OLD | NEW |
(Empty) | |
| 1 @// |
| 2 @// Copyright (c) 2013 The WebRTC project authors. All Rights Reserved. |
| 3 @// |
| 4 @// Use of this source code is governed by a BSD-style license |
| 5 @// that can be found in the LICENSE file in the root of the source |
| 6 @// tree. An additional intellectual property rights grant can be found |
| 7 @// in the file PATENTS. All contributing project authors may |
| 8 @// be found in the AUTHORS file in the root of the source tree. |
| 9 @// |
| 10 @// |
| 11 @// This is a modification of armSP_FFT_CToC_SC32_Radix4_fs_unsafe_s.s |
| 12 @// to support float instead of SC32. |
| 13 @// |
| 14 |
| 15 @// |
| 16 @// Description: |
| 17 @// Compute a first stage Radix 4 FFT stage for a N point complex signal |
| 18 @// |
| 19 @// |
| 20 |
| 21 |
| 22 @// Include standard headers |
| 23 |
| 24 #include "dl/api/armCOMM_s.h" |
| 25 #include "dl/api/omxtypes_s.h" |
| 26 |
| 27 @// Import symbols required from other files |
| 28 @// (For example tables) |
| 29 |
| 30 |
| 31 |
| 32 |
| 33 @// Set debugging level |
| 34 @//DEBUG_ON SETL {TRUE} |
| 35 |
| 36 |
| 37 |
| 38 @// Guarding implementation by the processor name |
| 39 |
| 40 |
| 41 |
| 42 @// Guarding implementation by the processor name |
| 43 |
| 44 |
| 45 @//Input Registers |
| 46 |
| 47 #define pSrc r0 |
| 48 #define pDst r2 |
| 49 #define pTwiddle r1 |
| 50 #define pPingPongBuf r5 |
| 51 #define subFFTNum r6 |
| 52 #define subFFTSize r7 |
| 53 |
| 54 |
| 55 @//Output Registers |
| 56 |
| 57 |
| 58 @//Local Scratch Registers |
| 59 |
| 60 #define grpSize r3 |
| 61 @// Reuse grpSize as setCount |
| 62 #define setCount r3 |
| 63 #define pointStep r4 |
| 64 #define outPointStep r4 |
| 65 #define setStep r8 |
| 66 #define step1 r9 |
| 67 #define step3 r10 |
| 68 |
| 69 @// Neon Registers |
| 70 |
| 71 #define dXr0 D0.F32 |
| 72 #define dXi0 D1.F32 |
| 73 #define dXr1 D2.F32 |
| 74 #define dXi1 D3.F32 |
| 75 #define dXr2 D4.F32 |
| 76 #define dXi2 D5.F32 |
| 77 #define dXr3 D6.F32 |
| 78 #define dXi3 D7.F32 |
| 79 #define dYr0 D8.F32 |
| 80 #define dYi0 D9.F32 |
| 81 #define dYr1 D10.F32 |
| 82 #define dYi1 D11.F32 |
| 83 #define dYr2 D12.F32 |
| 84 #define dYi2 D13.F32 |
| 85 #define dYr3 D14.F32 |
| 86 #define dYi3 D15.F32 |
| 87 #define qX0 Q0.F32 |
| 88 #define qX1 Q1.F32 |
| 89 #define qX2 Q2.F32 |
| 90 #define qX3 Q3.F32 |
| 91 #define qY0 Q4.F32 |
| 92 #define qY1 Q5.F32 |
| 93 #define qY2 Q6.F32 |
| 94 #define qY3 Q7.F32 |
| 95 #define dZr0 D16.F32 |
| 96 #define dZi0 D17.F32 |
| 97 #define dZr1 D18.F32 |
| 98 #define dZi1 D19.F32 |
| 99 #define dZr2 D20.F32 |
| 100 #define dZi2 D21.F32 |
| 101 #define dZr3 D22.F32 |
| 102 #define dZi3 D23.F32 |
| 103 #define qZ0 Q8.F32 |
| 104 #define qZ1 Q9.F32 |
| 105 #define qZ2 Q10.F32 |
| 106 #define qZ3 Q11.F32 |
| 107 |
| 108 |
| 109 .MACRO FFTSTAGE scaled, inverse, name |
| 110 |
| 111 @// Define stack arguments |
| 112 |
| 113 @// pT0+1 increments pT0 by 8 bytes |
| 114 @// pT0+pointStep = increment of 8*pointStep bytes = 2*grpSize bytes |
| 115 @// Note: outPointStep = pointStep for firststage |
| 116 |
| 117 MOV pointStep,subFFTNum,LSL #1 |
| 118 |
| 119 |
| 120 @// Update pSubFFTSize and pSubFFTNum regs |
| 121 VLD2 {dXr0,dXi0},[pSrc :128],pointStep @// data[0] |
| 122 @// subFFTSize = 1 for the first stage |
| 123 MOV subFFTSize,#4 |
| 124 |
| 125 @// Note: setCount = subFFTNum/4 (reuse the grpSize reg for setCount) |
| 126 LSR grpSize,subFFTNum,#2 |
| 127 VLD2 {dXr1,dXi1},[pSrc :128],pointStep @// data[1] |
| 128 MOV subFFTNum,grpSize |
| 129 |
| 130 |
| 131 @// Calculate the step of input data for the next set |
| 132 @//MOV setStep,pointStep,LSL #1 |
| 133 MOV setStep,grpSize,LSL #4 |
| 134 VLD2 {dXr2,dXi2},[pSrc :128],pointStep @// data[2] |
| 135 @// setStep = 3*pointStep |
| 136 ADD setStep,setStep,pointStep |
| 137 @// setStep = - 3*pointStep+16 |
| 138 RSB setStep,setStep,#16 |
| 139 |
| 140 @// data[3] & update pSrc for the next set |
| 141 VLD2 {dXr3,dXi3},[pSrc :128],setStep |
| 142 @// step1 = 2*pointStep |
| 143 MOV step1,pointStep,LSL #1 |
| 144 |
| 145 VADD qY0,qX0,qX2 |
| 146 |
| 147 @// step3 = -pointStep |
| 148 RSB step3,pointStep,#0 |
| 149 |
| 150 @// grp = 0 a special case since all the twiddle factors are 1 |
| 151 @// Loop on the sets : 2 sets at a time |
| 152 |
| 153 radix4fsGrpZeroSetLoop\name : |
| 154 |
| 155 |
| 156 |
| 157 @// Decrement setcount |
| 158 SUBS setCount,setCount,#2 |
| 159 |
| 160 |
| 161 @// finish first stage of 4 point FFT |
| 162 |
| 163 |
| 164 VSUB qY2,qX0,qX2 |
| 165 |
| 166 VLD2 {dXr0,dXi0},[pSrc :128],step1 @// data[0] |
| 167 VADD qY1,qX1,qX3 |
| 168 VLD2 {dXr2,dXi2},[pSrc :128],step3 @// data[2] |
| 169 VSUB qY3,qX1,qX3 |
| 170 |
| 171 |
| 172 @// finish second stage of 4 point FFT |
| 173 |
| 174 .ifeqs "\inverse", "TRUE" |
| 175 |
| 176 VLD2 {dXr1,dXi1},[pSrc :128],step1 @// data[1] |
| 177 VADD qZ0,qY0,qY1 |
| 178 |
| 179 @// data[3] & update pSrc for the next set, but not if it's the |
| 180 @// last iteration so that we don't read past the end of the |
| 181 @// input array. |
| 182 BEQ radix4SkipLastUpdateInv\name |
| 183 VLD2 {dXr3,dXi3},[pSrc :128],setStep |
| 184 radix4SkipLastUpdateInv\name: |
| 185 VSUB dZr3,dYr2,dYi3 |
| 186 |
| 187 VST2 {dZr0,dZi0},[pDst :128],outPointStep |
| 188 VADD dZi3,dYi2,dYr3 |
| 189 |
| 190 VSUB qZ1,qY0,qY1 |
| 191 VST2 {dZr3,dZi3},[pDst :128],outPointStep |
| 192 |
| 193 VADD dZr2,dYr2,dYi3 |
| 194 VST2 {dZr1,dZi1},[pDst :128],outPointStep |
| 195 VSUB dZi2,dYi2,dYr3 |
| 196 |
| 197 VADD qY0,qX0,qX2 @// u0 for next iteration |
| 198 VST2 {dZr2,dZi2},[pDst :128],setStep |
| 199 |
| 200 |
| 201 .else |
| 202 |
| 203 VLD2 {dXr1,dXi1},[pSrc :128],step1 @// data[1] |
| 204 VADD qZ0,qY0,qY1 |
| 205 |
| 206 @// data[3] & update pSrc for the next set, but not if it's the |
| 207 @// last iteration so that we don't read past the end of the |
| 208 @// input array. |
| 209 BEQ radix4SkipLastUpdateFwd\name |
| 210 VLD2 {dXr3,dXi3},[pSrc :128],setStep |
| 211 radix4SkipLastUpdateFwd\name: |
| 212 VADD dZr2,dYr2,dYi3 |
| 213 |
| 214 VST2 {dZr0,dZi0},[pDst :128],outPointStep |
| 215 VSUB dZi2,dYi2,dYr3 |
| 216 |
| 217 VSUB qZ1,qY0,qY1 |
| 218 VST2 {dZr2,dZi2},[pDst :128],outPointStep |
| 219 |
| 220 VSUB dZr3,dYr2,dYi3 |
| 221 VST2 {dZr1,dZi1},[pDst :128],outPointStep |
| 222 VADD dZi3,dYi2,dYr3 |
| 223 |
| 224 VADD qY0,qX0,qX2 @// u0 for next iteration |
| 225 VST2 {dZr3,dZi3},[pDst :128],setStep |
| 226 |
| 227 .endif |
| 228 |
| 229 BGT radix4fsGrpZeroSetLoop\name |
| 230 |
| 231 @// reset pSrc to pDst for the next stage |
| 232 SUB pSrc,pDst,pointStep @// pDst -= 2*grpSize |
| 233 MOV pDst,pPingPongBuf |
| 234 |
| 235 |
| 236 .endm |
| 237 |
| 238 |
| 239 |
| 240 M_START armSP_FFTFwd_CToC_FC32_Radix4_fs_OutOfPlace_unsafe,r4 |
| 241 FFTSTAGE "FALSE","FALSE",fwd |
| 242 M_END |
| 243 |
| 244 |
| 245 |
| 246 M_START armSP_FFTInv_CToC_FC32_Radix4_fs_OutOfPlace_unsafe,r4 |
| 247 FFTSTAGE "FALSE","TRUE",inv |
| 248 M_END |
| 249 |
| 250 |
| 251 .end |
OLD | NEW |