Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(7)

Unified Diff: third_party/openmax_dl/dl/sp/src/armSP_FFT_CToC_FC32_Radix4_fs_unsafe_s.S

Issue 12317152: Add openmax dl routines for review. MUST NOT BE LANDED (Closed) Base URL: http://git.chromium.org/chromium/src.git@master
Patch Set: Created 7 years, 9 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
Index: third_party/openmax_dl/dl/sp/src/armSP_FFT_CToC_FC32_Radix4_fs_unsafe_s.S
diff --git a/third_party/openmax_dl/dl/sp/src/armSP_FFT_CToC_FC32_Radix4_fs_unsafe_s.S b/third_party/openmax_dl/dl/sp/src/armSP_FFT_CToC_FC32_Radix4_fs_unsafe_s.S
new file mode 100644
index 0000000000000000000000000000000000000000..ff62dd132b820513873bab1c0f47f2c78bc2dbea
--- /dev/null
+++ b/third_party/openmax_dl/dl/sp/src/armSP_FFT_CToC_FC32_Radix4_fs_unsafe_s.S
@@ -0,0 +1,251 @@
+@//
+@// Copyright (c) 2013 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.
+@//
+@//
+@// This is a modification of armSP_FFT_CToC_SC32_Radix4_fs_unsafe_s.s
+@// to support float instead of SC32.
+@//
+
+@//
+@// Description:
+@// Compute a first stage Radix 4 FFT stage for a N point complex signal
+@//
+@//
+
+
+@// Include standard headers
+
+#include "dl/api/armCOMM_s.h"
+#include "dl/api/omxtypes_s.h"
+
+@// Import symbols required from other files
+@// (For example tables)
+
+
+
+
+@// Set debugging level
+@//DEBUG_ON SETL {TRUE}
+
+
+
+@// Guarding implementation by the processor name
+
+
+
+@// Guarding implementation by the processor name
+
+
+@//Input Registers
+
+#define pSrc r0
+#define pDst r2
+#define pTwiddle r1
+#define pPingPongBuf r5
+#define subFFTNum r6
+#define subFFTSize r7
+
+
+@//Output Registers
+
+
+@//Local Scratch Registers
+
+#define grpSize r3
+@// Reuse grpSize as setCount
+#define setCount r3
+#define pointStep r4
+#define outPointStep r4
+#define setStep r8
+#define step1 r9
+#define step3 r10
+
+@// Neon Registers
+
+#define dXr0 D0.F32
+#define dXi0 D1.F32
+#define dXr1 D2.F32
+#define dXi1 D3.F32
+#define dXr2 D4.F32
+#define dXi2 D5.F32
+#define dXr3 D6.F32
+#define dXi3 D7.F32
+#define dYr0 D8.F32
+#define dYi0 D9.F32
+#define dYr1 D10.F32
+#define dYi1 D11.F32
+#define dYr2 D12.F32
+#define dYi2 D13.F32
+#define dYr3 D14.F32
+#define dYi3 D15.F32
+#define qX0 Q0.F32
+#define qX1 Q1.F32
+#define qX2 Q2.F32
+#define qX3 Q3.F32
+#define qY0 Q4.F32
+#define qY1 Q5.F32
+#define qY2 Q6.F32
+#define qY3 Q7.F32
+#define dZr0 D16.F32
+#define dZi0 D17.F32
+#define dZr1 D18.F32
+#define dZi1 D19.F32
+#define dZr2 D20.F32
+#define dZi2 D21.F32
+#define dZr3 D22.F32
+#define dZi3 D23.F32
+#define qZ0 Q8.F32
+#define qZ1 Q9.F32
+#define qZ2 Q10.F32
+#define qZ3 Q11.F32
+
+
+ .MACRO FFTSTAGE scaled, inverse, name
+
+ @// Define stack arguments
+
+ @// pT0+1 increments pT0 by 8 bytes
+ @// pT0+pointStep = increment of 8*pointStep bytes = 2*grpSize bytes
+ @// Note: outPointStep = pointStep for firststage
+
+ MOV pointStep,subFFTNum,LSL #1
+
+
+ @// Update pSubFFTSize and pSubFFTNum regs
+ VLD2 {dXr0,dXi0},[pSrc :128],pointStep @// data[0]
+ @// subFFTSize = 1 for the first stage
+ MOV subFFTSize,#4
+
+ @// Note: setCount = subFFTNum/4 (reuse the grpSize reg for setCount)
+ LSR grpSize,subFFTNum,#2
+ VLD2 {dXr1,dXi1},[pSrc :128],pointStep @// data[1]
+ MOV subFFTNum,grpSize
+
+
+ @// Calculate the step of input data for the next set
+ @//MOV setStep,pointStep,LSL #1
+ MOV setStep,grpSize,LSL #4
+ VLD2 {dXr2,dXi2},[pSrc :128],pointStep @// data[2]
+ @// setStep = 3*pointStep
+ ADD setStep,setStep,pointStep
+ @// setStep = - 3*pointStep+16
+ RSB setStep,setStep,#16
+
+ @// data[3] & update pSrc for the next set
+ VLD2 {dXr3,dXi3},[pSrc :128],setStep
+ @// step1 = 2*pointStep
+ MOV step1,pointStep,LSL #1
+
+ VADD qY0,qX0,qX2
+
+ @// step3 = -pointStep
+ RSB step3,pointStep,#0
+
+ @// grp = 0 a special case since all the twiddle factors are 1
+ @// Loop on the sets : 2 sets at a time
+
+radix4fsGrpZeroSetLoop\name :
+
+
+
+ @// Decrement setcount
+ SUBS setCount,setCount,#2
+
+
+ @// finish first stage of 4 point FFT
+
+
+ VSUB qY2,qX0,qX2
+
+ VLD2 {dXr0,dXi0},[pSrc :128],step1 @// data[0]
+ VADD qY1,qX1,qX3
+ VLD2 {dXr2,dXi2},[pSrc :128],step3 @// data[2]
+ VSUB qY3,qX1,qX3
+
+
+ @// finish second stage of 4 point FFT
+
+ .ifeqs "\inverse", "TRUE"
+
+ VLD2 {dXr1,dXi1},[pSrc :128],step1 @// data[1]
+ VADD qZ0,qY0,qY1
+
+ @// data[3] & update pSrc for the next set, but not if it's the
+ @// last iteration so that we don't read past the end of the
+ @// input array.
+ BEQ radix4SkipLastUpdateInv\name
+ VLD2 {dXr3,dXi3},[pSrc :128],setStep
+radix4SkipLastUpdateInv\name:
+ VSUB dZr3,dYr2,dYi3
+
+ VST2 {dZr0,dZi0},[pDst :128],outPointStep
+ VADD dZi3,dYi2,dYr3
+
+ VSUB qZ1,qY0,qY1
+ VST2 {dZr3,dZi3},[pDst :128],outPointStep
+
+ VADD dZr2,dYr2,dYi3
+ VST2 {dZr1,dZi1},[pDst :128],outPointStep
+ VSUB dZi2,dYi2,dYr3
+
+ VADD qY0,qX0,qX2 @// u0 for next iteration
+ VST2 {dZr2,dZi2},[pDst :128],setStep
+
+
+ .else
+
+ VLD2 {dXr1,dXi1},[pSrc :128],step1 @// data[1]
+ VADD qZ0,qY0,qY1
+
+ @// data[3] & update pSrc for the next set, but not if it's the
+ @// last iteration so that we don't read past the end of the
+ @// input array.
+ BEQ radix4SkipLastUpdateFwd\name
+ VLD2 {dXr3,dXi3},[pSrc :128],setStep
+radix4SkipLastUpdateFwd\name:
+ VADD dZr2,dYr2,dYi3
+
+ VST2 {dZr0,dZi0},[pDst :128],outPointStep
+ VSUB dZi2,dYi2,dYr3
+
+ VSUB qZ1,qY0,qY1
+ VST2 {dZr2,dZi2},[pDst :128],outPointStep
+
+ VSUB dZr3,dYr2,dYi3
+ VST2 {dZr1,dZi1},[pDst :128],outPointStep
+ VADD dZi3,dYi2,dYr3
+
+ VADD qY0,qX0,qX2 @// u0 for next iteration
+ VST2 {dZr3,dZi3},[pDst :128],setStep
+
+ .endif
+
+ BGT radix4fsGrpZeroSetLoop\name
+
+ @// reset pSrc to pDst for the next stage
+ SUB pSrc,pDst,pointStep @// pDst -= 2*grpSize
+ MOV pDst,pPingPongBuf
+
+
+ .endm
+
+
+
+ M_START armSP_FFTFwd_CToC_FC32_Radix4_fs_OutOfPlace_unsafe,r4
+ FFTSTAGE "FALSE","FALSE",fwd
+ M_END
+
+
+
+ M_START armSP_FFTInv_CToC_FC32_Radix4_fs_OutOfPlace_unsafe,r4
+ FFTSTAGE "FALSE","TRUE",inv
+ M_END
+
+
+ .end

Powered by Google App Engine
This is Rietveld 408576698