| @ |
| @ Copyright (c) 2012 The WebRTC project authors. All Rights Reserved. |
| @ |
| @ Use of this source code is governed by a BSD-style license |
| @ that can be found in the LICENSE file in the root of the source |
| @ tree. An additional intellectual property rights grant can be found |
| @ in the file PATENTS. All contributing project authors may |
| @ be found in the AUTHORS file in the root of the source tree. |
| @ |
| @ Reference code in transform.c. Bit not exact due to how rounding is |
| @ done in C code and ARM instructions, but quality by assembly code is |
| @ not worse. |
| |
| #include "webrtc/modules/audio_coding/codecs/isac/fix/source/settings.h" |
| #include "webrtc/system_wrappers/interface/asm_defines.h" |
| |
| GLOBAL_FUNCTION WebRtcIsacfix_Spec2TimeNeon |
| GLOBAL_FUNCTION WebRtcIsacfix_Time2SpecNeon |
| GLOBAL_LABEL WebRtcIsacfix_kSinTab1 |
| GLOBAL_LABEL WebRtcIsacfix_kCosTab1 |
| GLOBAL_LABEL WebRtcIsacfix_kSinTab2 |
| |
| @ void WebRtcIsacfix_Time2SpecNeon(int16_t* inre1Q9, |
| @ int16_t* inre2Q9, |
| @ int16_t* outreQ7, |
| @ int16_t* outimQ7); |
| |
| DEFINE_FUNCTION WebRtcIsacfix_Time2SpecNeon |
| .align 2 |
| push {r4-r11,lr} |
| sub sp, sp, #(16 + FRAMESAMPLES * 4) |
| |
| str r0, [sp] @ inre1Q9 |
| str r1, [sp, #4] @ inre2Q9 |
| str r2, [sp, #8] @ outreQ7 |
| str r3, [sp, #12] @ outimQ7 |
| |
| mov r8, #(FRAMESAMPLES - 16) |
| add r12, r0, r8 @ &inreQ7[FRAMESAMPLES / 2 - 4] |
| add r11, r1, r8 @ &inimQ7[FRAMESAMPLES / 2 - 4] |
| add r4, sp, #16 @ tmpreQ16; |
| add r5, sp, #(16 + FRAMESAMPLES * 2) @ tmpimQ16; |
| |
| adr r9, WebRtcIsacfix_kCosTab1 |
| #if defined(__APPLE__) |
| mov r6, #:lower16:(WebRtcIsacfix_kSinTab1 - WebRtcIsacfix_kCosTab1) |
| #else |
| mov r6, #(WebRtcIsacfix_kSinTab1 - WebRtcIsacfix_kCosTab1) |
| #endif |
| add r10, r9, r6 @ WebRtcIsacfix_kSinTab1 |
| |
| vmov.u32 q14, #0 @ Initialize the maximum values for tmpInIm. |
| vmov.u32 q15, #0 @ Initialize the maximum values for tmpInRe. |
| movw r6, #16921 @ 0.5 / sqrt(240) in Q19 |
| lsl r6, #5 @ Together with vqdmulh, net effect is ">> 26". |
| mov r8, #(FRAMESAMPLES / 2) @ loop counter |
| vdup.s32 q11, r6 |
| |
| Time2Spec_TransformAndFindMax: |
| @ Use ">> 26", instead of ">> 7", ">> 16" and then ">> 3" as in the C code. |
| |
| subs r8, #8 |
| |
| vld1.16 {q0}, [r9, :64]! @ WebRtcIsacfix_kCosTab1[] |
| vld1.16 {q2}, [r0]! @ inre1Q9[] |
| vmull.s16 q8, d0, d4 @ WebRtcIsacfix_kCosTab1[k] * inre1Q9[k] |
| vld1.16 {q1}, [r10, :64]! @ WebRtcIsacfix_kSinTab1[] |
| vmull.s16 q9, d1, d5 @ WebRtcIsacfix_kCosTab1[k] * inre1Q9[k] |
| vld1.16 {q3}, [r1]! @ inre2Q9[] |
| vmlal.s16 q8, d2, d6 @ WebRtcIsacfix_kSinTab1[k] * inre2Q9[k] |
| vmlal.s16 q9, d3, d7 @ WebRtcIsacfix_kSinTab1[k] * inre2Q9[k] |
| vmull.s16 q12, d0, d6 @ WebRtcIsacfix_kCosTab1[k] * inre2Q9[k] |
| vmull.s16 q13, d1, d7 @ WebRtcIsacfix_kCosTab1[k] * inre2Q9[k] |
| vmlsl.s16 q12, d2, d4 @ WebRtcIsacfix_kSinTab1[k] * inre1Q9[k] |
| vmlsl.s16 q13, d3, d5 @ WebRtcIsacfix_kSinTab1[k] * inre1Q9[k] |
| |
| vqdmulh.s32 q0, q8, q11 @ xrQ16 * factQ19 |
| vqdmulh.s32 q1, q9, q11 @ xrQ16 * factQ19 |
| vqdmulh.s32 q2, q12, q11 @ xrQ16 * factQ19 |
| vqdmulh.s32 q3, q13, q11 @ xrQ16 * factQ19 |
| |
| @ Find the absolute maximum in the vectors and store them. |
| vabs.s32 q8, q0 |
| vabs.s32 q9, q1 |
| vabs.s32 q12, q2 |
| vst1.32 {q0, q1}, [r4]! @ tmpreQ16[k] |
| vabs.s32 q13, q3 |
| vmax.u32 q14, q8 @ Use u32 so we don't lose the value 0x80000000. |
| vmax.u32 q15, q12 |
| vst1.32 {q2, q3}, [r5]! @ tmpimQ16[k] |
| vmax.u32 q15, q13 |
| vmax.u32 q14, q9 @ Maximum for outre1Q16[]. |
| |
| bgt Time2Spec_TransformAndFindMax |
| |
| @ Find the maximum value in the Neon registers |
| vmax.u32 d28, d29 |
| vmax.u32 d30, d31 |
| vpmax.u32 d28, d28, d28 @ Both 32 bits words hold the same value tmpInIm. |
| vpmax.u32 d30, d30, d30 @ Both 32 bits words hold the same value tmpInRe. |
| vmax.s32 d30, d28, d30 @ if (yrQ16 > xrQ16) {xrQ16 = yrQ16}; |
| |
| ldr r4, [sp] @ inre1Q9 |
| vcls.s32 d31, d30 @ sh = WebRtcSpl_NormW32(tmpInRe); |
| ldr r5, [sp, #4] @ inre2Q9 |
| vmov.i32 d30, #24 |
| add r6, sp, #16 @ tmpreQ16; |
| vsub.s32 d31, d31, d30 @ sh = sh - 24; |
| add r7, sp, #(16 + FRAMESAMPLES * 2) @ tmpimQ16; |
| vdup.s32 q8, d31[0] @ sh |
| |
| mov r8, #(FRAMESAMPLES / 2) @ loop counter |
| |
| Time2Spec_PreFftShift: |
| subs r8, #16 |
| |
| vld1.32 {q0, q1}, [r6]! @ tmpreQ16[] |
| vrshl.s32 q0, q0, q8 |
| vld1.32 {q2, q3}, [r6]! @ tmpreQ16[] |
| vrshl.s32 q1, q1, q8 |
| vld1.32 {q10, q11}, [r7]! @ tmpimQ16[] |
| vrshl.s32 q2, q2, q8 |
| vld1.32 {q12, q13}, [r7]! @ tmpimQ16[] |
| vrshl.s32 q3, q3, q8 |
| vrshl.s32 q10, q10, q8 |
| vrshl.s32 q11, q11, q8 |
| vrshl.s32 q12, q12, q8 |
| vrshl.s32 q13, q13, q8 |
| |
| vmovn.s32 d0, q0 |
| vmovn.s32 d1, q1 |
| vmovn.s32 d2, q2 |
| vmovn.s32 d3, q3 |
| vmovn.s32 d4, q10 |
| vmovn.s32 d5, q11 |
| vmovn.s32 d6, q12 |
| vmovn.s32 d7, q13 |
| |
| vst1.16 {q0, q1}, [r4]! @ inre1Q9[] |
| vst1.16 {q2, q3}, [r5]! @ inre2Q9[] |
| |
| bgt Time2Spec_PreFftShift |
| |
| vmov.s32 r10, d16[0] @ Store value of sh. |
| ldr r0, [sp] @ inre1Q9 |
| ldr r1, [sp, #4] @ inre2Q9 |
| mov r2, #-1 |
| CALL_FUNCTION WebRtcIsacfix_FftRadix16Fastest |
| |
| vdup.s32 q8, r10 @ sh |
| mov r8, #(FRAMESAMPLES - 8) |
| ldr r2, [sp, #8] @ outreQ7 |
| ldr r3, [sp, #12] @ outimQ7 |
| add r11, r2, r8 @ &outRe1Q16[FRAMESAMPLES / 2 - 4] |
| add r12, r3, r8 @ &outim2Q16[FRAMESAMPLES / 2 - 4] |
| ldr r6, [sp] @ inre1Q9 |
| ldr r7, [sp, #4] @ inre2Q9 |
| add r4, r6, r8 @ &inre1Q9[FRAMESAMPLES / 2 - 4] |
| add r5, r7, r8 @ &inre2Q9[FRAMESAMPLES / 2 - 4] |
| adr r10, WebRtcIsacfix_kSinTab2 |
| |
| add r9, r10, #(120*2 - 8) @ &WebRtcIsacfix_kSinTab2[119 - 4] |
| |
| vneg.s32 q15, q8 @ -sh |
| vmov.i32 q0, #23 |
| vsub.s32 q15, q15, q0 @ -sh - 23 |
| |
| mov r8, #(FRAMESAMPLES / 4) @ loop counter |
| |
| @ Pre-load variables. |
| vld1.16 {d2}, [r4] @ inre1Q9[FRAMESAMPLES / 2 - 4 - i] |
| vld1.16 {d3}, [r5] @ inre2Q9[FRAMESAMPLES / 2 - 4 - i] |
| vld1.16 {d0}, [r6]! @ inre1Q9 |
| vld1.16 {d1}, [r7]! @ inre2Q9 |
| |
| Time2Spec_PostFftTransform: |
| @ By vshl, we effectively did "<< (-sh - 23)", instead of "<< (-sh)", |
| @ ">> 14" and then ">> 9" as in the C code. |
| |
| vld1.16 {d6}, [r9, :64] @ kCosTab2[] |
| vneg.s16 d6, d6 |
| vld1.16 {d7}, [r10, :64]! @ WebRtcIsacfix_kSinTab2[] |
| vrev64.16 q1, q1 @ Reverse samples in 2nd half of xrQ16[]. |
| vqadd.s16 d4, d0, d2 @ xrQ16 |
| vqsub.s16 d5, d1, d3 @ xiQ16 |
| vrev64.16 d6, d6 |
| |
| sub r9, #8 @ Update pointers for kCosTab2[]. |
| sub r4, #8 @ Update pointers for inre1Q9[]. |
| sub r5, #8 @ Update pointers for inr22Q9[]. |
| subs r8, #4 @ Update loop counter. |
| |
| vqadd.s16 d1, d1, d3 @ yrQ16 |
| vqsub.s16 d0, d2, d0 @ yiQ16 |
| |
| vmull.s16 q12, d6, d4 @ kCosTab2[k] * xrQ16 |
| vmlsl.s16 q12, d7, d5 @ WebRtcIsacfix_kSinTab2[k] * xiQ16 |
| vmull.s16 q13, d7, d4 @ WebRtcIsacfix_kSinTab2[k] * xrQ16 |
| vmlal.s16 q13, d6, d5 @ kCosTab2[k] * xiQ16 |
| vmull.s16 q9, d7, d1 @ WebRtcIsacfix_kSinTab2[k] * yrQ16 |
| vmlal.s16 q9, d6, d0 @ kCosTab2[k] * yiQ16 |
| vmull.s16 q10, d7, d0 @ WebRtcIsacfix_kSinTab2[k] * yiQ16 |
| vmlsl.s16 q10, d6, d1 @ kCosTab2[k] * yrQ16 |
| |
| vshl.s32 q12, q12, q15 |
| vshl.s32 q13, q13, q15 |
| vshl.s32 q9, q9, q15 |
| vshl.s32 q10, q10, q15 |
| |
| vneg.s32 q8, q9 |
| vld1.16 {d0}, [r6]! @ inre1Q9 |
| vmovn.s32 d24, q12 |
| vld1.16 {d1}, [r7]! @ inre2Q9 |
| vmovn.s32 d25, q13 |
| vld1.16 {d2}, [r4] @ inre1Q9[FRAMESAMPLES / 2 - 4 - i] |
| vmovn.s32 d5, q10 |
| vld1.16 {d3}, [r5] @ inre2Q9[FRAMESAMPLES / 2 - 4 - i] |
| vmovn.s32 d4, q8 |
| vst1.16 {d24}, [r2]! @ outreQ7[k] |
| vrev64.16 q2, q2 @ Reverse the order of the samples. |
| vst1.16 {d25}, [r3]! @ outimQ7[k] |
| vst1.16 {d4}, [r11] @ outreQ7[FRAMESAMPLES / 2 - 1 - k] |
| vst1.16 {d5}, [r12] @ outimQ7[FRAMESAMPLES / 2 - 1 - k] |
| sub r11, #8 @ Update pointers for outreQ7[]. |
| sub r12, #8 @ Update pointers for outimQ7[]. |
| |
| bgt Time2Spec_PostFftTransform |
| |
| add sp, sp, #(16 + FRAMESAMPLES * 4) |
| pop {r4-r11,pc} |
| |
| .align 8 |
| @ Cosine table 1 in Q14 |
| WebRtcIsacfix_kCosTab1: |
| _WebRtcIsacfix_kCosTab1: @ Label for iOS |
| .short 16384, 16383, 16378, 16371, 16362, 16349, 16333, 16315 |
| .short 16294, 16270, 16244, 16214, 16182, 16147, 16110, 16069 |
| .short 16026, 15980, 15931, 15880, 15826, 15769, 15709, 15647 |
| .short 15582, 15515, 15444, 15371, 15296, 15218, 15137, 15053 |
| .short 14968, 14879, 14788, 14694, 14598, 14500, 14399, 14295 |
| .short 14189, 14081, 13970, 13856, 13741, 13623, 13502, 13380 |
| .short 13255, 13128, 12998, 12867, 12733, 12597, 12458, 12318 |
| .short 12176, 12031, 11885, 11736, 11585, 11433, 11278, 11121 |
| .short 10963, 10803, 10641, 10477, 10311, 10143, 9974, 9803 |
| .short 9630, 9456, 9280, 9102, 8923, 8743, 8561, 8377 |
| .short 8192, 8006, 7818, 7629, 7438, 7246, 7053, 6859 |
| .short 6664, 6467, 6270, 6071, 5872, 5671, 5469, 5266 |
| .short 5063, 4859, 4653, 4447, 4240, 4033, 3825, 3616 |
| .short 3406, 3196, 2986, 2775, 2563, 2351, 2139, 1926 |
| .short 1713, 1499, 1285, 1072, 857, 643, 429, 214 |
| .short 0, -214, -429, -643, -857, -1072, -1285, -1499 |
| .short -1713, -1926, -2139, -2351, -2563, -2775, -2986, -3196 |
| .short -3406, -3616, -3825, -4033, -4240, -4447, -4653, -4859 |
| .short -5063, -5266, -5469, -5671, -5872, -6071, -6270, -6467 |
| .short -6664, -6859, -7053, -7246, -7438, -7629, -7818, -8006 |
| .short -8192, -8377, -8561, -8743, -8923, -9102, -9280, -9456 |
| .short -9630, -9803, -9974, -10143, -10311, -10477, -10641, -10803 |
| .short -10963, -11121, -11278, -11433, -11585, -11736, -11885, -12031 |
| .short -12176, -12318, -12458, -12597, -12733, -12867, -12998, -13128 |
| .short -13255, -13380, -13502, -13623, -13741, -13856, -13970, -14081 |
| .short -14189, -14295, -14399, -14500, -14598, -14694, -14788, -14879 |
| .short -14968, -15053, -15137, -15218, -15296, -15371, -15444, -15515 |
| .short -15582, -15647, -15709, -15769, -15826, -15880, -15931, -15980 |
| .short -16026, -16069, -16110, -16147, -16182, -16214, -16244, -16270 |
| .short -16294, -16315, -16333, -16349, -16362, -16371, -16378, -16383 |
| |
| .align 8 |
| @ Sine table 2 in Q14 |
| WebRtcIsacfix_kSinTab2: |
| _WebRtcIsacfix_kSinTab2: @ Label for iOS |
| .short 16384, -16381, 16375, -16367, 16356, -16342, 16325, -16305 |
| .short 16283, -16257, 16229, -16199, 16165, -16129, 16090, -16048 |
| .short 16003, -15956, 15906, -15853, 15798, -15739, 15679, -15615 |
| .short 15549, -15480, 15408, -15334, 15257, -15178, 15095, -15011 |
| .short 14924, -14834, 14741, -14647, 14549, -14449, 14347, -14242 |
| .short 14135, -14025, 13913, -13799, 13682, -13563, 13441, -13318 |
| .short 13192, -13063, 12933, -12800, 12665, -12528, 12389, -12247 |
| .short 12104, -11958, 11810, -11661, 11509, -11356, 11200, -11042 |
| .short 10883, -10722, 10559, -10394, 10227, -10059, 9889, -9717 |
| .short 9543, -9368, 9191, -9013, 8833, -8652, 8469, -8285 |
| .short 8099, -7912, 7723, -7534, 7342, -7150, 6957, -6762 |
| .short 6566, -6369, 6171, -5971, 5771, -5570, 5368, -5165 |
| .short 4961, -4756, 4550, -4344, 4137, -3929, 3720, -3511 |
| .short 3301, -3091, 2880, -2669, 2457, -2245, 2032, -1819 |
| .short 1606, -1392, 1179, -965, 750, -536, 322, -107 |
| |
| @ Table kCosTab2 was removed since its data is redundant with kSinTab2. |
| |
| .align 8 |
| @ Sine table 1 in Q14 |
| WebRtcIsacfix_kSinTab1: |
| _WebRtcIsacfix_kSinTab1: @ Label for iOS |
| .short 0, 214, 429, 643, 857, 1072, 1285, 1499 |
| .short 1713, 1926, 2139, 2351, 2563, 2775, 2986, 3196 |
| .short 3406, 3616, 3825, 4033, 4240, 4447, 4653, 4859 |
| .short 5063, 5266, 5469, 5671, 5872, 6071, 6270, 6467 |
| .short 6664, 6859, 7053, 7246, 7438, 7629, 7818, 8006 |
| .short 8192, 8377, 8561, 8743, 8923, 9102, 9280, 9456 |
| .short 9630, 9803, 9974, 10143, 10311, 10477, 10641, 10803 |
| .short 10963, 11121, 11278, 11433, 11585, 11736, 11885, 12031 |
| .short 12176, 12318, 12458, 12597, 12733, 12867, 12998, 13128 |
| .short 13255, 13380, 13502, 13623, 13741, 13856, 13970, 14081 |
| .short 14189, 14295, 14399, 14500, 14598, 14694, 14788, 14879 |
| .short 14968, 15053, 15137, 15218, 15296, 15371, 15444, 15515 |
| .short 15582, 15647, 15709, 15769, 15826, 15880, 15931, 15980 |
| .short 16026, 16069, 16110, 16147, 16182, 16214, 16244, 16270 |
| .short 16294, 16315, 16333, 16349, 16362, 16371, 16378, 16383 |
| .short 16384, 16383, 16378, 16371, 16362, 16349, 16333, 16315 |
| .short 16294, 16270, 16244, 16214, 16182, 16147, 16110, 16069 |
| .short 16026, 15980, 15931, 15880, 15826, 15769, 15709, 15647 |
| .short 15582, 15515, 15444, 15371, 15296, 15218, 15137, 15053 |
| .short 14968, 14879, 14788, 14694, 14598, 14500, 14399, 14295 |
| .short 14189, 14081, 13970, 13856, 13741, 13623, 13502, 13380 |
| .short 13255, 13128, 12998, 12867, 12733, 12597, 12458, 12318 |
| .short 12176, 12031, 11885, 11736, 11585, 11433, 11278, 11121 |
| .short 10963, 10803, 10641, 10477, 10311, 10143, 9974, 9803 |
| .short 9630, 9456, 9280, 9102, 8923, 8743, 8561, 8377 |
| .short 8192, 8006, 7818, 7629, 7438, 7246, 7053, 6859 |
| .short 6664, 6467, 6270, 6071, 5872, 5671, 5469, 5266 |
| .short 5063, 4859, 4653, 4447, 4240, 4033, 3825, 3616 |
| .short 3406, 3196, 2986, 2775, 2563, 2351, 2139, 1926 |
| .short 1713, 1499, 1285, 1072, 857, 643, 429, 214 |
| |
| @ void WebRtcIsacfix_Spec2TimeNeon(int16_t *inreQ7, |
| @ int16_t *inimQ7, |
| @ int32_t *outre1Q16, |
| @ int32_t *outre2Q16); |
| |
| DEFINE_FUNCTION WebRtcIsacfix_Spec2TimeNeon |
| .align 2 |
| push {r4-r11,lr} |
| |
| sub sp, sp, #16 |
| str r0, [sp] @ inreQ7 |
| str r1, [sp, #4] @ inimQ7 |
| str r2, [sp, #8] @ outre1Q16 |
| str r3, [sp, #12] @ outre2Q16 |
| |
| mov r8, #(FRAMESAMPLES - 16) |
| add r12, r0, r8 @ &inreQ7[FRAMESAMPLES / 2 - 8] |
| add r11, r1, r8 @ &inimQ7[FRAMESAMPLES / 2 - 8] |
| add r4, r2, r8, lsl #1 @ &outRe1Q16[FRAMESAMPLES / 2 - 8] |
| add r6, r3, r8, lsl #1 @ &outRe2Q16[FRAMESAMPLES / 2 - 8] |
| |
| mov r8, #(FRAMESAMPLES / 2) @ loop counter |
| adr r10, WebRtcIsacfix_kSinTab2 |
| add r9, r10, #(120*2 - 16) @ &WebRtcIsacfix_kSinTab2[119 - 8] |
| |
| vpush {q4-q7} |
| |
| mov r5, #-32 |
| mov r7, #-16 |
| vmov.u32 q6, #0 @ Initialize the maximum values for tmpInIm. |
| vmov.u32 q7, #0 @ Initialize the maximum values for tmpInRe. |
| |
| TransformAndFindMax: |
| @ Use ">> 5", instead of "<< 9" and then ">> 14" as in the C code. |
| @ Bit-exact. |
| |
| subs r8, #16 |
| |
| vld1.16 {q0}, [r9, :64] @ kCosTab2[] |
| sub r9, #16 |
| vld1.16 {q2}, [r0]! @ inreQ7[] |
| vneg.s16 q0, q0 |
| vld1.16 {q3}, [r1]! @ inimQ7[] |
| vrev64.16 d0, d0 |
| vrev64.16 d1, d1 |
| vld1.16 {q1}, [r10, :64]! @ WebRtcIsacfix_kSinTab2[] |
| vswp d0, d1 |
| |
| vmull.s16 q8, d2, d6 @ WebRtcIsacfix_kSinTab2[k] * inimQ7[k] |
| vmull.s16 q9, d3, d7 @ WebRtcIsacfix_kSinTab2[k] * inimQ7[k] |
| vmlal.s16 q8, d0, d4 @ kCosTab2[k] * inreQ7[k] |
| vmlal.s16 q9, d1, d5 @ kCosTab2[k] * inreQ7[k] |
| vmull.s16 q12, d0, d6 @ kCosTab2[k] * inimQ7[k] |
| vmull.s16 q13, d1, d7 @ kCosTab2[k] * inimQ7[k] |
| vmlsl.s16 q12, d2, d4 @ WebRtcIsacfix_kSinTab2[k] * inreQ7[k] |
| vmlsl.s16 q13, d3, d5 @ WebRtcIsacfix_kSinTab2[k] * inreQ7[k] |
| |
| vld1.16 {q2}, [r11], r7 @ inimQ7[FRAMESAMPLES / 2 - 8 + i] |
| vld1.16 {q3}, [r12], r7 @ inreQ7[FRAMESAMPLES / 2 - 8 + i] |
| |
| vrev64.16 q2, q2 @ Reverse the order of the samples |
| vrev64.16 q3, q3 @ Reverse the order of the samples |
| |
| vmull.s16 q14, d2, d5 @ WebRtcIsacfix_kSinTab2[k] * inimQ7[k] |
| vmull.s16 q15, d3, d4 @ WebRtcIsacfix_kSinTab2[k] * inimQ7[k] |
| vmlsl.s16 q14, d0, d7 @ q14 -= kCosTab2[k] * inreQ7[k] |
| vmlsl.s16 q15, d1, d6 @ q15 -= kCosTab2[k] * inreQ7[k] |
| |
| vmull.s16 q10, d0, d5 @ kCosTab2[k] * inimQ7[] |
| vmull.s16 q11, d1, d4 @ kCosTab2[k] * inimQ7[] |
| vmlal.s16 q10, d2, d7 @ q10 += WebRtcIsacfix_kSinTab2[k] * inreQ7[] |
| vmlal.s16 q11, d3, d6 @ q11 += WebRtcIsacfix_kSinTab2[k] * inreQ7[] |
| |
| vshr.s32 q8, q8, #5 @ xrQ16 |
| vshr.s32 q9, q9, #5 @ xrQ16 |
| vshr.s32 q12, q12, #5 @ xiQ16 |
| vshr.s32 q13, q13, #5 @ xiQ16 |
| vshr.s32 q14, q14, #5 @ yiQ16 |
| vshr.s32 q15, q15, #5 @ yiQ16 |
| |
| vneg.s32 q10, q10 |
| vneg.s32 q11, q11 |
| |
| @ xrQ16 - yiQ16 |
| vsub.s32 q0, q8, q14 |
| vsub.s32 q1, q9, q15 |
| |
| vshr.s32 q10, q10, #5 @ yrQ16 |
| vshr.s32 q11, q11, #5 @ yrQ16 |
| |
| @ xrQ16 + yiQ16 |
| vadd.s32 q3, q8, q14 |
| vadd.s32 q2, q9, q15 |
| |
| @ yrQ16 + xiQ16 |
| vadd.s32 q4, q10, q12 |
| vadd.s32 q5, q11, q13 |
| |
| @ yrQ16 - xiQ16 |
| vsub.s32 q8, q11, q13 |
| vsub.s32 q9, q10, q12 |
| |
| @ Reverse the order of the samples |
| vrev64.32 q2, q2 |
| vrev64.32 q3, q3 |
| vrev64.32 q8, q8 |
| vrev64.32 q9, q9 |
| vswp d4, d5 |
| vswp d6, d7 |
| |
| vst1.32 {q0, q1}, [r2]! @ outre1Q16[k] |
| vswp d16, d17 |
| vswp d18, d19 |
| vst1.32 {q2, q3}, [r4], r5 @ outre1Q16[FRAMESAMPLES / 2 - 1 - k] |
| |
| @ Find the absolute maximum in the vectors and store them in q6 and q7. |
| vabs.s32 q10, q0 |
| vabs.s32 q14, q4 |
| vabs.s32 q11, q1 |
| vabs.s32 q15, q5 |
| vabs.s32 q12, q2 |
| vmax.u32 q6, q10 @ Use u32 so we don't lose the value 0x80000000. |
| vmax.u32 q7, q14 @ Maximum for outre2Q16[]. |
| vabs.s32 q0, q8 |
| vmax.u32 q6, q11 @ Maximum for outre1Q16[]. |
| vmax.u32 q7, q15 |
| vabs.s32 q13, q3 |
| vmax.u32 q6, q12 |
| vmax.u32 q7, q0 |
| vabs.s32 q1, q9 |
| vst1.32 {q4, q5}, [r3]! @ outre2Q16[k] |
| vst1.32 {q8, q9}, [r6], r5 @ outre2Q16[FRAMESAMPLES / 2 - 1 - k] |
| vmax.u32 q6, q13 |
| vmax.u32 q7, q1 |
| |
| bgt TransformAndFindMax |
| |
| adr r10, WebRtcIsacfix_kSinTab1 |
| #if defined(__APPLE__) |
| mov r2, #:lower16:(WebRtcIsacfix_kSinTab1 - WebRtcIsacfix_kCosTab1) |
| #else |
| mov r2, #(WebRtcIsacfix_kSinTab1 - WebRtcIsacfix_kCosTab1) |
| #endif |
| |
| sub r11, r10, r2 @ WebRtcIsacfix_kCosTab1 |
| |
| @ Find the maximum value in the Neon registers |
| vmax.u32 d12, d13 |
| vmax.u32 d14, d15 |
| vpmax.u32 d12, d12, d12 @ Both 32 bits words hold the same value tmpInIm. |
| vpmax.u32 d14, d14, d14 @ Both 32 bits words hold the same value tmpInRe. |
| vmax.s32 d0, d12, d14 @ if (tmpInIm>tmpInRe) tmpInRe = tmpInIm; |
| |
| vpop {q4-q7} |
| |
| ldr r4, [sp] @ inreQ7 |
| vcls.s32 d1, d0 @ sh = WebRtcSpl_NormW32(tmpInRe); |
| ldr r5, [sp, #4] @ inimQ7 |
| vmov.i32 d0, #24 @ sh = sh-24; |
| ldr r6, [sp, #8] @ outre1Q16 |
| vsub.s32 d1, d1, d0 |
| ldr r7, [sp, #12] @ outre2Q16 |
| vdup.s32 q8, d1[0] @ sh |
| |
| mov r8, #(FRAMESAMPLES / 2) |
| |
| PreFftShift: |
| subs r8, #16 |
| vld1.32 {q0, q1}, [r6]! @ outre1Q16[] |
| vld1.32 {q2, q3}, [r6]! @ outre1Q16[] |
| vrshl.s32 q0, q0, q8 |
| vrshl.s32 q1, q1, q8 |
| vrshl.s32 q2, q2, q8 |
| vrshl.s32 q3, q3, q8 |
| vld1.32 {q10, q11}, [r7]! @ outre2Q16[] |
| vld1.32 {q12, q13}, [r7]! @ outre2Q16[] |
| vrshl.s32 q10, q10, q8 |
| vrshl.s32 q11, q11, q8 |
| vrshl.s32 q12, q12, q8 |
| vrshl.s32 q13, q13, q8 |
| |
| vmovn.s32 d0, q0 |
| vmovn.s32 d1, q1 |
| vmovn.s32 d2, q2 |
| vmovn.s32 d3, q3 |
| vmovn.s32 d4, q10 |
| vmovn.s32 d5, q11 |
| vmovn.s32 d6, q12 |
| vmovn.s32 d7, q13 |
| |
| vst1.16 {q0, q1}, [r4]! @ inreQ7[] |
| vst1.16 {q2, q3}, [r5]! @ inimQ7[] |
| |
| bgt PreFftShift |
| |
| vmov.s32 r8, d16[0] @ Store value of sh. |
| ldr r0, [sp] @ inreQ7 |
| ldr r1, [sp, #4] @ inimQ7 |
| mov r2, #1 |
| CALL_FUNCTION WebRtcIsacfix_FftRadix16Fastest |
| |
| vdup.s32 q8, r8 @ sh |
| mov r9, r11 @ WebRtcIsacfix_kCosTab1 |
| ldr r4, [sp] @ inreQ7 |
| ldr r5, [sp, #4] @ inimQ7 |
| ldr r6, [sp, #8] @ outre1Q16 |
| ldr r7, [sp, #12] @ outre2Q16 |
| mov r8, #(FRAMESAMPLES / 2) |
| vneg.s32 q15, q8 @ -sh |
| movw r0, #273 |
| lsl r0, #15 @ Together with vqdmulh, net effect is ">> 16". |
| vdup.s32 q14, r0 |
| |
| PostFftShiftDivide: |
| subs r8, #16 |
| |
| vld1.16 {q0, q1}, [r4]! @ inreQ7 |
| vmovl.s16 q10, d0 |
| vmovl.s16 q11, d1 |
| vld1.16 {q2, q3}, [r5]! @ inimQ7 |
| vmovl.s16 q8, d2 |
| vmovl.s16 q9, d3 |
| |
| vshl.s32 q10, q10, q15 |
| vshl.s32 q11, q11, q15 |
| vshl.s32 q8, q8, q15 |
| vshl.s32 q9, q9, q15 |
| |
| vqdmulh.s32 q10, q10, q14 |
| vqdmulh.s32 q11, q11, q14 |
| vqdmulh.s32 q8, q8, q14 |
| vqdmulh.s32 q9, q9, q14 |
| |
| vmovl.s16 q0, d4 |
| vmovl.s16 q1, d5 |
| vmovl.s16 q2, d6 |
| vmovl.s16 q3, d7 |
| |
| vshl.s32 q0, q0, q15 |
| vshl.s32 q1, q1, q15 |
| vshl.s32 q2, q2, q15 |
| vshl.s32 q3, q3, q15 |
| |
| @ WEBRTC_SPL_MUL_16_32_RSFT16(273, outre2Q16[k]) |
| vqdmulh.s32 q0, q0, q14 |
| vqdmulh.s32 q1, q1, q14 |
| vst1.32 {q10, q11}, [r6]! @ outre1Q16[] |
| vqdmulh.s32 q2, q2, q14 |
| vqdmulh.s32 q3, q3, q14 |
| vst1.32 {q8, q9}, [r6]! @ outre1Q16[] |
| vst1.32 {q0, q1}, [r7]! @ outre2Q16[] |
| vst1.32 {q2, q3}, [r7]! @ outre2Q16[] |
| |
| bgt PostFftShiftDivide |
| |
| mov r8, #(FRAMESAMPLES / 2) |
| ldr r2, [sp, #8] @ outre1Q16 |
| ldr r3, [sp, #12] @ outre2Q16 |
| movw r0, #31727 |
| lsl r0, #16 @ With vqdmulh and vrshrn, net effect is ">> 25". |
| |
| DemodulateAndSeparate: |
| subs r8, #8 |
| |
| vld1.16 {q0}, [r9, :64]! @ WebRtcIsacfix_kCosTab1[] |
| vmovl.s16 q10, d0 @ WebRtcIsacfix_kCosTab1[] |
| vld1.16 {q1}, [r10, :64]! @ WebRtcIsacfix_kSinTab1[] |
| vmovl.s16 q11, d1 @ WebRtcIsacfix_kCosTab1[] |
| vld1.32 {q2, q3}, [r2] @ outre1Q16 |
| vmovl.s16 q12, d2 @ WebRtcIsacfix_kSinTab1[] |
| vld1.32 {q14, q15}, [r3] @ outre2Q16 |
| vmovl.s16 q13, d3 @ WebRtcIsacfix_kSinTab1[] |
| |
| vmull.s32 q0, d20, d4 @ WebRtcIsacfix_kCosTab1[k] * outre1Q16[k] |
| vmull.s32 q1, d21, d5 @ WebRtcIsacfix_kCosTab1[k] * outre1Q16[k] |
| vmull.s32 q8, d22, d6 @ WebRtcIsacfix_kCosTab1[k] * outre1Q16[k] |
| vmull.s32 q9, d23, d7 @ WebRtcIsacfix_kCosTab1[k] * outre1Q16[k] |
| |
| vmlsl.s32 q0, d24, d28 @ += WebRtcIsacfix_kSinTab1[k] * outre2Q16[k] |
| vmlsl.s32 q1, d25, d29 @ += WebRtcIsacfix_kSinTab1[k] * outre2Q16[k] |
| vmlsl.s32 q8, d26, d30 @ += WebRtcIsacfix_kSinTab1[k] * outre2Q16[k] |
| vmlsl.s32 q9, d27, d31 @ += WebRtcIsacfix_kSinTab1[k] * outre2Q16[k] |
| |
| vrshrn.s64 d0, q0, #10 @ xrQ16 |
| vrshrn.s64 d1, q1, #10 @ xrQ16 |
| vrshrn.s64 d2, q8, #10 @ xrQ16 |
| vrshrn.s64 d3, q9, #10 @ xrQ16 |
| |
| vmull.s32 q8, d20, d28 @ WebRtcIsacfix_kCosTab1[k] * outre2Q16[k] |
| vmull.s32 q9, d21, d29 @ WebRtcIsacfix_kCosTab1[k] * outre2Q16[k] |
| vmull.s32 q14, d22, d30 @ WebRtcIsacfix_kCosTab1[k] * outre2Q16[k] |
| vmull.s32 q15, d23, d31 @ WebRtcIsacfix_kCosTab1[k] * outre2Q16[k] |
| |
| vmlal.s32 q8, d24, d4 @ += WebRtcIsacfix_kSinTab1[k] * outre1Q16[k] |
| vmlal.s32 q9, d25, d5 @ += WebRtcIsacfix_kSinTab1[k] * outre1Q16[k] |
| vmlal.s32 q14, d26, d6 @ += WebRtcIsacfix_kSinTab1[k] * outre1Q16[k] |
| vmlal.s32 q15, d27, d7 @ += WebRtcIsacfix_kSinTab1[k] * outre1Q16[k] |
| |
| vdup.s32 q11, r0 @ generic -> Neon doesn't cost extra cycles. |
| |
| vrshrn.s64 d24, q8, #10 @ xiQ16 |
| vrshrn.s64 d25, q9, #10 @ xiQ16 |
| vqdmulh.s32 q0, q0, q11 |
| vrshrn.s64 d26, q14, #10 @ xiQ16 |
| vrshrn.s64 d27, q15, #10 @ xiQ16 |
| |
| @ WEBRTC_SPL_MUL_16_32_RSFT11(factQ11, xrQ16) |
| @ WEBRTC_SPL_MUL_16_32_RSFT11(factQ11, xiQ16) |
| |
| vqdmulh.s32 q1, q1, q11 |
| vqdmulh.s32 q2, q12, q11 |
| vqdmulh.s32 q3, q13, q11 |
| |
| vst1.16 {q0, q1}, [r2]! @ outre1Q16[] |
| vst1.16 {q2, q3}, [r3]! @ outre2Q16[] |
| |
| bgt DemodulateAndSeparate |
| |
| add sp, sp, #16 |
| pop {r4-r11,pc} |
