVING_R1_DIRTY(g, scale); | ||
| 372 | b = scale8_LEAVING_R1_DIRTY(b, scale); | ||
| 373 | cleanup_R1(); | ||
| 374 | #else | ||
| 375 | #error "No implementation for nscale8x3 available." | ||
| 376 | #endif | ||
| 377 | 27063 | } | |
| 378 | |||
| 379 | /// Scale three one-byte values by a fourth one, which is treated as | ||
| 380 | /// the numerator of a fraction whose demominator is 256. | ||
| 381 | /// | ||
| 382 | /// In other words, it computes r,g,b * (scale / 256), ensuring | ||
| 383 | /// that non-zero values passed in remain non-zero, no matter how low the scale | ||
| 384 | /// argument. | ||
| 385 | /// | ||
| 386 | /// @warning This function always modifies its arguments in place! | ||
| 387 | /// @param r first value to scale | ||
| 388 | /// @param g second value to scale | ||
| 389 | /// @param b third value to scale | ||
| 390 | /// @param scale scale factor, in n/256 units | ||
| 391 | 46 | LIB8STATIC void nscale8x3_video(uint8_t &r, uint8_t &g, uint8_t &b, | |
| 392 | fract8 scale) { | ||
| 393 | #if SCALE8_C == 1 | ||
| 394 |
1/2✓ Branch 0 taken 46 times.
✗ Branch 1 not taken.
|
46 | uint8_t nonzeroscale = (scale != 0) ? 1 : 0; |
| 395 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 46 times.
|
46 | r = (r == 0) ? 0 : (((int)r * (int)(scale)) >> 8) + nonzeroscale; |
| 396 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 46 times.
|
46 | g = (g == 0) ? 0 : (((int)g * (int)(scale)) >> 8) + nonzeroscale; |
| 397 |
1/2✓ Branch 0 taken 46 times.
✗ Branch 1 not taken.
|
46 | b = (b == 0) ? 0 : (((int)b * (int)(scale)) >> 8) + nonzeroscale; |
| 398 | #elif SCALE8_AVRASM == 1 | ||
| 399 | nscale8_video_LEAVING_R1_DIRTY(r, scale); | ||
| 400 | nscale8_video_LEAVING_R1_DIRTY(g, scale); | ||
| 401 | nscale8_video_LEAVING_R1_DIRTY(b, scale); | ||
| 402 | cleanup_R1(); | ||
| 403 | #else | ||
| 404 | #error "No implementation for nscale8x3 available." | ||
| 405 | #endif | ||
| 406 | 46 | } | |
| 407 | |||
| 408 | /// Scale two one-byte values by a third one, which is treated as | ||
| 409 | /// the numerator of a fraction whose demominator is 256. | ||
| 410 | /// | ||
| 411 | /// In other words, it computes i,j * (scale / 256). | ||
| 412 | /// | ||
| 413 | /// @warning This function always modifies its arguments in place! | ||
| 414 | /// @param i first value to scale | ||
| 415 | /// @param j second value to scale | ||
| 416 | /// @param scale scale factor, in n/256 units | ||
| 417 | LIB8STATIC void nscale8x2(uint8_t &i, uint8_t &j, fract8 scale) { | ||
| 418 | #if SCALE8_C == 1 | ||
| 419 | #if FASTLED_SCALE8_FIXED == 1 | ||
| 420 | uint16_t scale_fixed = scale + 1; | ||
| 421 | i = (((uint16_t)i) * scale_fixed) >> 8; | ||
| 422 | j = (((uint16_t)j) * scale_fixed) >> 8; | ||
| 423 | #else | ||
| 424 | i = ((uint16_t)i * (uint16_t)(scale)) >> 8; | ||
| 425 | j = ((uint16_t)j * (uint16_t)(scale)) >> 8; | ||
| 426 | #endif | ||
| 427 | #elif SCALE8_AVRASM == 1 | ||
| 428 | i = scale8_LEAVING_R1_DIRTY(i, scale); | ||
| 429 | j = scale8_LEAVING_R1_DIRTY(j, scale); | ||
| 430 | cleanup_R1(); | ||
| 431 | #else | ||
| 432 | #error "No implementation for nscale8x2 available." | ||
| 433 | #endif | ||
| 434 | } | ||
| 435 | |||
| 436 | /// Scale two one-byte values by a third one, which is treated as | ||
| 437 | /// the numerator of a fraction whose demominator is 256. | ||
| 438 | /// | ||
| 439 | /// In other words, it computes i,j * (scale / 256), ensuring | ||
| 440 | /// that non-zero values passed in remain non zero, no matter how low the scale | ||
| 441 | /// argument. | ||
| 442 | /// | ||
| 443 | /// @warning This function always modifies its arguments in place! | ||
| 444 | /// @param i first value to scale | ||
| 445 | /// @param j second value to scale | ||
| 446 | /// @param scale scale factor, in n/256 units | ||
| 447 | LIB8STATIC void nscale8x2_video(uint8_t &i, uint8_t &j, fract8 scale) { | ||
| 448 | #if SCALE8_C == 1 | ||
| 449 | uint8_t nonzeroscale = (scale != 0) ? 1 : 0; | ||
| 450 | i = (i == 0) ? 0 : (((int)i * (int)(scale)) >> 8) + nonzeroscale; | ||
| 451 | j = (j == 0) ? 0 : (((int)j * (int)(scale)) >> 8) + nonzeroscale; | ||
| 452 | #elif SCALE8_AVRASM == 1 | ||
| 453 | nscale8_video_LEAVING_R1_DIRTY(i, scale); | ||
| 454 | nscale8_video_LEAVING_R1_DIRTY(j, scale); | ||
| 455 | cleanup_R1(); | ||
| 456 | #else | ||
| 457 | #error "No implementation for nscale8x2 available." | ||
| 458 | #endif | ||
| 459 | } | ||
| 460 | |||
| 461 | /// Scale a 16-bit unsigned value by an 8-bit value, which is treated | ||
| 462 | /// as the numerator of a fraction whose denominator is 256. | ||
| 463 | /// | ||
| 464 | /// In other words, it computes i * (scale / 256) | ||
| 465 | /// @param i input value to scale | ||
| 466 | /// @param scale scale factor, in n/256 units | ||
| 467 | /// @returns scaled value | ||
| 468 | LIB8STATIC_ALWAYS_INLINE uint16_t scale16by8(uint16_t i, fract8 scale) { | ||
| 469 | if (scale == 0) { | ||
| 470 | return 0; // Fixes non zero output when scale == 0 and | ||
| 471 | // FASTLED_SCALE8_FIXED==1 | ||
| 472 | } | ||
| 473 | #if SCALE16BY8_C == 1 | ||
| 474 | uint16_t result; | ||
| 475 | #if FASTLED_SCALE8_FIXED == 1 | ||
| 476 | result = (((uint32_t)(i) * (1 + ((uint32_t)scale))) >> 8); | ||
| 477 | #else | ||
| 478 | result = (i * scale) / 256; | ||
| 479 | #endif | ||
| 480 | return result; | ||
| 481 | #elif SCALE16BY8_AVRASM == 1 | ||
| 482 | #if FASTLED_SCALE8_FIXED == 1 | ||
| 483 | uint16_t result = 0; | ||
| 484 | asm volatile( | ||
| 485 | // result.A = HighByte( (i.A x scale) + i.A ) | ||
| 486 | " mul %A[i], %[scale] \n\t" | ||
| 487 | " add r0, %A[i] \n\t" | ||
| 488 | // " adc r1, [zero] \n\t" | ||
| 489 | // " mov %A[result], r1 \n\t" | ||
| 490 | " adc %A[result], r1 \n\t" | ||
| 491 | |||
| 492 | // result.A-B += i.B x scale | ||
| 493 | " mul %B[i], %[scale] \n\t" | ||
| 494 | " add %A[result], r0 \n\t" | ||
| 495 | " adc %B[result], r1 \n\t" | ||
| 496 | |||
| 497 | // cleanup r1 | ||
| 498 | " clr __zero_reg__ \n\t" | ||
| 499 | |||
| 500 | // result.A-B += i.B | ||
| 501 | " add %A[result], %B[i] \n\t" | ||
| 502 | " adc %B[result], __zero_reg__ \n\t" | ||
| 503 | |||
| 504 | : [result] "+r"(result) | ||
| 505 | : [i] "r"(i), [scale] "r"(scale) | ||
| 506 | : "r0", "r1"); | ||
| 507 | return result; | ||
| 508 | #else | ||
| 509 | uint16_t result = 0; | ||
| 510 | asm volatile( | ||
| 511 | // result.A = HighByte(i.A x j ) | ||
| 512 | " mul %A[i], %[scale] \n\t" | ||
| 513 | " mov %A[result], r1 \n\t" | ||
| 514 | //" clr %B[result] \n\t" | ||
| 515 | |||
| 516 | // result.A-B += i.B x j | ||
| 517 | " mul %B[i], %[scale] \n\t" | ||
| 518 | " add %A[result], r0 \n\t" | ||
| 519 | " adc %B[result], r1 \n\t" | ||
| 520 | |||
| 521 | // cleanup r1 | ||
| 522 | " clr __zero_reg__ \n\t" | ||
| 523 | |||
| 524 | : [result] "+r"(result) | ||
| 525 | : [i] "r"(i), [scale] "r"(scale) | ||
| 526 | : "r0", "r1"); | ||
| 527 | return result; | ||
| 528 | #endif | ||
| 529 | #else | ||
| 530 | #error "No implementation for scale16by8 available." | ||
| 531 | #endif | ||
| 532 | } | ||
| 533 | |||
| 534 | /// Scale a 16-bit unsigned value by an 16-bit value, which is treated | ||
| 535 | /// as the numerator of a fraction whose denominator is 65536. | ||
| 536 | /// In other words, it computes i * (scale / 65536) | ||
| 537 | /// @param i input value to scale | ||
| 538 | /// @param scale scale factor, in n/65536 units | ||
| 539 | /// @returns scaled value | ||
| 540 | 82662 | LIB8STATIC uint16_t scale16(uint16_t i, fract16 scale) { | |
| 541 | #if SCALE16_C == 1 | ||
| 542 | uint16_t result; | ||
| 543 | #if FASTLED_SCALE8_FIXED == 1 | ||
| 544 | result = ((uint32_t)(i) * (1 + (uint32_t)(scale))) / 65536; | ||
| 545 | #else | ||
| 546 | 82662 | result = ((uint32_t)(i) * (uint32_t)(scale)) / 65536; | |
| 547 | #endif | ||
| 548 | 82662 | return result; | |
| 549 | #elif SCALE16_AVRASM == 1 | ||
| 550 | #if FASTLED_SCALE8_FIXED == 1 | ||
| 551 | // implemented sort of like | ||
| 552 | // result = ((i * scale) + i ) / 65536 | ||
| 553 | // | ||
| 554 | // why not like this, you may ask? | ||
| 555 | // result = (i * (scale+1)) / 65536 | ||
| 556 | // the answer is that if scale is 65535, then scale+1 | ||
| 557 | // will be zero, which is not what we want. | ||
| 558 | uint32_t result; | ||
| 559 | asm volatile( | ||
| 560 | // result.A-B = i.A x scale.A | ||
| 561 | " mul %A[i], %A[scale] \n\t" | ||
| 562 | // save results... | ||
| 563 | // basic idea: | ||
| 564 | //" mov %A[result], r0 \n\t" | ||
| 565 | //" mov %B[result], r1 \n\t" | ||
| 566 | // which can be written as... | ||
| 567 | " movw %A[result], r0 \n\t" | ||
| 568 | // Because we're going to add i.A-B to | ||
| 569 | // result.A-D, we DO need to keep both | ||
| 570 | // the r0 and r1 portions of the product | ||
| 571 | // UNlike in the 'unfixed scale8' version. | ||
| 572 | // So the movw here is needed. | ||
| 573 | : [result] "=r"(result) | ||
| 574 | : [i] "r"(i), [scale] "r"(scale) | ||
| 575 | : "r0", "r1"); | ||
| 576 | |||
| 577 | asm volatile( | ||
| 578 | // result.C-D = i.B x scale.B | ||
| 579 | " mul %B[i], %B[scale] \n\t" | ||
| 580 | //" mov %C[result], r0 \n\t" | ||
| 581 | //" mov %D[result], r1 \n\t" | ||
| 582 | " movw %C[result], r0 \n\t" | ||
| 583 | : [result] "+r"(result) | ||
| 584 | : [i] "r"(i), [scale] "r"(scale) | ||
| 585 | : "r0", "r1"); | ||
| 586 | |||
| 587 | const uint8_t zero = 0; | ||
| 588 | asm volatile( | ||
| 589 | // result.B-D += i.B x scale.A | ||
| 590 | " mul %B[i], %A[scale] \n\t" | ||
| 591 | |||
| 592 | " add %B[result], r0 \n\t" | ||
| 593 | " adc %C[result], r1 \n\t" | ||
| 594 | " adc %D[result], %[zero] \n\t" | ||
| 595 | |||
| 596 | // result.B-D += i.A x scale.B | ||
| 597 | " mul %A[i], %B[scale] \n\t" | ||
| 598 | |||
| 599 | " add %B[result], r0 \n\t" | ||
| 600 | " adc %C[result], r1 \n\t" | ||
| 601 | " adc %D[result], %[zero] \n\t" | ||
| 602 | |||
| 603 | // cleanup r1 | ||
| 604 | " clr r1 \n\t" | ||
| 605 | |||
| 606 | : [result] "+r"(result) | ||
| 607 | : [i] "r"(i), [scale] "r"(scale), [zero] "r"(zero) | ||
| 608 | : "r0", "r1"); | ||
| 609 | |||
| 610 | asm volatile( | ||
| 611 | // result.A-D += i.A-B | ||
| 612 | " add %A[result], %A[i] \n\t" | ||
| 613 | " adc %B[result], %B[i] \n\t" | ||
| 614 | " adc %C[result], %[zero] \n\t" | ||
| 615 | " adc %D[result], %[zero] \n\t" | ||
| 616 | : [result] "+r"(result) | ||
| 617 | : [i] "r"(i), [zero] "r"(zero)); | ||
| 618 | |||
| 619 | result = result >> 16; | ||
| 620 | return result; | ||
| 621 | #else | ||
| 622 | uint32_t result; | ||
| 623 | asm volatile( | ||
| 624 | // result.A-B = i.A x scale.A | ||
| 625 | " mul %A[i], %A[scale] \n\t" | ||
| 626 | // save results... | ||
| 627 | // basic idea: | ||
| 628 | //" mov %A[result], r0 \n\t" | ||
| 629 | //" mov %B[result], r1 \n\t" | ||
| 630 | // which can be written as... | ||
| 631 | " movw %A[result], r0 \n\t" | ||
| 632 | // We actually don't need to do anything with r0, | ||
| 633 | // as result.A is never used again here, so we | ||
| 634 | // could just move the high byte, but movw is | ||
| 635 | // one clock cycle, just like mov, so might as | ||
| 636 | // well, in case we want to use this code for | ||
| 637 | // a generic 16x16 multiply somewhere. | ||
| 638 | |||
| 639 | : [result] "=r"(result) | ||
| 640 | : [i] "r"(i), [scale] "r"(scale) | ||
| 641 | : "r0", "r1"); | ||
| 642 | |||
| 643 | asm volatile( | ||
| 644 | // result.C-D = i.B x scale.B | ||
| 645 | " mul %B[i], %B[scale] \n\t" | ||
| 646 | //" mov %C[result], r0 \n\t" | ||
| 647 | //" mov %D[result], r1 \n\t" | ||
| 648 | " movw %C[result], r0 \n\t" | ||
| 649 | : [result] "+r"(result) | ||
| 650 | : [i] "r"(i), [scale] "r"(scale) | ||
| 651 | : "r0", "r1"); | ||
| 652 | |||
| 653 | const uint8_t zero = 0; | ||
| 654 | asm volatile( | ||
| 655 | // result.B-D += i.B x scale.A | ||
| 656 | " mul %B[i], %A[scale] \n\t" | ||
| 657 | |||
| 658 | " add %B[result], r0 \n\t" | ||
| 659 | " adc %C[result], r1 \n\t" | ||
| 660 | " adc %D[result], %[zero] \n\t" | ||
| 661 | |||
| 662 | // result.B-D += i.A x scale.B | ||
| 663 | " mul %A[i], %B[scale] \n\t" | ||
| 664 | |||
| 665 | " add %B[result], r0 \n\t" | ||
| 666 | " adc %C[result], r1 \n\t" | ||
| 667 | " adc %D[result], %[zero] \n\t" | ||
| 668 | |||
| 669 | // cleanup r1 | ||
| 670 | " clr r1 \n\t" | ||
| 671 | |||
| 672 | : [result] "+r"(result) | ||
| 673 | : [i] "r"(i), [scale] "r"(scale), [zero] "r"(zero) | ||
| 674 | : "r0", "r1"); | ||
| 675 | |||
| 676 | result = result >> 16; | ||
| 677 | return result; | ||
| 678 | #endif | ||
| 679 | #else | ||
| 680 | #error "No implementation for scale16 available." | ||
| 681 | #endif | ||
| 682 | } | ||
| 683 | /// @} Scaling | ||
| 684 | |||
| 685 | /// @defgroup Dimming Dimming and Brightening Functions | ||
| 686 | /// Functions to dim or brighten data. | ||
| 687 | /// | ||
| 688 | /// The eye does not respond in a linear way to light. | ||
| 689 | /// High speed PWM'd LEDs at 50% duty cycle appear far | ||
| 690 | /// brighter then the "half as bright" you might expect. | ||
| 691 | /// | ||
| 692 | /// If you want your midpoint brightness LEDs (128) to | ||
| 693 | /// appear half as bright as "full" brightness (255), you | ||
| 694 | /// have to apply a "dimming function". | ||
| 695 | /// | ||
| 696 | /// @note These are approximations of gamma correction with | ||
| 697 | /// a gamma value of 2.0. | ||
| 698 | /// @see @ref GammaFuncs | ||
| 699 | /// @{ | ||
| 700 | |||
| 701 | /// Adjust a scaling value for dimming. | ||
| 702 | /// @see scale8() | ||
| 703 | LIB8STATIC uint8_t dim8_raw(uint8_t x) { return scale8(x, x); } | ||
| 704 | |||
| 705 | /// Adjust a scaling value for dimming for video (value will never go below 1) | ||
| 706 | /// @see scale8_video() | ||
| 707 | LIB8STATIC uint8_t dim8_video(uint8_t x) { return scale8_video(x, x); } | ||
| 708 | |||
| 709 | /// Linear version of the dimming function that halves for values < 128 | ||
| 710 | LIB8STATIC uint8_t dim8_lin(uint8_t x) { | ||
| 711 | if (x & 0x80) { | ||
| 712 | x = scale8(x, x); | ||
| 713 | } else { | ||
| 714 | x += 1; | ||
| 715 | x /= 2; | ||
| 716 | } | ||
| 717 | return x; | ||
| 718 | } | ||
| 719 | |||
| 720 | /// Brighten a value (inverse of dim8_raw()) | ||
| 721 | LIB8STATIC uint8_t brighten8_raw(uint8_t x) { | ||
| 722 | uint8_t ix = 255 - x; | ||
| 723 | return 255 - scale8(ix, ix); | ||
| 724 | } | ||
| 725 | |||
| 726 | /// Brighten a value (inverse of dim8_video()) | ||
| 727 | LIB8STATIC uint8_t brighten8_video(uint8_t x) { | ||
| 728 | uint8_t ix = 255 - x; | ||
| 729 | return 255 - scale8_video(ix, ix); | ||
| 730 | } | ||
| 731 | |||
| 732 | /// Brighten a value (inverse of dim8_lin()) | ||
| 733 | LIB8STATIC uint8_t brighten8_lin(uint8_t x) { | ||
| 734 | uint8_t ix = 255 - x; | ||
| 735 | if (ix & 0x80) { | ||
| 736 | ix = scale8(ix, ix); | ||
| 737 | } else { | ||
| 738 | ix += 1; | ||
| 739 | ix /= 2; | ||
| 740 | } | ||
| 741 | return 255 - ix; | ||
| 742 | } | ||
| 743 | |||
| 744 | /// @} Dimming | ||
| 745 | /// @} lib8tion | ||
| 746 | |||
| 747 | FASTLED_NAMESPACE_END | ||
| 748 |