tnt has uploaded this change for review. ( https://gerrit.osmocom.org/12415
Change subject: convolve: Remove support for step, offset parameters ...................................................................... convolve: Remove support for step, offset parameters - Those are not used any where - Those are not supported by the sse/neon accelerated versions - And I see very little use cases for those. Change-Id: Ic850269a0ed5d98c0ea68980afd31016ed555b48 Signed-off-by: Sylvain Munaut <[email protected]> --- M Transceiver52M/Channelizer.cpp M Transceiver52M/Resampler.cpp M Transceiver52M/Synthesis.cpp M Transceiver52M/arch/arm/convolve.c M Transceiver52M/arch/common/convolve.h M Transceiver52M/arch/common/convolve_base.c M Transceiver52M/arch/x86/convolve.c M Transceiver52M/arch/x86/convolve_sse_3.c M Transceiver52M/arch/x86/convolve_sse_3.h M Transceiver52M/sigProcLib.cpp M tests/Transceiver52M/convolve_test.c 11 files changed, 124 insertions(+), 166 deletions(-) git pull ssh://gerrit.osmocom.org:29418/osmo-trx refs/changes/15/12415/1 diff --git a/Transceiver52M/Channelizer.cpp b/Transceiver52M/Channelizer.cpp index 2d817b0..a18dd03 100644 --- a/Transceiver52M/Channelizer.cpp +++ b/Transceiver52M/Channelizer.cpp @@ -88,7 +88,7 @@ convolve_real(hInputs[i], blockLen, subFilters[i], hLen, hOutputs[i], blockLen, - 0, blockLen, 1, 0); + 0, blockLen); } cxvec_fft(fftHandle); diff --git a/Transceiver52M/Resampler.cpp b/Transceiver52M/Resampler.cpp index 6b9621c..ecd8865 100644 --- a/Transceiver52M/Resampler.cpp +++ b/Transceiver52M/Resampler.cpp @@ -143,7 +143,7 @@ convolve_real(in, in_len, reinterpret_cast<float *>(partitions[path]), filt_len, &out[2 * i], out_len - i, - n, 1, 1, 0); + n, 1); } return out_len; diff --git a/Transceiver52M/Synthesis.cpp b/Transceiver52M/Synthesis.cpp index 262c638..6b62156 100644 --- a/Transceiver52M/Synthesis.cpp +++ b/Transceiver52M/Synthesis.cpp @@ -102,7 +102,7 @@ convolve_real(hInputs[i], blockLen, subFilters[i], hLen, hOutputs[i], blockLen, - 0, blockLen, 1, 0); + 0, blockLen); } /* Interleave into output vector */ diff --git a/Transceiver52M/arch/arm/convolve.c b/Transceiver52M/arch/arm/convolve.c index 912d0c2..5b5bce5 100644 --- a/Transceiver52M/arch/arm/convolve.c +++ b/Transceiver52M/arch/arm/convolve.c @@ -29,17 +29,15 @@ int _base_convolve_real(float *x, int x_len, float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset); + int start, int len); int _base_convolve_complex(float *x, int x_len, float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset); + int start, int len); int bounds_check(int x_len, int h_len, int y_len, - int start, int len, int step); + int start, int len); #ifdef HAVE_NEON /* Calls into NEON assembler */ @@ -69,35 +67,32 @@ int convolve_real(float *x, int x_len, float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset) + int start, int len) { void (*conv_func)(float *, float *, float *, int) = NULL; - if (bounds_check(x_len, h_len, y_len, start, len, step) < 0) + if (bounds_check(x_len, h_len, y_len, start, len) < 0) return -1; memset(y, 0, len * 2 * sizeof(float)); #ifdef HAVE_NEON - if (step <= 4) { - switch (h_len) { - case 4: - conv_func = neon_conv_real4; - break; - case 8: - conv_func = neon_conv_real8; - break; - case 12: - conv_func = neon_conv_real12; - break; - case 16: - conv_func = neon_conv_real16; - break; - case 20: - conv_func = neon_conv_real20; - break; - } + switch (h_len) { + case 4: + conv_func = neon_conv_real4; + break; + case 8: + conv_func = neon_conv_real8; + break; + case 12: + conv_func = neon_conv_real12; + break; + case 16: + conv_func = neon_conv_real16; + break; + case 20: + conv_func = neon_conv_real20; + break; } #endif if (conv_func) { @@ -107,7 +102,7 @@ _base_convolve_real(x, x_len, h, h_len, y, y_len, - start, len, step, offset); + start, len); } return len; @@ -118,18 +113,17 @@ int convolve_complex(float *x, int x_len, float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset) + int start, int len) { void (*conv_func)(float *, float *, float *, int, int) = NULL; - if (bounds_check(x_len, h_len, y_len, start, len, step) < 0) + if (bounds_check(x_len, h_len, y_len, start, len) < 0) return -1; memset(y, 0, len * 2 * sizeof(float)); #ifdef HAVE_NEON - if (step <= 4 && !(h_len % 4)) + if (!(h_len % 4)) conv_func = neon_conv_cmplx_4n; #endif if (conv_func) { @@ -139,7 +133,7 @@ _base_convolve_complex(x, x_len, h, h_len, y, y_len, - start, len, step, offset); + start, len); } return len; diff --git a/Transceiver52M/arch/common/convolve.h b/Transceiver52M/arch/common/convolve.h index 095b04c..e30f7ec 100644 --- a/Transceiver52M/arch/common/convolve.h +++ b/Transceiver52M/arch/common/convolve.h @@ -6,26 +6,22 @@ int convolve_real(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset); + int start, int len); int convolve_complex(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset); + int start, int len); int base_convolve_real(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset); + int start, int len); int base_convolve_complex(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset); + int start, int len); void convolve_init(void); diff --git a/Transceiver52M/arch/common/convolve_base.c b/Transceiver52M/arch/common/convolve_base.c index 2eb7124..9bb8d3d 100644 --- a/Transceiver52M/arch/common/convolve_base.c +++ b/Transceiver52M/arch/common/convolve_base.c @@ -41,17 +41,17 @@ /* Base vector complex-complex multiply and accumulate */ static void mac_real_vec_n(const float *x, const float *h, float *y, - int len, int step, int offset) + int len) { - for (int i = offset; i < len; i += step) + for (int i=0; i<len; i++) mac_real(&x[2 * i], &h[2 * i], y); } /* Base vector complex-complex multiply and accumulate */ static void mac_cmplx_vec_n(const float *x, const float *h, float *y, - int len, int step, int offset) + int len) { - for (int i = offset; i < len; i += step) + for (int i=0; i<len; i++) mac_cmplx(&x[2 * i], &h[2 * i], y); } @@ -59,14 +59,12 @@ int _base_convolve_real(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset) + int start, int len) { for (int i = 0; i < len; i++) { mac_real_vec_n(&x[2 * (i - (h_len - 1) + start)], h, - &y[2 * i], h_len, - step, offset); + &y[2 * i], h_len); } return len; @@ -76,14 +74,13 @@ int _base_convolve_complex(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset) + int start, int len) { for (int i = 0; i < len; i++) { mac_cmplx_vec_n(&x[2 * (i - (h_len - 1) + start)], h, &y[2 * i], - h_len, step, offset); + h_len); } return len; @@ -91,10 +88,10 @@ /* Buffer validity checks */ int bounds_check(int x_len, int h_len, int y_len, - int start, int len, int step) + int start, int len) { if ((x_len < 1) || (h_len < 1) || - (y_len < 1) || (len < 1) || (step < 1)) { + (y_len < 1) || (len < 1)) { fprintf(stderr, "Convolve: Invalid input\n"); return -1; } @@ -113,10 +110,9 @@ int base_convolve_real(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset) + int start, int len) { - if (bounds_check(x_len, h_len, y_len, start, len, step) < 0) + if (bounds_check(x_len, h_len, y_len, start, len) < 0) return -1; memset(y, 0, len * 2 * sizeof(float)); @@ -124,17 +120,16 @@ return _base_convolve_real(x, x_len, h, h_len, y, y_len, - start, len, step, offset); + start, len); } /* API: Non-aligned (no SSE) complex-complex */ int base_convolve_complex(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset) + int start, int len) { - if (bounds_check(x_len, h_len, y_len, start, len, step) < 0) + if (bounds_check(x_len, h_len, y_len, start, len) < 0) return -1; memset(y, 0, len * 2 * sizeof(float)); @@ -142,7 +137,7 @@ return _base_convolve_complex(x, x_len, h, h_len, y, y_len, - start, len, step, offset); + start, len); } /* Aligned filter tap allocation */ diff --git a/Transceiver52M/arch/x86/convolve.c b/Transceiver52M/arch/x86/convolve.c index eb38f64..209d377 100644 --- a/Transceiver52M/arch/x86/convolve.c +++ b/Transceiver52M/arch/x86/convolve.c @@ -30,25 +30,25 @@ /* Architecture dependant function pointers */ struct convolve_cpu_context { void (*conv_cmplx_4n) (const float *, int, const float *, int, float *, - int, int, int, int, int); + int, int, int); void (*conv_cmplx_8n) (const float *, int, const float *, int, float *, - int, int, int, int, int); + int, int, int); void (*conv_cmplx) (const float *, int, const float *, int, float *, - int, int, int, int, int); + int, int, int); void (*conv_real4) (const float *, int, const float *, int, float *, - int, int, int, int, int); + int, int, int); void (*conv_real8) (const float *, int, const float *, int, float *, - int, int, int, int, int); + int, int, int); void (*conv_real12) (const float *, int, const float *, int, float *, - int, int, int, int, int); + int, int, int); void (*conv_real16) (const float *, int, const float *, int, float *, - int, int, int, int, int); + int, int, int); void (*conv_real20) (const float *, int, const float *, int, float *, - int, int, int, int, int); + int, int, int); void (*conv_real4n) (const float *, int, const float *, int, float *, - int, int, int, int, int); + int, int, int); void (*conv_real) (const float *, int, const float *, int, float *, int, - int, int, int, int); + int, int); }; static struct convolve_cpu_context c; @@ -56,17 +56,15 @@ int _base_convolve_real(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset); + int start, int len); int _base_convolve_complex(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, - int step, int offset); + int start, int len); int bounds_check(int x_len, int h_len, int y_len, - int start, int len, int step); + int start, int len); /* API: Initalize convolve module */ void convolve_init(void) @@ -99,46 +97,37 @@ /* API: Aligned complex-real */ int convolve_real(const float *x, int x_len, const float *h, int h_len, - float *y, int y_len, int start, int len, int step, int offset) + float *y, int y_len, int start, int len) { - if (bounds_check(x_len, h_len, y_len, start, len, step) < 0) + if (bounds_check(x_len, h_len, y_len, start, len) < 0) return -1; memset(y, 0, len * 2 * sizeof(float)); - if (step <= 4) { - switch (h_len) { - case 4: - c.conv_real4(x, x_len, h, h_len, y, y_len, start, len, - step, offset); - break; - case 8: - c.conv_real8(x, x_len, h, h_len, y, y_len, start, len, - step, offset); - break; - case 12: - c.conv_real12(x, x_len, h, h_len, y, y_len, start, len, - step, offset); - break; - case 16: - c.conv_real16(x, x_len, h, h_len, y, y_len, start, len, - step, offset); - break; - case 20: - c.conv_real20(x, x_len, h, h_len, y, y_len, start, len, - step, offset); - break; - default: - if (!(h_len % 4)) - c.conv_real4n(x, x_len, h, h_len, y, y_len, - start, len, step, offset); - else - c.conv_real(x, x_len, h, h_len, y, y_len, start, - len, step, offset); - } - } else - c.conv_real(x, x_len, h, h_len, y, y_len, start, len, step, - offset); + switch (h_len) { + case 4: + c.conv_real4(x, x_len, h, h_len, y, y_len, start, len); + break; + case 8: + c.conv_real8(x, x_len, h, h_len, y, y_len, start, len); + break; + case 12: + c.conv_real12(x, x_len, h, h_len, y, y_len, start, len); + break; + case 16: + c.conv_real16(x, x_len, h, h_len, y, y_len, start, len); + break; + case 20: + c.conv_real20(x, x_len, h, h_len, y, y_len, start, len); + break; + default: + if (!(h_len % 4)) + c.conv_real4n(x, x_len, h, h_len, y, y_len, + start, len); + else + c.conv_real(x, x_len, h, h_len, y, y_len, start, + len); + } return len; } @@ -147,26 +136,19 @@ int convolve_complex(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset) + int start, int len) { - if (bounds_check(x_len, h_len, y_len, start, len, step) < 0) + if (bounds_check(x_len, h_len, y_len, start, len) < 0) return -1; memset(y, 0, len * 2 * sizeof(float)); - if (step <= 4) { - if (!(h_len % 8)) - c.conv_cmplx_8n(x, x_len, h, h_len, y, y_len, start, - len, step, offset); - else if (!(h_len % 4)) - c.conv_cmplx_4n(x, x_len, h, h_len, y, y_len, start, - len, step, offset); - else - c.conv_cmplx(x, x_len, h, h_len, y, y_len, start, len, - step, offset); - } else - c.conv_cmplx(x, x_len, h, h_len, y, y_len, start, len, step, - offset); + if (!(h_len % 8)) + c.conv_cmplx_8n(x, x_len, h, h_len, y, y_len, start, len); + else if (!(h_len % 4)) + c.conv_cmplx_4n(x, x_len, h, h_len, y, y_len, start, len); + else + c.conv_cmplx(x, x_len, h, h_len, y, y_len, start, len); return len; } diff --git a/Transceiver52M/arch/x86/convolve_sse_3.c b/Transceiver52M/arch/x86/convolve_sse_3.c index dbee3cc..8fd3b5e 100644 --- a/Transceiver52M/arch/x86/convolve_sse_3.c +++ b/Transceiver52M/arch/x86/convolve_sse_3.c @@ -34,12 +34,12 @@ void sse_conv_real4(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset) + int start, int len) { /* NOTE: The parameter list of this function has to match the parameter * list of _base_convolve_real() in convolve_base.c. This specific * implementation, ignores some of the parameters of - * _base_convolve_complex(), which are: x_len, y_len, offset, step */ + * _base_convolve_complex(), which are: x_len, y_len. */ __m128 m0, m1, m2, m3, m4, m5, m6, m7; @@ -75,7 +75,7 @@ void sse_conv_real8(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset) + int start, int len) { /* See NOTE in sse_conv_real4() */ @@ -126,7 +126,7 @@ void sse_conv_real12(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset) + int start, int len) { /* See NOTE in sse_conv_real4() */ @@ -192,7 +192,7 @@ void sse_conv_real16(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset) + int start, int len) { /* See NOTE in sse_conv_real4() */ @@ -271,7 +271,7 @@ void sse_conv_real20(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset) + int start, int len) { /* See NOTE in sse_conv_real4() */ @@ -361,7 +361,7 @@ void sse_conv_real4n(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset) + int start, int len) { /* See NOTE in sse_conv_real4() */ @@ -408,12 +408,12 @@ void sse_conv_cmplx_4n(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset) + int start, int len) { /* NOTE: The parameter list of this function has to match the parameter * list of _base_convolve_complex() in convolve_base.c. This specific * implementation, ignores some of the parameters of - * _base_convolve_complex(), which are: x_len, y_len, offset, step. */ + * _base_convolve_complex(), which are: x_len, y_len. */ __m128 m0, m1, m2, m3, m4, m5, m6, m7; @@ -466,7 +466,7 @@ void sse_conv_cmplx_8n(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset) + int start, int len) { /* See NOTE in sse_conv_cmplx_4n() */ diff --git a/Transceiver52M/arch/x86/convolve_sse_3.h b/Transceiver52M/arch/x86/convolve_sse_3.h index ac30ca5..d929ef6 100644 --- a/Transceiver52M/arch/x86/convolve_sse_3.h +++ b/Transceiver52M/arch/x86/convolve_sse_3.h @@ -23,46 +23,46 @@ void sse_conv_real4(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset); + int start, int len); /* 8-tap SSE complex-real convolution */ void sse_conv_real8(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset); + int start, int len); /* 12-tap SSE complex-real convolution */ void sse_conv_real12(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset); + int start, int len); /* 16-tap SSE complex-real convolution */ void sse_conv_real16(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset); + int start, int len); /* 20-tap SSE complex-real convolution */ void sse_conv_real20(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset); + int start, int len); /* 4*N-tap SSE complex-real convolution */ void sse_conv_real4n(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset); + int start, int len); /* 4*N-tap SSE complex-complex convolution */ void sse_conv_cmplx_4n(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset); + int start, int len); /* 8*N-tap SSE complex-complex convolution */ void sse_conv_cmplx_8n(const float *x, int x_len, const float *h, int h_len, float *y, int y_len, - int start, int len, int step, int offset); + int start, int len); diff --git a/Transceiver52M/sigProcLib.cpp b/Transceiver52M/sigProcLib.cpp index f720828..fcda5fa 100644 --- a/Transceiver52M/sigProcLib.cpp +++ b/Transceiver52M/sigProcLib.cpp @@ -285,8 +285,7 @@ static signalVector *convolve(const signalVector *x, const signalVector *h, signalVector *y, ConvType spanType, - size_t start = 0, size_t len = 0, - size_t step = 1, int offset = 0) + size_t start = 0, size_t len = 0) { int rc; size_t head = 0, tail = 0; @@ -354,22 +353,22 @@ rc = convolve_real((float *) _x->begin(), _x->size(), (float *) h->begin(), h->size(), (float *) y->begin(), y->size(), - start, len, step, offset); + start, len); } else if (!h->isReal() && h->isAligned()) { rc = convolve_complex((float *) _x->begin(), _x->size(), (float *) h->begin(), h->size(), (float *) y->begin(), y->size(), - start, len, step, offset); + start, len); } else if (h->isReal() && !h->isAligned()) { rc = base_convolve_real((float *) _x->begin(), _x->size(), (float *) h->begin(), h->size(), (float *) y->begin(), y->size(), - start, len, step, offset); + start, len); } else if (!h->isReal() && !h->isAligned()) { rc = base_convolve_complex((float *) _x->begin(), _x->size(), (float *) h->begin(), h->size(), (float *) y->begin(), y->size(), - start, len, step, offset); + start, len); } else { rc = -1; } @@ -1482,7 +1481,7 @@ /* Correlate */ if (!convolve(corr_in, sync->sequence, &corr, - CUSTOM, start, len, 1, 0)) { + CUSTOM, start, len)) { delete dec; return -1; } diff --git a/tests/Transceiver52M/convolve_test.c b/tests/Transceiver52M/convolve_test.c index 54bc7a1..8ca4b72 100644 --- a/tests/Transceiver52M/convolve_test.c +++ b/tests/Transceiver52M/convolve_test.c @@ -62,21 +62,17 @@ int y_len; int start; int len; - int step; - int offset; x_len=34; y_len=26; start=8; len=26; - step=1; - offset=1; reset_testvec(0); dump_floats(x,x_len,"x"); printf("\n"); dump_floats(h,h_len,"h"); printf("\n"); - convolve_complex(x, x_len, h, h_len, y, y_len, start, len, step, offset); + convolve_complex(x, x_len, h, h_len, y, y_len, start, len); dump_floats(y,y_len,"y"); printf("\n"); } @@ -88,21 +84,17 @@ int y_len; int start; int len; - int step; - int offset; x_len=34; y_len=26; start=8; len=26; - step=1; - offset=1; reset_testvec(0); - dump_floats(x,x_len,"x"); + dump_floats(x-30,2*x_len+30,"x"); printf("\n"); - dump_floats(h,h_len,"h"); + dump_floats(h,2*h_len,"h"); printf("\n"); - convolve_real(x, x_len, h, h_len, y, y_len, start, len, step, offset); + convolve_real(x, x_len, h, h_len, y, y_len, start, len); dump_floats(y,y_len,"y"); printf("\n"); } -- To view, visit https://gerrit.osmocom.org/12415 To unsubscribe, or for help writing mail filters, visit https://gerrit.osmocom.org/settings Gerrit-Project: osmo-trx Gerrit-Branch: master Gerrit-MessageType: newchange Gerrit-Change-Id: Ic850269a0ed5d98c0ea68980afd31016ed555b48 Gerrit-Change-Number: 12415 Gerrit-PatchSet: 1 Gerrit-Owner: tnt <[email protected]>
