| 1 | // Copyright 2014 Google Inc. All Rights Reserved. |
| 2 | // |
| 3 | // Use of this source code is governed by a BSD-style license |
| 4 | // that can be found in the COPYING file in the root of the source |
| 5 | // tree. An additional intellectual property rights grant can be found |
| 6 | // in the file PATENTS. All contributing project authors may |
| 7 | // be found in the AUTHORS file in the root of the source tree. |
| 8 | // ----------------------------------------------------------------------------- |
| 9 | // |
| 10 | // ARGB making functions (mips version). |
| 11 | // |
| 12 | // Author: Djordje Pesut (djordje.pesut@imgtec.com) |
| 13 | |
| 14 | #include "./dsp.h" |
| 15 | |
| 16 | #if defined(WEBP_USE_MIPS_DSP_R2) |
| 17 | |
| 18 | static void PackARGB(const uint8_t* a, const uint8_t* r, const uint8_t* g, |
| 19 | const uint8_t* b, int len, uint32_t* out) { |
| 20 | int temp0, temp1, temp2, temp3, offset; |
| 21 | const int rest = len & 1; |
| 22 | const uint32_t* const loop_end = out + len - rest; |
| 23 | const int step = 4; |
| 24 | __asm__ volatile ( |
| 25 | "xor %[offset], %[offset], %[offset] \n\t" |
| 26 | "beq %[loop_end], %[out], 0f \n\t" |
| 27 | "2: \n\t" |
| 28 | "lbux %[temp0], %[offset](%[a]) \n\t" |
| 29 | "lbux %[temp1], %[offset](%[r]) \n\t" |
| 30 | "lbux %[temp2], %[offset](%[g]) \n\t" |
| 31 | "lbux %[temp3], %[offset](%[b]) \n\t" |
| 32 | "ins %[temp1], %[temp0], 16, 16 \n\t" |
| 33 | "ins %[temp3], %[temp2], 16, 16 \n\t" |
| 34 | "addiu %[out], %[out], 4 \n\t" |
| 35 | "precr.qb.ph %[temp0], %[temp1], %[temp3] \n\t" |
| 36 | "sw %[temp0], -4(%[out]) \n\t" |
| 37 | "addu %[offset], %[offset], %[step] \n\t" |
| 38 | "bne %[loop_end], %[out], 2b \n\t" |
| 39 | "0: \n\t" |
| 40 | "beq %[rest], $zero, 1f \n\t" |
| 41 | "lbux %[temp0], %[offset](%[a]) \n\t" |
| 42 | "lbux %[temp1], %[offset](%[r]) \n\t" |
| 43 | "lbux %[temp2], %[offset](%[g]) \n\t" |
| 44 | "lbux %[temp3], %[offset](%[b]) \n\t" |
| 45 | "ins %[temp1], %[temp0], 16, 16 \n\t" |
| 46 | "ins %[temp3], %[temp2], 16, 16 \n\t" |
| 47 | "precr.qb.ph %[temp0], %[temp1], %[temp3] \n\t" |
| 48 | "sw %[temp0], 0(%[out]) \n\t" |
| 49 | "1: \n\t" |
| 50 | : [temp0]"=&r" (temp0), [temp1]"=&r" (temp1), [temp2]"=&r" (temp2), |
| 51 | [temp3]"=&r" (temp3), [offset]"=&r" (offset), [out]"+&r" (out) |
| 52 | : [a]"r" (a), [r]"r" (r), [g]"r" (g), [b]"r" (b), [step]"r" (step), |
| 53 | [loop_end]"r" (loop_end), [rest]"r" (rest) |
| 54 | : "memory" |
| 55 | ); |
| 56 | } |
| 57 | |
| 58 | static void PackRGB(const uint8_t* r, const uint8_t* g, const uint8_t* b, |
| 59 | int len, int step, uint32_t* out) { |
| 60 | int temp0, temp1, temp2, offset; |
| 61 | const int rest = len & 1; |
| 62 | const int a = 0xff; |
| 63 | const uint32_t* const loop_end = out + len - rest; |
| 64 | __asm__ volatile ( |
| 65 | "xor %[offset], %[offset], %[offset] \n\t" |
| 66 | "beq %[loop_end], %[out], 0f \n\t" |
| 67 | "2: \n\t" |
| 68 | "lbux %[temp0], %[offset](%[r]) \n\t" |
| 69 | "lbux %[temp1], %[offset](%[g]) \n\t" |
| 70 | "lbux %[temp2], %[offset](%[b]) \n\t" |
| 71 | "ins %[temp0], %[a], 16, 16 \n\t" |
| 72 | "ins %[temp2], %[temp1], 16, 16 \n\t" |
| 73 | "addiu %[out], %[out], 4 \n\t" |
| 74 | "precr.qb.ph %[temp0], %[temp0], %[temp2] \n\t" |
| 75 | "sw %[temp0], -4(%[out]) \n\t" |
| 76 | "addu %[offset], %[offset], %[step] \n\t" |
| 77 | "bne %[loop_end], %[out], 2b \n\t" |
| 78 | "0: \n\t" |
| 79 | "beq %[rest], $zero, 1f \n\t" |
| 80 | "lbux %[temp0], %[offset](%[r]) \n\t" |
| 81 | "lbux %[temp1], %[offset](%[g]) \n\t" |
| 82 | "lbux %[temp2], %[offset](%[b]) \n\t" |
| 83 | "ins %[temp0], %[a], 16, 16 \n\t" |
| 84 | "ins %[temp2], %[temp1], 16, 16 \n\t" |
| 85 | "precr.qb.ph %[temp0], %[temp0], %[temp2] \n\t" |
| 86 | "sw %[temp0], 0(%[out]) \n\t" |
| 87 | "1: \n\t" |
| 88 | : [temp0]"=&r" (temp0), [temp1]"=&r" (temp1), [temp2]"=&r" (temp2), |
| 89 | [offset]"=&r" (offset), [out]"+&r" (out) |
| 90 | : [a]"r" (a), [r]"r" (r), [g]"r" (g), [b]"r" (b), [step]"r" (step), |
| 91 | [loop_end]"r" (loop_end), [rest]"r" (rest) |
| 92 | : "memory" |
| 93 | ); |
| 94 | } |
| 95 | |
| 96 | //------------------------------------------------------------------------------ |
| 97 | // Entry point |
| 98 | |
| 99 | extern void VP8EncDspARGBInitMIPSdspR2(void); |
| 100 | |
| 101 | WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInitMIPSdspR2(void) { |
| 102 | VP8PackARGB = PackARGB; |
| 103 | VP8PackRGB = PackRGB; |
| 104 | } |
| 105 | |
| 106 | #else // !WEBP_USE_MIPS_DSP_R2 |
| 107 | |
| 108 | WEBP_DSP_INIT_STUB(VP8EncDspARGBInitMIPSdspR2) |
| 109 | |
| 110 | #endif // WEBP_USE_MIPS_DSP_R2 |
| 111 | |