search for: sig_shift

Hi all, This is Linfeng Zhang from Google. I'll work on ARM NEON optimization in the next few months. I'm submitting 2 patches in the following couple of emails, which have the new created celt_fir_neon(). I revised celt_fir_c() to not pass in argument "mem" in Patch 1. If there are concerns to this change, please let me know. Many thanks to your comments. Linfeng Zhang

...ord-i-4))); + for (;i<ord;i++) + rnum[i] = num[ord-i-1]; + rnum[ord] = rnum[ord+1] = rnum[ord+2] = 0; + (void)arch; + +#ifdef SMALL_FOOTPRINT + for (i=0;i<N-7;i+=8) + { + int16x8_t x_s16x8 = vld1q_s16(_x+i); + int32x4_t sum0_s32x4 = vshll_n_s16(vget_low_s16 (x_s16x8), SIG_SHIFT); + int32x4_t sum1_s32x4 = vshll_n_s16(vget_high_s16(x_s16x8), SIG_SHIFT); + for (j=0;j<ord;j+=4) + { + const int16x4_t rnum_s16x4 = vld1_s16(rnum+j); + x_s16x8 = vld1q_s16(x+i+j+0); + sum0_s32x4 = vmlal_lane_s16(sum0_s32x4, vget_low_s16 (x_s16x8), rnum_s16...

Hi Jean-Marc, I could be way off base here, but it seems to me that line 115 in pitch.c in the function "pitch_downsample": x_lp[i] = SHR32(HALF32(HALF32(x[1][(2*i-1)]+x[1][(2*i+1)])+x[1][2*i]), SIG_SHIFT+2); should actually be: x_lp[i] += SHR32(HALF32(HALF32(x[1][(2*i-1)]+x[1][(2*i+1)])+x[1][2*i]), SIG_SHIFT+2); Sorry if I'm totally misreading things and just wasting your time. Cheers, John Ridges

HI trying to build libopus beta compiler complains about more fixed point specific names referenced in common unconditional code Particularly: ./celt/x86/celt_lpc_sse.c(100): error: identifier "SIG_SHIFT" is undefined noA = EXTEND32(1) << SIG_SHIFT >> 1; ^ SIG_SHIFT is defined in arch.h provided FIXED_POINT is used (not on floating point build) and silk/x86/x86_silk_map.c unconditionally references alot names defined in fixed point sources for exam...

I rebased my previous 3 patches to the current master with minor changes. Patches 1 to 3 replace all my previous submitted patches. Patches 4 and 5 are new. Thanks, Linfeng Zhang

.....p] autocorrelation values */ @@ -87,6 +91,7 @@ int p #endif } +#ifndef OVERRIDE_CELT_FIR void celt_fir(const opus_val16 *x, const opus_val16 *num, opus_val16 *y, @@ -101,7 +106,7 @@ void celt_fir(const opus_val16 *x, opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); for (j=0;j<ord;j++) { - sum += MULT16_16(num[j],mem[j]); + sum = MAC16_16(sum, num[j], mem[j]); } for (j=ord-1;j>=1;j--) { @@ -111,6 +116,7 @@ void celt_fir(const opus_val16 *x, y[i] = ROUND16(sum, SIG_SHIFT); } } +#endif voi...

> On Nov 16, 2015, at 4:42 PM, Jonathan Lennox <jonathan at vidyo.com> wrote: > > I haven?t yet tried replacing SIG2WORD16 (or silk_ADD_SAT32/silk_SUB_SAT32) with Neon intrinsics. That?s an obvious next step. This doesn?t show any appreciable speed difference in my tests, but the code is obviously better by inspection (all three of these map directly to a single Aarch64

.....p] autocorrelation values */ @@ -87,6 +91,7 @@ int p #endif } +#ifndef OVERRIDE_CELT_FIR void celt_fir(const opus_val16 *x, const opus_val16 *num, opus_val16 *y, @@ -101,7 +106,7 @@ void celt_fir(const opus_val16 *x, opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); for (j=0;j<ord;j++) { - sum += MULT16_16(num[j],mem[j]); + sum = MAC16_16(sum, num[j], mem[j]); } for (j=ord-1;j>=1;j--) { @@ -111,6 +116,7 @@ void celt_fir(const opus_val16 *x, y[i] = ROUND16(sum, SIG_SHIFT); } } +#endif voi...

...#39;s it never worked. I found innov_save will write over memory it shouldn't be when I fill memory with 1's but not when I fill it with zeros. When my DSP gets to this code chunk: if (innov_save){ for (i=0;i<st->subframeSize;i++) innov_save[i] = EXTRACT16(PSHR32(innov[i], SIG_SHIFT)); } it will just start filling data in, which it shouldn't. I see that innov_save is set at the beginning of a for loop at: for (sub=0;sub<st->nbSubframes;sub++) { int offset; spx_word16_t *exc; spx_word16_t *sp; spx_word16_t *innov_save = NULL; spx_word...

Hi! I have some questions about that functions: fir_mem16, iir_mem16 and filter_mem16. Filtering is very slow on TI DSP, and i want to optimise it. Can somebody give me formulas which discribe work of this filters? Or any suggestions about how to transform code for better performance. I going to implement this functions in assembler, but it is hard to do without full understanding how functions

...rated. > celt_fir_permit_overflow(): x[i] is accumulated to the sum first and > then truncated saturated. Actually, the two branches are bit-exact for fixed-point. There is indeed a difference in where x[i] gets accumulated, but because in the SMALL_FOOTPRINT case it first gets shifted up by SIG_SHIFT, the result of the downshift (also by SIG_SHIFT) is the same no matter when it gets added. That being said, I thought adding at the beginning was nicer so I changed the remaining code to do that. > Maybe this is the reason why silk_LPC_analysis_filter() switched the FIR > from celt_fir() to...

..._t; typedef spx_word16_t spx_coef_t; typedef spx_word16_t spx_lsp_t; typedef spx_word32_t spx_sig_t; #define LPC_SCALING 8192 #define SIG_SCALING 16384 #define LSP_SCALING 8192. #define GAMMA_SCALING 32768. #define GAIN_SCALING 64 #define GAIN_SCALING_1 0.015625 #define LPC_SHIFT 13 #define SIG_SHIFT 14 #define VERY_SMALL 0 #ifdef ARM_ASM #include "fixed_arm.h" #elif defined (FIXED_DEBUG) #include "fixed_debug.h" #else #include "fixed_generic.h" #endif #else typedef float spx_mem_t; typedef float spx_coef_t; typedef float spx_lsp_t; typedef float spx_sig_...

..._t; typedef spx_word16_t spx_coef_t; typedef spx_word16_t spx_lsp_t; typedef spx_word32_t spx_sig_t; #define LPC_SCALING 8192 #define SIG_SCALING 16384 #define LSP_SCALING 8192. #define GAMMA_SCALING 32768. #define GAIN_SCALING 64 #define GAIN_SCALING_1 0.015625 #define LPC_SHIFT 13 #define SIG_SHIFT 14 #define VERY_SMALL 0 #ifdef ARM5E_ASM #include "fixed_arm5e.h" #elif defined (ARM4_ASM) #include "fixed_arm4.h" #elif defined (FIXED_DEBUG) #include "fixed_debug.h" #elif defined (C55X_ASM) #include "fixed_c55x.h" #else #include "fixed_generic.h...

...32 bit registers. Can anyone tell me what scaling values like #define LPC_SCALING 8192 #define SIG_SCALING 16384 #define LSP_SCALING 8192. #define GAMMA_SCALING 32768. #define GAIN_SCALING 64 #define GAIN_SCALING_1 0.015625 #define LPC_SHIFT 13 #define LSP_SHIFT 13 #define SIG_SHIFT 14 Would be required for a 24 bit register? 2. I would like to only include only the data tables as required to support 11kbps narrowband mode. I think they would be the following signed byte tables cdbk_nb[640] cdbk_nb_low1[320] cdbk_nb_high1[320] gain_cdbk_lbr[128] e...

...ig_t *x, spx_word16_t *y, spx_sig_t max_scale, int le "LOOP_END norm_max%=;\n\t" : "=&d" (max_val) : "a" (x), "a" (len) - : "R1", "R2" + : "R1", "R2", "ASTAT" BFIN_HWLOOP0_REGS ); sig_shift=0; @@ -74,7 +76,7 @@ int normalize16(const spx_sig_t *x, spx_word16_t *y, spx_sig_t max_scale, int le "R1 = ASHIFT R0 by %2.L;\n\t" "W[P1++] = R1;\n\t" : : "a" (x), "a" (y), "d" (-sig_shift), "a" (len-1) - : "I0", &...

Hi Jean-Marc, The original celt_fir() is a little bit messy. It has 2 branches chosen by #ifdef SMALL_FOOTPRINT. For floating-point, the 2 branches are identical (except the operation sequence of accumulating x[i] to sum, which is not a big deal). For fixed-point, the 2 branches are different. I separate them into 2 functions: the new celt_fir(), and celt_fir_permit_overflow() which is the

Hi Jean-Marc, Based on the wonderful Speex project, I've created SpeexOutLoud, essentially a Speex codec port for Windows Mobile 2003 devices. I've included a sample project intended to show the usage of SpeexOutLoud codec in a Pocket PC application based on .NET Compact Framework. I'd request you to please go through the attached build, and include it as a contribution to the

...emory it shouldn't be when > I fill memory with 1's but not when I fill it with zeros. Which is normal, see below. > When my DSP > gets to this code chunk: > if (innov_save){ > for (i=0;i<st->subframeSize;i++) > innov_save[i] = EXTRACT16(PSHR32(innov[i], SIG_SHIFT)); > } This bit of code is for copying the innov variable to a buffer owned by the wideband decoder -- but only if there is actually a wideband decoder. > it will just start filling data in, which it shouldn't. I see that > innov_save is set at the beginning of a for loop at: > f...

Alexander Chemeris a ?crit : > Ok, if I comment out inclusion of "filters_arm4.h" (or comment out its > only overridden function - normalize16()) it works fine. > > Also it works fine if I use -O0 for compilation. Specifying -O1 or -O2 lead > to segfault (if "filters_arm4.h" is included, sure). OK, so either I screwed up the alignment/constraints in the

Any speedup from the intrinsics may just be swamped by the rest of the encode/decode process. But I think you really want SIG2WORD16 to be (vqmovns_s32(PSHR32((x), SIG_SHIFT))) On 11/19/2015 2:52 PM, Jonathan Lennox wrote: >> On Nov 16, 2015, at 4:42 PM, Jonathan Lennox <jonathan at vidyo.com> wrote: >> >> I haven?t yet tried replacing SIG2WORD16 (or silk_ADD_SAT32/silk_SUB_SAT32) with Neon intrinsics. That?s an obvious next step. > This d...