michael@0: // Copyright (c) 2010 The Chromium Authors. All rights reserved. michael@0: // Use of this source code is governed by a BSD-style license that can be michael@0: // found in the LICENSE file. michael@0: michael@0: // contributor Siarhei Siamashka michael@0: michael@0: #include "yuv_convert.h" michael@0: #include "ycbcr_to_rgb565.h" michael@0: michael@0: michael@0: michael@0: #ifdef HAVE_YCBCR_TO_RGB565 michael@0: michael@0: namespace mozilla { michael@0: michael@0: namespace gfx { michael@0: michael@0: # if defined(MOZILLA_MAY_SUPPORT_NEON) michael@0: void __attribute((noinline,optimize("-fomit-frame-pointer"))) michael@0: yuv42x_to_rgb565_row_neon(uint16 *dst, michael@0: const uint8 *y, michael@0: const uint8 *u, michael@0: const uint8 *v, michael@0: int n, michael@0: int oddflag) michael@0: { michael@0: static __attribute__((aligned(16))) uint16 acc_r[8] = { michael@0: 22840, 22840, 22840, 22840, 22840, 22840, 22840, 22840, michael@0: }; michael@0: static __attribute__((aligned(16))) uint16 acc_g[8] = { michael@0: 17312, 17312, 17312, 17312, 17312, 17312, 17312, 17312, michael@0: }; michael@0: static __attribute__((aligned(16))) uint16 acc_b[8] = { michael@0: 28832, 28832, 28832, 28832, 28832, 28832, 28832, 28832, michael@0: }; michael@0: /* michael@0: * Registers: michael@0: * q0, q1 : d0, d1, d2, d3 - are used for initial loading of YUV data michael@0: * q2 : d4, d5 - are used for storing converted RGB data michael@0: * q3 : d6, d7 - are used for temporary storage michael@0: * michael@0: * q4-q7 - reserved michael@0: * michael@0: * q8, q9 : d16, d17, d18, d19 - are used for expanded Y data michael@0: * q10 : d20, d21 michael@0: * q11 : d22, d23 michael@0: * q12 : d24, d25 michael@0: * q13 : d26, d27 michael@0: * q13, q14, q15 - various constants (#16, #149, #204, #50, #104, #154) michael@0: */ michael@0: asm volatile ( michael@0: ".fpu neon\n" michael@0: /* Allow to build on targets not supporting neon, and force the object file michael@0: * target to avoid bumping the final binary target */ michael@0: ".arch armv7-a\n" michael@0: ".object_arch armv4t\n" michael@0: ".macro convert_macroblock size\n" michael@0: /* load up to 16 source pixels */ michael@0: ".if \\size == 16\n" michael@0: "pld [%[y], #64]\n" michael@0: "pld [%[u], #64]\n" michael@0: "pld [%[v], #64]\n" michael@0: "vld1.8 {d1}, [%[y]]!\n" michael@0: "vld1.8 {d3}, [%[y]]!\n" michael@0: "vld1.8 {d0}, [%[u]]!\n" michael@0: "vld1.8 {d2}, [%[v]]!\n" michael@0: ".elseif \\size == 8\n" michael@0: "vld1.8 {d1}, [%[y]]!\n" michael@0: "vld1.8 {d0[0]}, [%[u]]!\n" michael@0: "vld1.8 {d0[1]}, [%[u]]!\n" michael@0: "vld1.8 {d0[2]}, [%[u]]!\n" michael@0: "vld1.8 {d0[3]}, [%[u]]!\n" michael@0: "vld1.8 {d2[0]}, [%[v]]!\n" michael@0: "vld1.8 {d2[1]}, [%[v]]!\n" michael@0: "vld1.8 {d2[2]}, [%[v]]!\n" michael@0: "vld1.8 {d2[3]}, [%[v]]!\n" michael@0: ".elseif \\size == 4\n" michael@0: "vld1.8 {d1[0]}, [%[y]]!\n" michael@0: "vld1.8 {d1[1]}, [%[y]]!\n" michael@0: "vld1.8 {d1[2]}, [%[y]]!\n" michael@0: "vld1.8 {d1[3]}, [%[y]]!\n" michael@0: "vld1.8 {d0[0]}, [%[u]]!\n" michael@0: "vld1.8 {d0[1]}, [%[u]]!\n" michael@0: "vld1.8 {d2[0]}, [%[v]]!\n" michael@0: "vld1.8 {d2[1]}, [%[v]]!\n" michael@0: ".elseif \\size == 2\n" michael@0: "vld1.8 {d1[0]}, [%[y]]!\n" michael@0: "vld1.8 {d1[1]}, [%[y]]!\n" michael@0: "vld1.8 {d0[0]}, [%[u]]!\n" michael@0: "vld1.8 {d2[0]}, [%[v]]!\n" michael@0: ".elseif \\size == 1\n" michael@0: "vld1.8 {d1[0]}, [%[y]]!\n" michael@0: "vld1.8 {d0[0]}, [%[u]]!\n" michael@0: "vld1.8 {d2[0]}, [%[v]]!\n" michael@0: ".else\n" michael@0: ".error \"unsupported macroblock size\"\n" michael@0: ".endif\n" michael@0: michael@0: /* d1 - Y data (first 8 bytes) */ michael@0: /* d3 - Y data (next 8 bytes) */ michael@0: /* d0 - U data, d2 - V data */ michael@0: michael@0: /* split even and odd Y color components */ michael@0: "vuzp.8 d1, d3\n" /* d1 - evenY, d3 - oddY */ michael@0: /* clip upper and lower boundaries */ michael@0: "vqadd.u8 q0, q0, q4\n" michael@0: "vqadd.u8 q1, q1, q4\n" michael@0: "vqsub.u8 q0, q0, q5\n" michael@0: "vqsub.u8 q1, q1, q5\n" michael@0: michael@0: "vshr.u8 d4, d2, #1\n" /* d4 = V >> 1 */ michael@0: michael@0: "vmull.u8 q8, d1, d27\n" /* q8 = evenY * 149 */ michael@0: "vmull.u8 q9, d3, d27\n" /* q9 = oddY * 149 */ michael@0: michael@0: "vld1.16 {d20, d21}, [%[acc_r], :128]\n" /* q10 - initialize accumulator for red */ michael@0: "vsubw.u8 q10, q10, d4\n" /* red acc -= (V >> 1) */ michael@0: "vmlsl.u8 q10, d2, d28\n" /* red acc -= V * 204 */ michael@0: "vld1.16 {d22, d23}, [%[acc_g], :128]\n" /* q11 - initialize accumulator for green */ michael@0: "vmlsl.u8 q11, d2, d30\n" /* green acc -= V * 104 */ michael@0: "vmlsl.u8 q11, d0, d29\n" /* green acc -= U * 50 */ michael@0: "vld1.16 {d24, d25}, [%[acc_b], :128]\n" /* q12 - initialize accumulator for blue */ michael@0: "vmlsl.u8 q12, d0, d30\n" /* blue acc -= U * 104 */ michael@0: "vmlsl.u8 q12, d0, d31\n" /* blue acc -= U * 154 */ michael@0: michael@0: "vhsub.s16 q3, q8, q10\n" /* calculate even red components */ michael@0: "vhsub.s16 q10, q9, q10\n" /* calculate odd red components */ michael@0: "vqshrun.s16 d0, q3, #6\n" /* right shift, narrow and saturate even red components */ michael@0: "vqshrun.s16 d3, q10, #6\n" /* right shift, narrow and saturate odd red components */ michael@0: michael@0: "vhadd.s16 q3, q8, q11\n" /* calculate even green components */ michael@0: "vhadd.s16 q11, q9, q11\n" /* calculate odd green components */ michael@0: "vqshrun.s16 d1, q3, #6\n" /* right shift, narrow and saturate even green components */ michael@0: "vqshrun.s16 d4, q11, #6\n" /* right shift, narrow and saturate odd green components */ michael@0: michael@0: "vhsub.s16 q3, q8, q12\n" /* calculate even blue components */ michael@0: "vhsub.s16 q12, q9, q12\n" /* calculate odd blue components */ michael@0: "vqshrun.s16 d2, q3, #6\n" /* right shift, narrow and saturate even blue components */ michael@0: "vqshrun.s16 d5, q12, #6\n" /* right shift, narrow and saturate odd blue components */ michael@0: michael@0: "vzip.8 d0, d3\n" /* join even and odd red components */ michael@0: "vzip.8 d1, d4\n" /* join even and odd green components */ michael@0: "vzip.8 d2, d5\n" /* join even and odd blue components */ michael@0: michael@0: "vshll.u8 q3, d0, #8\n\t" michael@0: "vshll.u8 q8, d1, #8\n\t" michael@0: "vshll.u8 q9, d2, #8\n\t" michael@0: "vsri.u16 q3, q8, #5\t\n" michael@0: "vsri.u16 q3, q9, #11\t\n" michael@0: /* store pixel data to memory */ michael@0: ".if \\size == 16\n" michael@0: " vst1.16 {d6, d7}, [%[dst]]!\n" michael@0: " vshll.u8 q3, d3, #8\n\t" michael@0: " vshll.u8 q8, d4, #8\n\t" michael@0: " vshll.u8 q9, d5, #8\n\t" michael@0: " vsri.u16 q3, q8, #5\t\n" michael@0: " vsri.u16 q3, q9, #11\t\n" michael@0: " vst1.16 {d6, d7}, [%[dst]]!\n" michael@0: ".elseif \\size == 8\n" michael@0: " vst1.16 {d6, d7}, [%[dst]]!\n" michael@0: ".elseif \\size == 4\n" michael@0: " vst1.16 {d6}, [%[dst]]!\n" michael@0: ".elseif \\size == 2\n" michael@0: " vst1.16 {d6[0]}, [%[dst]]!\n" michael@0: " vst1.16 {d6[1]}, [%[dst]]!\n" michael@0: ".elseif \\size == 1\n" michael@0: " vst1.16 {d6[0]}, [%[dst]]!\n" michael@0: ".endif\n" michael@0: ".endm\n" michael@0: michael@0: "vmov.u8 d8, #15\n" /* add this to U/V to saturate upper boundary */ michael@0: "vmov.u8 d9, #20\n" /* add this to Y to saturate upper boundary */ michael@0: "vmov.u8 d10, #31\n" /* sub this from U/V to saturate lower boundary */ michael@0: "vmov.u8 d11, #36\n" /* sub this from Y to saturate lower boundary */ michael@0: michael@0: "vmov.u8 d26, #16\n" michael@0: "vmov.u8 d27, #149\n" michael@0: "vmov.u8 d28, #204\n" michael@0: "vmov.u8 d29, #50\n" michael@0: "vmov.u8 d30, #104\n" michael@0: "vmov.u8 d31, #154\n" michael@0: michael@0: "cmp %[oddflag], #0\n" michael@0: "beq 1f\n" michael@0: "convert_macroblock 1\n" michael@0: "sub %[n], %[n], #1\n" michael@0: "1:\n" michael@0: "subs %[n], %[n], #16\n" michael@0: "blt 2f\n" michael@0: "1:\n" michael@0: "convert_macroblock 16\n" michael@0: "subs %[n], %[n], #16\n" michael@0: "bge 1b\n" michael@0: "2:\n" michael@0: "tst %[n], #8\n" michael@0: "beq 3f\n" michael@0: "convert_macroblock 8\n" michael@0: "3:\n" michael@0: "tst %[n], #4\n" michael@0: "beq 4f\n" michael@0: "convert_macroblock 4\n" michael@0: "4:\n" michael@0: "tst %[n], #2\n" michael@0: "beq 5f\n" michael@0: "convert_macroblock 2\n" michael@0: "5:\n" michael@0: "tst %[n], #1\n" michael@0: "beq 6f\n" michael@0: "convert_macroblock 1\n" michael@0: "6:\n" michael@0: ".purgem convert_macroblock\n" michael@0: : [y] "+&r" (y), [u] "+&r" (u), [v] "+&r" (v), [dst] "+&r" (dst), [n] "+&r" (n) michael@0: : [acc_r] "r" (&acc_r[0]), [acc_g] "r" (&acc_g[0]), [acc_b] "r" (&acc_b[0]), michael@0: [oddflag] "r" (oddflag) michael@0: : "cc", "memory", michael@0: "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", michael@0: "d8", "d9", "d10", "d11", /* "d12", "d13", "d14", "d15", */ michael@0: "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", michael@0: "d24", "d25", "d26", "d27", "d28", "d29", "d30", "d31" michael@0: ); michael@0: } michael@0: # endif // MOZILLA_MAY_SUPPORT_NEON michael@0: michael@0: } // namespace gfx michael@0: michael@0: } // namespace mozilla michael@0: michael@0: #endif // HAVE_YCBCR_TO_RGB565