diff options
Diffstat (limited to 'drivers/staging/echo')
-rw-r--r-- | drivers/staging/echo/bit_operations.h | 187 | ||||
-rw-r--r-- | drivers/staging/echo/echo.c | 803 | ||||
-rw-r--r-- | drivers/staging/echo/echo.h | 7 | ||||
-rw-r--r-- | drivers/staging/echo/fir.h | 298 | ||||
-rw-r--r-- | drivers/staging/echo/mmx.h | 29 | ||||
-rw-r--r-- | drivers/staging/echo/oslec.h | 2 |
6 files changed, 654 insertions, 672 deletions
diff --git a/drivers/staging/echo/bit_operations.h b/drivers/staging/echo/bit_operations.h index abcd7a2..cecdcf3 100644 --- a/drivers/staging/echo/bit_operations.h +++ b/drivers/staging/echo/bit_operations.h @@ -36,14 +36,15 @@ \return The bit number of the highest set bit, or -1 if the word is zero. */ static __inline__ int top_bit(unsigned int bits) { - int res; - - __asm__ (" xorl %[res],%[res];\n" - " decl %[res];\n" - " bsrl %[bits],%[res]\n" - : [res] "=&r" (res) - : [bits] "rm" (bits)); - return res; + int res; + + __asm__(" xorl %[res],%[res];\n" + " decl %[res];\n" + " bsrl %[bits],%[res]\n" + :[res] "=&r" (res) + :[bits] "rm"(bits) + ); + return res; } /*! \brief Find the bit position of the lowest set bit in a word @@ -51,84 +52,75 @@ static __inline__ int top_bit(unsigned int bits) \return The bit number of the lowest set bit, or -1 if the word is zero. */ static __inline__ int bottom_bit(unsigned int bits) { - int res; - - __asm__ (" xorl %[res],%[res];\n" - " decl %[res];\n" - " bsfl %[bits],%[res]\n" - : [res] "=&r" (res) - : [bits] "rm" (bits)); - return res; + int res; + + __asm__(" xorl %[res],%[res];\n" + " decl %[res];\n" + " bsfl %[bits],%[res]\n" + :[res] "=&r" (res) + :[bits] "rm"(bits) + ); + return res; } #else static __inline__ int top_bit(unsigned int bits) { - int i; - - if (bits == 0) - return -1; - i = 0; - if (bits & 0xFFFF0000) - { - bits &= 0xFFFF0000; - i += 16; - } - if (bits & 0xFF00FF00) - { - bits &= 0xFF00FF00; - i += 8; - } - if (bits & 0xF0F0F0F0) - { - bits &= 0xF0F0F0F0; - i += 4; - } - if (bits & 0xCCCCCCCC) - { - bits &= 0xCCCCCCCC; - i += 2; - } - if (bits & 0xAAAAAAAA) - { - bits &= 0xAAAAAAAA; - i += 1; - } - return i; + int i; + + if (bits == 0) + return -1; + i = 0; + if (bits & 0xFFFF0000) { + bits &= 0xFFFF0000; + i += 16; + } + if (bits & 0xFF00FF00) { + bits &= 0xFF00FF00; + i += 8; + } + if (bits & 0xF0F0F0F0) { + bits &= 0xF0F0F0F0; + i += 4; + } + if (bits & 0xCCCCCCCC) { + bits &= 0xCCCCCCCC; + i += 2; + } + if (bits & 0xAAAAAAAA) { + bits &= 0xAAAAAAAA; + i += 1; + } + return i; } static __inline__ int bottom_bit(unsigned int bits) { - int i; - - if (bits == 0) - return -1; - i = 32; - if (bits & 0x0000FFFF) - { - bits &= 0x0000FFFF; - i -= 16; - } - if (bits & 0x00FF00FF) - { - bits &= 0x00FF00FF; - i -= 8; - } - if (bits & 0x0F0F0F0F) - { - bits &= 0x0F0F0F0F; - i -= 4; - } - if (bits & 0x33333333) - { - bits &= 0x33333333; - i -= 2; - } - if (bits & 0x55555555) - { - bits &= 0x55555555; - i -= 1; - } - return i; + int i; + + if (bits == 0) + return -1; + i = 32; + if (bits & 0x0000FFFF) { + bits &= 0x0000FFFF; + i -= 16; + } + if (bits & 0x00FF00FF) { + bits &= 0x00FF00FF; + i -= 8; + } + if (bits & 0x0F0F0F0F) { + bits &= 0x0F0F0F0F; + i -= 4; + } + if (bits & 0x33333333) { + bits &= 0x33333333; + i -= 2; + } + if (bits & 0x55555555) { + bits &= 0x55555555; + i -= 1; + } + return i; } #endif @@ -138,13 +130,14 @@ static __inline__ int bottom_bit(unsigned int bits) static __inline__ uint8_t bit_reverse8(uint8_t x) { #if defined(__i386__) || defined(__x86_64__) - /* If multiply is fast */ - return ((x*0x0802U & 0x22110U) | (x*0x8020U & 0x88440U))*0x10101U >> 16; + /* If multiply is fast */ + return ((x * 0x0802U & 0x22110U) | (x * 0x8020U & 0x88440U)) * + 0x10101U >> 16; #else - /* If multiply is slow, but we have a barrel shifter */ - x = (x >> 4) | (x << 4); - x = ((x & 0xCC) >> 2) | ((x & 0x33) << 2); - return ((x & 0xAA) >> 1) | ((x & 0x55) << 1); + /* If multiply is slow, but we have a barrel shifter */ + x = (x >> 4) | (x << 4); + x = ((x & 0xCC) >> 2) | ((x & 0x33) << 2); + return ((x & 0xAA) >> 1) | ((x & 0x55) << 1); #endif } @@ -184,7 +177,7 @@ uint16_t make_mask16(uint16_t x); \return The word with the single set bit. */ static __inline__ uint32_t least_significant_one32(uint32_t x) { - return (x & (-(int32_t) x)); + return (x & (-(int32_t) x)); } /*! \brief Find the most significant one in a word, and return a word @@ -194,10 +187,10 @@ static __inline__ uint32_t least_significant_one32(uint32_t x) static __inline__ uint32_t most_significant_one32(uint32_t x) { #if defined(__i386__) || defined(__x86_64__) - return 1 << top_bit(x); + return 1 << top_bit(x); #else - x = make_mask32(x); - return (x ^ (x >> 1)); + x = make_mask32(x); + return (x ^ (x >> 1)); #endif } @@ -206,8 +199,8 @@ static __inline__ uint32_t most_significant_one32(uint32_t x) \return 1 for odd, or 0 for even. */ static __inline__ int parity8(uint8_t x) { - x = (x ^ (x >> 4)) & 0x0F; - return (0x6996 >> x) & 1; + x = (x ^ (x >> 4)) & 0x0F; + return (0x6996 >> x) & 1; } /*! \brief Find the parity of a 16 bit word. @@ -215,9 +208,9 @@ static __inline__ int parity8(uint8_t x) \return 1 for odd, or 0 for even. */ static __inline__ int parity16(uint16_t x) { - x ^= (x >> 8); - x = (x ^ (x >> 4)) & 0x0F; - return (0x6996 >> x) & 1; + x ^= (x >> 8); + x = (x ^ (x >> 4)) & 0x0F; + return (0x6996 >> x) & 1; } /*! \brief Find the parity of a 32 bit word. @@ -225,10 +218,10 @@ static __inline__ int parity16(uint16_t x) \return 1 for odd, or 0 for even. */ static __inline__ int parity32(uint32_t x) { - x ^= (x >> 16); - x ^= (x >> 8); - x = (x ^ (x >> 4)) & 0x0F; - return (0x6996 >> x) & 1; + x ^= (x >> 16); + x ^= (x >> 8); + x = (x ^ (x >> 4)) & 0x0F; + return (0x6996 >> x) & 1; } #endif diff --git a/drivers/staging/echo/echo.c b/drivers/staging/echo/echo.c index ad18a95..b8f2c5e 100644 --- a/drivers/staging/echo/echo.c +++ b/drivers/staging/echo/echo.c @@ -74,7 +74,6 @@ Steve also has some nice notes on echo cancellers in echo.h - References: [1] Ochiai, Areseki, and Ogihara, "Echo Canceller with Two Echo @@ -105,7 +104,7 @@ Mark, Pawel, and Pavel. */ -#include <linux/kernel.h> /* We're doing kernel work */ +#include <linux/kernel.h> /* We're doing kernel work */ #include <linux/module.h> #include <linux/kernel.h> #include <linux/slab.h> @@ -115,8 +114,8 @@ #define MIN_TX_POWER_FOR_ADAPTION 64 #define MIN_RX_POWER_FOR_ADAPTION 64 -#define DTD_HANGOVER 600 /* 600 samples, or 75ms */ -#define DC_LOG2BETA 3 /* log2() of DC filter Beta */ +#define DTD_HANGOVER 600 /* 600 samples, or 75ms */ +#define DC_LOG2BETA 3 /* log2() of DC filter Beta */ /*-----------------------------------------------------------------------*\ FUNCTIONS @@ -124,59 +123,58 @@ /* adapting coeffs using the traditional stochastic descent (N)LMS algorithm */ - #ifdef __bfin__ -static void __inline__ lms_adapt_bg(struct oslec_state *ec, int clean, int shift) +static void __inline__ lms_adapt_bg(struct oslec_state *ec, int clean, + int shift) { - int i, j; - int offset1; - int offset2; - int factor; - int exp; - int16_t *phist; - int n; - - if (shift > 0) - factor = clean << shift; - else - factor = clean >> -shift; - - /* Update the FIR taps */ - - offset2 = ec->curr_pos; - offset1 = ec->taps - offset2; - phist = &ec->fir_state_bg.history[offset2]; - - /* st: and en: help us locate the assembler in echo.s */ - - //asm("st:"); - n = ec->taps; - for (i = 0, j = offset2; i < n; i++, j++) - { - exp = *phist++ * factor; - ec->fir_taps16[1][i] += (int16_t) ((exp+(1<<14)) >> 15); - } - //asm("en:"); - - /* Note the asm for the inner loop above generated by Blackfin gcc - 4.1.1 is pretty good (note even parallel instructions used): - - R0 = W [P0++] (X); - R0 *= R2; - R0 = R0 + R3 (NS) || - R1 = W [P1] (X) || - nop; - R0 >>>= 15; - R0 = R0 + R1; - W [P1++] = R0; - - A block based update algorithm would be much faster but the - above can't be improved on much. Every instruction saved in - the loop above is 2 MIPs/ch! The for loop above is where the - Blackfin spends most of it's time - about 17 MIPs/ch measured - with speedtest.c with 256 taps (32ms). Write-back and - Write-through cache gave about the same performance. - */ + int i, j; + int offset1; + int offset2; + int factor; + int exp; + int16_t *phist; + int n; + + if (shift > 0) + factor = clean << shift; + else + factor = clean >> -shift; + + /* Update the FIR taps */ + + offset2 = ec->curr_pos; + offset1 = ec->taps - offset2; + phist = &ec->fir_state_bg.history[offset2]; + + /* st: and en: help us locate the assembler in echo.s */ + + //asm("st:"); + n = ec->taps; + for (i = 0, j = offset2; i < n; i++, j++) { + exp = *phist++ * factor; + ec->fir_taps16[1][i] += (int16_t) ((exp + (1 << 14)) >> 15); + } + //asm("en:"); + + /* Note the asm for the inner loop above generated by Blackfin gcc + 4.1.1 is pretty good (note even parallel instructions used): + + R0 = W [P0++] (X); + R0 *= R2; + R0 = R0 + R3 (NS) || + R1 = W [P1] (X) || + nop; + R0 >>>= 15; + R0 = R0 + R1; + W [P1++] = R0; + + A block based update algorithm would be much faster but the + above can't be improved on much. Every instruction saved in + the loop above is 2 MIPs/ch! The for loop above is where the + Blackfin spends most of it's time - about 17 MIPs/ch measured + with speedtest.c with 256 taps (32ms). Write-back and + Write-through cache gave about the same performance. + */ } /* @@ -198,94 +196,90 @@ static void __inline__ lms_adapt_bg(struct oslec_state *ec, int clean, int shift */ #else -static __inline__ void lms_adapt_bg(struct oslec_state *ec, int clean, int shift) +static __inline__ void lms_adapt_bg(struct oslec_state *ec, int clean, + int shift) { - int i; - - int offset1; - int offset2; - int factor; - int exp; - - if (shift > 0) - factor = clean << shift; - else - factor = clean >> -shift; - - /* Update the FIR taps */ - - offset2 = ec->curr_pos; - offset1 = ec->taps - offset2; - - for (i = ec->taps - 1; i >= offset1; i--) - { - exp = (ec->fir_state_bg.history[i - offset1]*factor); - ec->fir_taps16[1][i] += (int16_t) ((exp+(1<<14)) >> 15); - } - for ( ; i >= 0; i--) - { - exp = (ec->fir_state_bg.history[i + offset2]*factor); - ec->fir_taps16[1][i] += (int16_t) ((exp+(1<<14)) >> 15); - } + int i; + + int offset1; + int offset2; + int factor; + int exp; + + if (shift > 0) + factor = clean << shift; + else + factor = clean >> -shift; + + /* Update the FIR taps */ + + offset2 = ec->curr_pos; + offset1 = ec->taps - offset2; + + for (i = ec->taps - 1; i >= offset1; i--) { + exp = (ec->fir_state_bg.history[i - offset1] * factor); + ec->fir_taps16[1][i] += (int16_t) ((exp + (1 << 14)) >> 15); + } + for (; i >= 0; i--) { + exp = (ec->fir_state_bg.history[i + offset2] * factor); + ec->fir_taps16[1][i] += (int16_t) ((exp + (1 << 14)) >> 15); + } } #endif - struct oslec_state *oslec_create(int len, int adaption_mode) { - struct oslec_state *ec; - int i; - - ec = kzalloc(sizeof(*ec), GFP_KERNEL); - if (!ec) - return NULL; - - ec->taps = len; - ec->log2taps = top_bit(len); - ec->curr_pos = ec->taps - 1; - - for (i = 0; i < 2; i++) { - ec->fir_taps16[i] = kcalloc(ec->taps, sizeof(int16_t), GFP_KERNEL); - if (!ec->fir_taps16[i]) - goto error_oom; - } - - fir16_create(&ec->fir_state, - ec->fir_taps16[0], - ec->taps); - fir16_create(&ec->fir_state_bg, - ec->fir_taps16[1], - ec->taps); - - for(i=0; i<5; i++) { - ec->xvtx[i] = ec->yvtx[i] = ec->xvrx[i] = ec->yvrx[i] = 0; - } - - ec->cng_level = 1000; - oslec_adaption_mode(ec, adaption_mode); - - ec->snapshot = kcalloc(ec->taps, sizeof(int16_t), GFP_KERNEL); - if (!ec->snapshot) - goto error_oom; - - ec->cond_met = 0; - ec->Pstates = 0; - ec->Ltxacc = ec->Lrxacc = ec->Lcleanacc = ec->Lclean_bgacc = 0; - ec->Ltx = ec->Lrx = ec->Lclean = ec->Lclean_bg = 0; - ec->tx_1 = ec->tx_2 = ec->rx_1 = ec->rx_2 = 0; - ec->Lbgn = ec->Lbgn_acc = 0; - ec->Lbgn_upper = 200; - ec->Lbgn_upper_acc = ec->Lbgn_upper << 13; - - return ec; - -error_oom: - for (i = 0; i < 2; i++) - kfree(ec->fir_taps16[i]); - - kfree(ec); - return NULL; + struct oslec_state *ec; + int i; + + ec = kzalloc(sizeof(*ec), GFP_KERNEL); + if (!ec) + return NULL; + + ec->taps = len; + ec->log2taps = top_bit(len); + ec->curr_pos = ec->taps - 1; + + for (i = 0; i < 2; i++) { + ec->fir_taps16[i] = + kcalloc(ec->taps, sizeof(int16_t), GFP_KERNEL); + if (!ec->fir_taps16[i]) + goto error_oom; + } + + fir16_create(&ec->fir_state, ec->fir_taps16[0], ec->taps); + fir16_create(&ec->fir_state_bg, ec->fir_taps16[1], ec->taps); + + for (i = 0; i < 5; i++) { + ec->xvtx[i] = ec->yvtx[i] = ec->xvrx[i] = ec->yvrx[i] = 0; + } + + ec->cng_level = 1000; + oslec_adaption_mode(ec, adaption_mode); + + ec->snapshot = kcalloc(ec->taps, sizeof(int16_t), GFP_KERNEL); + if (!ec->snapshot) + goto error_oom; + + ec->cond_met = 0; + ec->Pstates = 0; + ec->Ltxacc = ec->Lrxacc = ec->Lcleanacc = ec->Lclean_bgacc = 0; + ec->Ltx = ec->Lrx = ec->Lclean = ec->Lclean_bg = 0; + ec->tx_1 = ec->tx_2 = ec->rx_1 = ec->rx_2 = 0; + ec->Lbgn = ec->Lbgn_acc = 0; + ec->Lbgn_upper = 200; + ec->Lbgn_upper_acc = ec->Lbgn_upper << 13; + + return ec; + + error_oom: + for (i = 0; i < 2; i++) + kfree(ec->fir_taps16[i]); + + kfree(ec); + return NULL; } + EXPORT_SYMBOL_GPL(oslec_create); void oslec_free(struct oslec_state *ec) @@ -294,293 +288,300 @@ void oslec_free(struct oslec_state *ec) fir16_free(&ec->fir_state); fir16_free(&ec->fir_state_bg); - for (i = 0; i < 2; i++) + for (i = 0; i < 2; i++) kfree(ec->fir_taps16[i]); kfree(ec->snapshot); kfree(ec); } + EXPORT_SYMBOL_GPL(oslec_free); void oslec_adaption_mode(struct oslec_state *ec, int adaption_mode) { - ec->adaption_mode = adaption_mode; + ec->adaption_mode = adaption_mode; } + EXPORT_SYMBOL_GPL(oslec_adaption_mode); void oslec_flush(struct oslec_state *ec) { - int i; + int i; - ec->Ltxacc = ec->Lrxacc = ec->Lcleanacc = ec->Lclean_bgacc = 0; - ec->Ltx = ec->Lrx = ec->Lclean = ec->Lclean_bg = 0; - ec->tx_1 = ec->tx_2 = ec->rx_1 = ec->rx_2 = 0; + ec->Ltxacc = ec->Lrxacc = ec->Lcleanacc = ec->Lclean_bgacc = 0; + ec->Ltx = ec->Lrx = ec->Lclean = ec->Lclean_bg = 0; + ec->tx_1 = ec->tx_2 = ec->rx_1 = ec->rx_2 = 0; - ec->Lbgn = ec->Lbgn_acc = 0; - ec->Lbgn_upper = 200; - ec->Lbgn_upper_acc = ec->Lbgn_upper << 13; + ec->Lbgn = ec->Lbgn_acc = 0; + ec->Lbgn_upper = 200; + ec->Lbgn_upper_acc = ec->Lbgn_upper << 13; - ec->nonupdate_dwell = 0; + ec->nonupdate_dwell = 0; - fir16_flush(&ec->fir_state); - fir16_flush(&ec->fir_state_bg); - ec->fir_state.curr_pos = ec->taps - 1; - ec->fir_state_bg.curr_pos = ec->taps - 1; - for (i = 0; i < 2; i++) - memset(ec->fir_taps16[i], 0, ec->taps*sizeof(int16_t)); + fir16_flush(&ec->fir_state); + fir16_flush(&ec->fir_state_bg); + ec->fir_state.curr_pos = ec->taps - 1; + ec->fir_state_bg.curr_pos = ec->taps - 1; + for (i = 0; i < 2; i++) + memset(ec->fir_taps16[i], 0, ec->taps * sizeof(int16_t)); - ec->curr_pos = ec->taps - 1; - ec->Pstates = 0; + ec->curr_pos = ec->taps - 1; + ec->Pstates = 0; } + EXPORT_SYMBOL_GPL(oslec_flush); -void oslec_snapshot(struct oslec_state *ec) { - memcpy(ec->snapshot, ec->fir_taps16[0], ec->taps*sizeof(int16_t)); +void oslec_snapshot(struct oslec_state *ec) +{ + memcpy(ec->snapshot, ec->fir_taps16[0], ec->taps * sizeof(int16_t)); } + EXPORT_SYMBOL_GPL(oslec_snapshot); /* Dual Path Echo Canceller ------------------------------------------------*/ int16_t oslec_update(struct oslec_state *ec, int16_t tx, int16_t rx) { - int32_t echo_value; - int clean_bg; - int tmp, tmp1; - - /* Input scaling was found be required to prevent problems when tx - starts clipping. Another possible way to handle this would be the - filter coefficent scaling. */ - - ec->tx = tx; ec->rx = rx; - tx >>=1; - rx >>=1; - - /* - Filter DC, 3dB point is 160Hz (I think), note 32 bit precision required - otherwise values do not track down to 0. Zero at DC, Pole at (1-Beta) - only real axis. Some chip sets (like Si labs) don't need - this, but something like a $10 X100P card does. Any DC really slows - down convergence. - - Note: removes some low frequency from the signal, this reduces - the speech quality when listening to samples through headphones - but may not be obvious through a telephone handset. - - Note that the 3dB frequency in radians is approx Beta, e.g. for - Beta = 2^(-3) = 0.125, 3dB freq is 0.125 rads = 159Hz. - */ - - if (ec->adaption_mode & ECHO_CAN_USE_RX_HPF) { - tmp = rx << 15; + int32_t echo_value; + int clean_bg; + int tmp, tmp1; + + /* Input scaling was found be required to prevent problems when tx + starts clipping. Another possible way to handle this would be the + filter coefficent scaling. */ + + ec->tx = tx; + ec->rx = rx; + tx >>= 1; + rx >>= 1; + + /* + Filter DC, 3dB point is 160Hz (I think), note 32 bit precision required + otherwise values do not track down to 0. Zero at DC, Pole at (1-Beta) + only real axis. Some chip sets (like Si labs) don't need + this, but something like a $10 X100P card does. Any DC really slows + down convergence. + + Note: removes some low frequency from the signal, this reduces + the speech quality when listening to samples through headphones + but may not be obvious through a telephone handset. + + Note that the 3dB frequency in radians is approx Beta, e.g. for + Beta = 2^(-3) = 0.125, 3dB freq is 0.125 rads = 159Hz. + */ + + if (ec->adaption_mode & ECHO_CAN_USE_RX_HPF) { + tmp = rx << 15; #if 1 - /* Make sure the gain of the HPF is 1.0. This can still saturate a little under - impulse conditions, and it might roll to 32768 and need clipping on sustained peak - level signals. However, the scale of such clipping is small, and the error due to - any saturation should not markedly affect the downstream processing. */ - tmp -= (tmp >> 4); + /* Make sure the gain of the HPF is 1.0. This can still saturate a little under + impulse conditions, and it might roll to 32768 and need clipping on sustained peak + level signals. However, the scale of such clipping is small, and the error due to + any saturation should not markedly affect the downstream processing. */ + tmp -= (tmp >> 4); #endif - ec->rx_1 += -(ec->rx_1>>DC_LOG2BETA) + tmp - ec->rx_2; + ec->rx_1 += -(ec->rx_1 >> DC_LOG2BETA) + tmp - ec->rx_2; + + /* hard limit filter to prevent clipping. Note that at this stage + rx should be limited to +/- 16383 due to right shift above */ + tmp1 = ec->rx_1 >> 15; + if (tmp1 > 16383) + tmp1 = 16383; + if (tmp1 < -16383) + tmp1 = -16383; + rx = tmp1; + ec->rx_2 = tmp; + } - /* hard limit filter to prevent clipping. Note that at this stage - rx should be limited to +/- 16383 due to right shift above */ - tmp1 = ec->rx_1 >> 15; - if (tmp1 > 16383) tmp1 = 16383; - if (tmp1 < -16383) tmp1 = -16383; - rx = tmp1; - ec->rx_2 = tmp; - } + /* Block average of power in the filter states. Used for + adaption power calculation. */ - /* Block average of power in the filter states. Used for - adaption power calculation. */ + { + int new, old; + + /* efficient "out with the old and in with the new" algorithm so + we don't have to recalculate over the whole block of + samples. */ + new = (int)tx *(int)tx; + old = (int)ec->fir_state.history[ec->fir_state.curr_pos] * + (int)ec->fir_state.history[ec->fir_state.curr_pos]; + ec->Pstates += + ((new - old) + (1 << ec->log2taps)) >> ec->log2taps; + if (ec->Pstates < 0) + ec->Pstates = 0; + } - { - int new, old; + /* Calculate short term average levels using simple single pole IIRs */ - /* efficient "out with the old and in with the new" algorithm so - we don't have to recalculate over the whole block of - samples. */ - new = (int)tx * (int)tx; - old = (int)ec->fir_state.history[ec->fir_state.curr_pos] * - (int)ec->fir_state.history[ec->fir_state.curr_pos]; - ec->Pstates += ((new - old) + (1<<ec->log2taps)) >> ec->log2taps; - if (ec->Pstates < 0) ec->Pstates = 0; - } - - /* Calculate short term average levels using simple single pole IIRs */ - - ec->Ltxacc += abs(tx) - ec->Ltx; - ec->Ltx = (ec->Ltxacc + (1<<4)) >> 5; - ec->Lrxacc += abs(rx) - ec->Lrx; - ec->Lrx = (ec->Lrxacc + (1<<4)) >> 5; - - /* Foreground filter ---------------------------------------------------*/ - - ec->fir_state.coeffs = ec->fir_taps16[0]; - echo_value = fir16(&ec->fir_state, tx); - ec->clean = rx - echo_value; - ec->Lcleanacc += abs(ec->clean) - ec->Lclean; - ec->Lclean = (ec->Lcleanacc + (1<<4)) >> 5; - - /* Background filter ---------------------------------------------------*/ - - echo_value = fir16(&ec->fir_state_bg, tx); - clean_bg = rx - echo_value; - ec->Lclean_bgacc += abs(clean_bg) - ec->Lclean_bg; - ec->Lclean_bg = (ec->Lclean_bgacc + (1<<4)) >> 5; - - /* Background Filter adaption -----------------------------------------*/ - - /* Almost always adap bg filter, just simple DT and energy - detection to minimise adaption in cases of strong double talk. - However this is not critical for the dual path algorithm. - */ - ec->factor = 0; - ec->shift = 0; - if ((ec->nonupdate_dwell == 0)) { - int P, logP, shift; - - /* Determine: - - f = Beta * clean_bg_rx/P ------ (1) - - where P is the total power in the filter states. - - The Boffins have shown that if we obey (1) we converge - quickly and avoid instability. - - The correct factor f must be in Q30, as this is the fixed - point format required by the lms_adapt_bg() function, - therefore the scaled version of (1) is: - - (2^30) * f = (2^30) * Beta * clean_bg_rx/P - factor = (2^30) * Beta * clean_bg_rx/P ----- (2) - - We have chosen Beta = 0.25 by experiment, so: - - factor = (2^30) * (2^-2) * clean_bg_rx/P - - (30 - 2 - log2(P)) - factor = clean_bg_rx 2 ----- (3) - - To avoid a divide we approximate log2(P) as top_bit(P), - which returns the position of the highest non-zero bit in - P. This approximation introduces an error as large as a - factor of 2, but the algorithm seems to handle it OK. - - Come to think of it a divide may not be a big deal on a - modern DSP, so its probably worth checking out the cycles - for a divide versus a top_bit() implementation. - */ - - P = MIN_TX_POWER_FOR_ADAPTION + ec->Pstates; - logP = top_bit(P) + ec->log2taps; - shift = 30 - 2 - logP; - ec->shift = shift; - - lms_adapt_bg(ec, clean_bg, shift); - } - - /* very simple DTD to make sure we dont try and adapt with strong - near end speech */ - - ec->adapt = 0; - if ((ec->Lrx > MIN_RX_POWER_FOR_ADAPTION) && (ec->Lrx > ec->Ltx)) - ec->nonupdate_dwell = DTD_HANGOVER; - if (ec->nonupdate_dwell) - ec->nonupdate_dwell--; + ec->Ltxacc += abs(tx) - ec->Ltx; + ec->Ltx = (ec->Ltxacc + (1 << 4)) >> 5; + ec->Lrxacc += abs(rx) - ec->Lrx; + ec->Lrx = (ec->Lrxacc + (1 << 4)) >> 5; - /* Transfer logic ------------------------------------------------------*/ + /* Foreground filter --------------------------------------------------- */ - /* These conditions are from the dual path paper [1], I messed with - them a bit to improve performance. */ + ec->fir_state.coeffs = ec->fir_taps16[0]; + echo_value = fir16(&ec->fir_state, tx); + ec->clean = rx - echo_value; + ec->Lcleanacc += abs(ec->clean) - ec->Lclean; + ec->Lclean = (ec->Lcleanacc + (1 << 4)) >> 5; - if ((ec->adaption_mode & ECHO_CAN_USE_ADAPTION) && - (ec->nonupdate_dwell == 0) && - (8*ec->Lclean_bg < 7*ec->Lclean) /* (ec->Lclean_bg < 0.875*ec->Lclean) */ && - (8*ec->Lclean_bg < ec->Ltx) /* (ec->Lclean_bg < 0.125*ec->Ltx) */ ) - { - if (ec->cond_met == 6) { - /* BG filter has had better results for 6 consecutive samples */ - ec->adapt = 1; - memcpy(ec->fir_taps16[0], ec->fir_taps16[1], ec->taps*sizeof(int16_t)); - } - else - ec->cond_met++; - } - else - ec->cond_met = 0; + /* Background filter --------------------------------------------------- */ - /* Non-Linear Processing ---------------------------------------------------*/ + echo_value = fir16(&ec->fir_state_bg, tx); + clean_bg = rx - echo_value; + ec->Lclean_bgacc += abs(clean_bg) - ec->Lclean_bg; + ec->Lclean_bg = (ec->Lclean_bgacc + (1 << 4)) >> 5; - ec->clean_nlp = ec->clean; - if (ec->adaption_mode & ECHO_CAN_USE_NLP) - { - /* Non-linear processor - a fancy way to say "zap small signals, to avoid - residual echo due to (uLaw/ALaw) non-linearity in the channel.". */ + /* Background Filter adaption ----------------------------------------- */ - if ((16*ec->Lclean < ec->Ltx)) - { - /* Our e/c has improved echo by at least 24 dB (each factor of 2 is 6dB, - so 2*2*2*2=16 is the same as 6+6+6+6=24dB) */ - if (ec->adaption_mode & ECHO_CAN_USE_CNG) - { - ec->cng_level = ec->Lbgn; - - /* Very elementary comfort noise generation. Just random - numbers rolled off very vaguely Hoth-like. DR: This - noise doesn't sound quite right to me - I suspect there - are some overlfow issues in the filtering as it's too - "crackly". TODO: debug this, maybe just play noise at - high level or look at spectrum. - */ - - ec->cng_rndnum = 1664525U*ec->cng_rndnum + 1013904223U; - ec->cng_filter = ((ec->cng_rndnum & 0xFFFF) - 32768 + 5*ec->cng_filter) >> 3; - ec->clean_nlp = (ec->cng_filter*ec->cng_level*8) >> 14; - - } - else if (ec->adaption_mode & ECHO_CAN_USE_CLIP) - { - /* This sounds much better than CNG */ - if (ec->clean_nlp > ec->Lbgn) - ec->clean_nlp = ec->Lbgn; - if (ec->clean_nlp < -ec->Lbgn) - ec->clean_nlp = -ec->Lbgn; + /* Almost always adap bg filter, just simple DT and energy + detection to minimise adaption in cases of strong double talk. + However this is not critical for the dual path algorithm. + */ + ec->factor = 0; + ec->shift = 0; + if ((ec->nonupdate_dwell == 0)) { + int P, logP, shift; + + /* Determine: + + f = Beta * clean_bg_rx/P ------ (1) + + where P is the total power in the filter states. + + The Boffins have shown that if we obey (1) we converge + quickly and avoid instability. + + The correct factor f must be in Q30, as this is the fixed + point format required by the lms_adapt_bg() function, + therefore the scaled version of (1) is: + + (2^30) * f = (2^30) * Beta * clean_bg_rx/P + factor = (2^30) * Beta * clean_bg_rx/P ----- (2) + + We have chosen Beta = 0.25 by experiment, so: + + factor = (2^30) * (2^-2) * clean_bg_rx/P + + (30 - 2 - log2(P)) + factor = clean_bg_rx 2 ----- (3) + + To avoid a divide we approximate log2(P) as top_bit(P), + which returns the position of the highest non-zero bit in + P. This approximation introduces an error as large as a + factor of 2, but the algorithm seems to handle it OK. + + Come to think of it a divide may not be a big deal on a + modern DSP, so its probably worth checking out the cycles + for a divide versus a top_bit() implementation. + */ + + P = MIN_TX_POWER_FOR_ADAPTION + ec->Pstates; + logP = top_bit(P) + ec->log2taps; + shift = 30 - 2 - logP; + ec->shift = shift; + + lms_adapt_bg(ec, clean_bg, shift); } - else - { - /* just mute the residual, doesn't sound very good, used mainly - in G168 tests */ - ec->clean_nlp = 0; - } - } - else { - /* Background noise estimator. I tried a few algorithms - here without much luck. This very simple one seems to - work best, we just average the level using a slow (1 sec - time const) filter if the current level is less than a - (experimentally derived) constant. This means we dont - include high level signals like near end speech. When - combined with CNG or especially CLIP seems to work OK. - */ - if (ec->Lclean < 40) { - ec->Lbgn_acc += abs(ec->clean) - ec->Lbgn; - ec->Lbgn = (ec->Lbgn_acc + (1<<11)) >> 12; - } - } - } - - /* Roll around the taps buffer */ - if (ec->curr_pos <= 0) - ec->curr_pos = ec->taps; - ec->curr_pos--; - - if (ec->adaption_mode & ECHO_CAN_DISABLE) - ec->clean_nlp = rx; - - /* Output scaled back up again to match input scaling */ - - return (int16_t) ec->clean_nlp << 1; + + /* very simple DTD to make sure we dont try and adapt with strong + near end speech */ + + ec->adapt = 0; + if ((ec->Lrx > MIN_RX_POWER_FOR_ADAPTION) && (ec->Lrx > ec->Ltx)) + ec->nonupdate_dwell = DTD_HANGOVER; + if (ec->nonupdate_dwell) + ec->nonupdate_dwell--; + + /* Transfer logic ------------------------------------------------------ */ + + /* These conditions are from the dual path paper [1], I messed with + them a bit to improve performance. */ + + if ((ec->adaption_mode & ECHO_CAN_USE_ADAPTION) && + (ec->nonupdate_dwell == 0) && + (8 * ec->Lclean_bg < + 7 * ec->Lclean) /* (ec->Lclean_bg < 0.875*ec->Lclean) */ && + (8 * ec->Lclean_bg < + ec->Ltx) /* (ec->Lclean_bg < 0.125*ec->Ltx) */ ) { + if (ec->cond_met == 6) { + /* BG filter has had better results for 6 consecutive samples */ + ec->adapt = 1; + memcpy(ec->fir_taps16[0], ec->fir_taps16[1], + ec->taps * sizeof(int16_t)); + } else + ec->cond_met++; + } else + ec->cond_met = 0; + + /* Non-Linear Processing --------------------------------------------------- */ + + ec->clean_nlp = ec->clean; + if (ec->adaption_mode & ECHO_CAN_USE_NLP) { + /* Non-linear processor - a fancy way to say "zap small signals, to avoid + residual echo due to (uLaw/ALaw) non-linearity in the channel.". */ + + if ((16 * ec->Lclean < ec->Ltx)) { + /* Our e/c has improved echo by at least 24 dB (each factor of 2 is 6dB, + so 2*2*2*2=16 is the same as 6+6+6+6=24dB) */ + if (ec->adaption_mode & ECHO_CAN_USE_CNG) { + ec->cng_level = ec->Lbgn; + + /* Very elementary comfort noise generation. Just random + numbers rolled off very vaguely Hoth-like. DR: This + noise doesn't sound quite right to me - I suspect there + are some overlfow issues in the filtering as it's too + "crackly". TODO: debug this, maybe just play noise at + high level or look at spectrum. + */ + + ec->cng_rndnum = + 1664525U * ec->cng_rndnum + 1013904223U; + ec->cng_filter = + ((ec->cng_rndnum & 0xFFFF) - 32768 + + 5 * ec->cng_filter) >> 3; + ec->clean_nlp = + (ec->cng_filter * ec->cng_level * 8) >> 14; + + } else if (ec->adaption_mode & ECHO_CAN_USE_CLIP) { + /* This sounds much better than CNG */ + if (ec->clean_nlp > ec->Lbgn) + ec->clean_nlp = ec->Lbgn; + if (ec->clean_nlp < -ec->Lbgn) + ec->clean_nlp = -ec->Lbgn; + } else { + /* just mute the residual, doesn't sound very good, used mainly + in G168 tests */ + ec->clean_nlp = 0; + } + } else { + /* Background noise estimator. I tried a few algorithms + here without much luck. This very simple one seems to + work best, we just average the level using a slow (1 sec + time const) filter if the current level is less than a + (experimentally derived) constant. This means we dont + include high level signals like near end speech. When + combined with CNG or especially CLIP seems to work OK. + */ + if (ec->Lclean < 40) { + ec->Lbgn_acc += abs(ec->clean) - ec->Lbgn; + ec->Lbgn = (ec->Lbgn_acc + (1 << 11)) >> 12; + } + } + } + + /* Roll around the taps buffer */ + if (ec->curr_pos <= 0) + ec->curr_pos = ec->taps; + ec->curr_pos--; + + if (ec->adaption_mode & ECHO_CAN_DISABLE) + ec->clean_nlp = rx; + + /* Output scaled back up again to match input scaling */ + + return (int16_t) ec->clean_nlp << 1; } + EXPORT_SYMBOL_GPL(oslec_update); /* This function is seperated from the echo canceller is it is usually called @@ -604,28 +605,32 @@ EXPORT_SYMBOL_GPL(oslec_update); precision, which noise shapes things, giving very clean DC removal. */ -int16_t oslec_hpf_tx(struct oslec_state *ec, int16_t tx) { - int tmp, tmp1; +int16_t oslec_hpf_tx(struct oslec_state * ec, int16_t tx) +{ + int tmp, tmp1; - if (ec->adaption_mode & ECHO_CAN_USE_TX_HPF) { - tmp = tx << 15; + if (ec->adaption_mode & ECHO_CAN_USE_TX_HPF) { + tmp = tx << 15; #if 1 - /* Make sure the gain of the HPF is 1.0. The first can still saturate a little under - impulse conditions, and it might roll to 32768 and need clipping on sustained peak - level signals. However, the scale of such clipping is small, and the error due to - any saturation should not markedly affect the downstream processing. */ - tmp -= (tmp >> 4); + /* Make sure the gain of the HPF is 1.0. The first can still saturate a little under + impulse conditions, and it might roll to 32768 and need clipping on sustained peak + level signals. However, the scale of such clipping is small, and the error due to + any saturation should not markedly affect the downstream processing. */ + tmp -= (tmp >> 4); #endif - ec->tx_1 += -(ec->tx_1>>DC_LOG2BETA) + tmp - ec->tx_2; - tmp1 = ec->tx_1 >> 15; - if (tmp1 > 32767) tmp1 = 32767; - if (tmp1 < -32767) tmp1 = -32767; - tx = tmp1; - ec->tx_2 = tmp; - } - - return tx; + ec->tx_1 += -(ec->tx_1 >> DC_LOG2BETA) + tmp - ec->tx_2; + tmp1 = ec->tx_1 >> 15; + if (tmp1 > 32767) + tmp1 = 32767; + if (tmp1 < -32767) + tmp1 = -32767; + tx = tmp1; + ec->tx_2 = tmp; + } + + return tx; } + EXPORT_SYMBOL_GPL(oslec_hpf_tx); MODULE_LICENSE("GPL"); diff --git a/drivers/staging/echo/echo.h b/drivers/staging/echo/echo.h index 2a1d1d8..9fb9543 100644 --- a/drivers/staging/echo/echo.h +++ b/drivers/staging/echo/echo.h @@ -124,9 +124,8 @@ a minor burden. G.168 echo canceller descriptor. This defines the working state for a line echo canceller. */ -struct oslec_state -{ - int16_t tx,rx; +struct oslec_state { + int16_t tx, rx; int16_t clean; int16_t clean_nlp; @@ -170,4 +169,4 @@ struct oslec_state int16_t *snapshot; }; -#endif /* __ECHO_H */ +#endif /* __ECHO_H */ diff --git a/drivers/staging/echo/fir.h b/drivers/staging/echo/fir.h index e760471..5645cb1 100644 --- a/drivers/staging/echo/fir.h +++ b/drivers/staging/echo/fir.h @@ -72,8 +72,7 @@ 16 bit integer FIR descriptor. This defines the working state for a single instance of an FIR filter using 16 bit integer coefficients. */ -typedef struct -{ +typedef struct { int taps; int curr_pos; const int16_t *coeffs; @@ -85,8 +84,7 @@ typedef struct instance of an FIR filter using 32 bit integer coefficients, and filtering 16 bit integer data. */ -typedef struct -{ +typedef struct { int taps; int curr_pos; const int32_t *coeffs; @@ -97,39 +95,37 @@ typedef struct Floating point FIR descriptor. This defines the working state for a single instance of an FIR filter using floating point coefficients and data. */ -typedef struct -{ +typedef struct { int taps; int curr_pos; const float *coeffs; float *history; } fir_float_state_t; -static __inline__ const int16_t *fir16_create(fir16_state_t *fir, - const int16_t *coeffs, - int taps) +static __inline__ const int16_t *fir16_create(fir16_state_t * fir, + const int16_t * coeffs, int taps) { fir->taps = taps; fir->curr_pos = taps - 1; fir->coeffs = coeffs; #if defined(USE_MMX) || defined(USE_SSE2) || defined(__bfin__) - fir->history = kcalloc(2*taps, sizeof(int16_t), GFP_KERNEL); + fir->history = kcalloc(2 * taps, sizeof(int16_t), GFP_KERNEL); #else fir->history = kcalloc(taps, sizeof(int16_t), GFP_KERNEL); #endif return fir->history; } -static __inline__ void fir16_flush(fir16_state_t *fir) +static __inline__ void fir16_flush(fir16_state_t * fir) { #if defined(USE_MMX) || defined(USE_SSE2) || defined(__bfin__) - memset(fir->history, 0, 2*fir->taps*sizeof(int16_t)); + memset(fir->history, 0, 2 * fir->taps * sizeof(int16_t)); #else - memset(fir->history, 0, fir->taps*sizeof(int16_t)); + memset(fir->history, 0, fir->taps * sizeof(int16_t)); #endif } -static __inline__ void fir16_free(fir16_state_t *fir) +static __inline__ void fir16_free(fir16_state_t * fir) { kfree(fir->history); } @@ -137,166 +133,162 @@ static __inline__ void fir16_free(fir16_state_t *fir) #ifdef __bfin__ static inline int32_t dot_asm(short *x, short *y, int len) { - int dot; - - len--; - - __asm__ - ( - "I0 = %1;\n\t" - "I1 = %2;\n\t" - "A0 = 0;\n\t" - "R0.L = W[I0++] || R1.L = W[I1++];\n\t" - "LOOP dot%= LC0 = %3;\n\t" - "LOOP_BEGIN dot%=;\n\t" - "A0 += R0.L * R1.L (IS) || R0.L = W[I0++] || R1.L = W[I1++];\n\t" - "LOOP_END dot%=;\n\t" - "A0 += R0.L*R1.L (IS);\n\t" - "R0 = A0;\n\t" - "%0 = R0;\n\t" - : "=&d" (dot) - : "a" (x), "a" (y), "a" (len) - : "I0", "I1", "A1", "A0", "R0", "R1" - ); - - return dot; + int dot; + + len--; + + __asm__("I0 = %1;\n\t" + "I1 = %2;\n\t" + "A0 = 0;\n\t" + "R0.L = W[I0++] || R1.L = W[I1++];\n\t" + "LOOP dot%= LC0 = %3;\n\t" + "LOOP_BEGIN dot%=;\n\t" + "A0 += R0.L * R1.L (IS) || R0.L = W[I0++] || R1.L = W[I1++];\n\t" + "LOOP_END dot%=;\n\t" + "A0 += R0.L*R1.L (IS);\n\t" + "R0 = A0;\n\t" + "%0 = R0;\n\t" + :"=&d"(dot) + :"a"(x), "a"(y), "a"(len) + :"I0", "I1", "A1", "A0", "R0", "R1" + ); + + return dot; } #endif -static __inline__ int16_t fir16(fir16_state_t *fir, int16_t sample) +static __inline__ int16_t fir16(fir16_state_t * fir, int16_t sample) { - int32_t y; + int32_t y; #if defined(USE_MMX) - int i; - mmx_t *mmx_coeffs; - mmx_t *mmx_hist; - - fir->history[fir->curr_pos] = sample; - fir->history[fir->curr_pos + fir->taps] = sample; - - mmx_coeffs = (mmx_t *) fir->coeffs; - mmx_hist = (mmx_t *) &fir->history[fir->curr_pos]; - i = fir->taps; - pxor_r2r(mm4, mm4); - /* 8 samples per iteration, so the filter must be a multiple of 8 long. */ - while (i > 0) - { - movq_m2r(mmx_coeffs[0], mm0); - movq_m2r(mmx_coeffs[1], mm2); - movq_m2r(mmx_hist[0], mm1); - movq_m2r(mmx_hist[1], mm3); - mmx_coeffs += 2; - mmx_hist += 2; - pmaddwd_r2r(mm1, mm0); - pmaddwd_r2r(mm3, mm2); - paddd_r2r(mm0, mm4); - paddd_r2r(mm2, mm4); - i -= 8; - } - movq_r2r(mm4, mm0); - psrlq_i2r(32, mm0); - paddd_r2r(mm0, mm4); - movd_r2m(mm4, y); - emms(); + int i; + mmx_t *mmx_coeffs; + mmx_t *mmx_hist; + + fir->history[fir->curr_pos] = sample; + fir->history[fir->curr_pos + fir->taps] = sample; + + mmx_coeffs = (mmx_t *) fir->coeffs; + mmx_hist = (mmx_t *) & fir->history[fir->curr_pos]; + i = fir->taps; + pxor_r2r(mm4, mm4); + /* 8 samples per iteration, so the filter must be a multiple of 8 long. */ + while (i > 0) { + movq_m2r(mmx_coeffs[0], mm0); + movq_m2r(mmx_coeffs[1], mm2); + movq_m2r(mmx_hist[0], mm1); + movq_m2r(mmx_hist[1], mm3); + mmx_coeffs += 2; + mmx_hist += 2; + pmaddwd_r2r(mm1, mm0); + pmaddwd_r2r(mm3, mm2); + paddd_r2r(mm0, mm4); + paddd_r2r(mm2, mm4); + i -= 8; + } + movq_r2r(mm4, mm0); + psrlq_i2r(32, mm0); + paddd_r2r(mm0, mm4); + movd_r2m(mm4, y); + emms(); #elif defined(USE_SSE2) - int i; - xmm_t *xmm_coeffs; - xmm_t *xmm_hist; - - fir->history[fir->curr_pos] = sample; - fir->history[fir->curr_pos + fir->taps] = sample; - - xmm_coeffs = (xmm_t *) fir->coeffs; - xmm_hist = (xmm_t *) &fir->history[fir->curr_pos]; - i = fir->taps; - pxor_r2r(xmm4, xmm4); - /* 16 samples per iteration, so the filter must be a multiple of 16 long. */ - while (i > 0) - { - movdqu_m2r(xmm_coeffs[0], xmm0); - movdqu_m2r(xmm_coeffs[1], xmm2); - movdqu_m2r(xmm_hist[0], xmm1); - movdqu_m2r(xmm_hist[1], xmm3); - xmm_coeffs += 2; - xmm_hist += 2; - pmaddwd_r2r(xmm1, xmm0); - pmaddwd_r2r(xmm3, xmm2); - paddd_r2r(xmm0, xmm4); - paddd_r2r(xmm2, xmm4); - i -= 16; - } - movdqa_r2r(xmm4, xmm0); - psrldq_i2r(8, xmm0); - paddd_r2r(xmm0, xmm4); - movdqa_r2r(xmm4, xmm0); - psrldq_i2r(4, xmm0); - paddd_r2r(xmm0, xmm4); - movd_r2m(xmm4, y); + int i; + xmm_t *xmm_coeffs; + xmm_t *xmm_hist; + + fir->history[fir->curr_pos] = sample; + fir->history[fir->curr_pos + fir->taps] = sample; + + xmm_coeffs = (xmm_t *) fir->coeffs; + xmm_hist = (xmm_t *) & fir->history[fir->curr_pos]; + i = fir->taps; + pxor_r2r(xmm4, xmm4); + /* 16 samples per iteration, so the filter must be a multiple of 16 long. */ + while (i > 0) { + movdqu_m2r(xmm_coeffs[0], xmm0); + movdqu_m2r(xmm_coeffs[1], xmm2); + movdqu_m2r(xmm_hist[0], xmm1); + movdqu_m2r(xmm_hist[1], xmm3); + xmm_coeffs += 2; + xmm_hist += 2; + pmaddwd_r2r(xmm1, xmm0); + pmaddwd_r2r(xmm3, xmm2); + paddd_r2r(xmm0, xmm4); + paddd_r2r(xmm2, xmm4); + i -= 16; + } + movdqa_r2r(xmm4, xmm0); + psrldq_i2r(8, xmm0); + paddd_r2r(xmm0, xmm4); + movdqa_r2r(xmm4, xmm0); + psrldq_i2r(4, xmm0); + paddd_r2r(xmm0, xmm4); + movd_r2m(xmm4, y); #elif defined(__bfin__) - fir->history[fir->curr_pos] = sample; - fir->history[fir->curr_pos + fir->taps] = sample; - y = dot_asm((int16_t*)fir->coeffs, &fir->history[fir->curr_pos], fir->taps); + fir->history[fir->curr_pos] = sample; + fir->history[fir->curr_pos + fir->taps] = sample; + y = dot_asm((int16_t *) fir->coeffs, &fir->history[fir->curr_pos], + fir->taps); #else - int i; - int offset1; - int offset2; - - fir->history[fir->curr_pos] = sample; - - offset2 = fir->curr_pos; - offset1 = fir->taps - offset2; - y = 0; - for (i = fir->taps - 1; i >= offset1; i--) - y += fir->coeffs[i]*fir->history[i - offset1]; - for ( ; i >= 0; i--) - y += fir->coeffs[i]*fir->history[i + offset2]; + int i; + int offset1; + int offset2; + + fir->history[fir->curr_pos] = sample; + + offset2 = fir->curr_pos; + offset1 = fir->taps - offset2; + y = 0; + for (i = fir->taps - 1; i >= offset1; i--) + y += fir->coeffs[i] * fir->history[i - offset1]; + for (; i >= 0; i--) + y += fir->coeffs[i] * fir->history[i + offset2]; #endif - if (fir->curr_pos <= 0) - fir->curr_pos = fir->taps; - fir->curr_pos--; - return (int16_t) (y >> 15); + if (fir->curr_pos <= 0) + fir->curr_pos = fir->taps; + fir->curr_pos--; + return (int16_t) (y >> 15); } -static __inline__ const int16_t *fir32_create(fir32_state_t *fir, - const int32_t *coeffs, - int taps) +static __inline__ const int16_t *fir32_create(fir32_state_t * fir, + const int32_t * coeffs, int taps) { - fir->taps = taps; - fir->curr_pos = taps - 1; - fir->coeffs = coeffs; - fir->history = kcalloc(taps, sizeof(int16_t), GFP_KERNEL); - return fir->history; + fir->taps = taps; + fir->curr_pos = taps - 1; + fir->coeffs = coeffs; + fir->history = kcalloc(taps, sizeof(int16_t), GFP_KERNEL); + return fir->history; } -static __inline__ void fir32_flush(fir32_state_t *fir) +static __inline__ void fir32_flush(fir32_state_t * fir) { - memset(fir->history, 0, fir->taps*sizeof(int16_t)); + memset(fir->history, 0, fir->taps * sizeof(int16_t)); } -static __inline__ void fir32_free(fir32_state_t *fir) +static __inline__ void fir32_free(fir32_state_t * fir) { - kfree(fir->history); + kfree(fir->history); } -static __inline__ int16_t fir32(fir32_state_t *fir, int16_t sample) +static __inline__ int16_t fir32(fir32_state_t * fir, int16_t sample) { - int i; - int32_t y; - int offset1; - int offset2; - - fir->history[fir->curr_pos] = sample; - offset2 = fir->curr_pos; - offset1 = fir->taps - offset2; - y = 0; - for (i = fir->taps - 1; i >= offset1; i--) - y += fir->coeffs[i]*fir->history[i - offset1]; - for ( ; i >= 0; i--) - y += fir->coeffs[i]*fir->history[i + offset2]; - if (fir->curr_pos <= 0) - fir->curr_pos = fir->taps; - fir->curr_pos--; - return (int16_t) (y >> 15); + int i; + int32_t y; + int offset1; + int offset2; + + fir->history[fir->curr_pos] = sample; + offset2 = fir->curr_pos; + offset1 = fir->taps - offset2; + y = 0; + for (i = fir->taps - 1; i >= offset1; i--) + y += fir->coeffs[i] * fir->history[i - offset1]; + for (; i >= 0; i--) + y += fir->coeffs[i] * fir->history[i + offset2]; + if (fir->curr_pos <= 0) + fir->curr_pos = fir->taps; + fir->curr_pos--; + return (int16_t) (y >> 15); } #endif diff --git a/drivers/staging/echo/mmx.h b/drivers/staging/echo/mmx.h index b5a3964..35412ef 100644 --- a/drivers/staging/echo/mmx.h +++ b/drivers/staging/echo/mmx.h @@ -27,24 +27,23 @@ * values by ULL, lest they be truncated by the compiler) */ -typedef union { - long long q; /* Quadword (64-bit) value */ - unsigned long long uq; /* Unsigned Quadword */ - int d[2]; /* 2 Doubleword (32-bit) values */ - unsigned int ud[2]; /* 2 Unsigned Doubleword */ - short w[4]; /* 4 Word (16-bit) values */ - unsigned short uw[4]; /* 4 Unsigned Word */ - char b[8]; /* 8 Byte (8-bit) values */ - unsigned char ub[8]; /* 8 Unsigned Byte */ - float s[2]; /* Single-precision (32-bit) value */ -} mmx_t; /* On an 8-byte (64-bit) boundary */ +typedef union { + long long q; /* Quadword (64-bit) value */ + unsigned long long uq; /* Unsigned Quadword */ + int d[2]; /* 2 Doubleword (32-bit) values */ + unsigned int ud[2]; /* 2 Unsigned Doubleword */ + short w[4]; /* 4 Word (16-bit) values */ + unsigned short uw[4]; /* 4 Unsigned Word */ + char b[8]; /* 8 Byte (8-bit) values */ + unsigned char ub[8]; /* 8 Unsigned Byte */ + float s[2]; /* Single-precision (32-bit) value */ +} mmx_t; /* On an 8-byte (64-bit) boundary */ /* SSE registers */ typedef union { char b[16]; } xmm_t; - #define mmx_i2r(op,imm,reg) \ __asm__ __volatile__ (#op " %0, %%" #reg \ : /* nothing */ \ @@ -63,7 +62,6 @@ typedef union { #define mmx_r2r(op,regs,regd) \ __asm__ __volatile__ (#op " %" #regs ", %" #regd) - #define emms() __asm__ __volatile__ ("emms") #define movd_m2r(var,reg) mmx_m2r (movd, var, reg) @@ -192,16 +190,13 @@ typedef union { #define pxor_m2r(var,reg) mmx_m2r (pxor, var, reg) #define pxor_r2r(regs,regd) mmx_r2r (pxor, regs, regd) - /* 3DNOW extensions */ #define pavgusb_m2r(var,reg) mmx_m2r (pavgusb, var, reg) #define pavgusb_r2r(regs,regd) mmx_r2r (pavgusb, regs, regd) - /* AMD MMX extensions - also available in intel SSE */ - #define mmx_m2ri(op,mem,reg,imm) \ __asm__ __volatile__ (#op " %1, %0, %%" #reg \ : /* nothing */ \ @@ -216,7 +211,6 @@ typedef union { : /* nothing */ \ : "m" (mem)) - #define maskmovq(regs,maskreg) mmx_r2ri (maskmovq, regs, maskreg) #define movntq_r2m(mmreg,var) mmx_r2m (movntq, mmreg, var) @@ -284,5 +278,4 @@ typedef union { #define punpcklqdq_r2r(regs,regd) mmx_r2r (punpcklqdq, regs, regd) #define punpckhqdq_r2r(regs,regd) mmx_r2r (punpckhqdq, regs, regd) - #endif /* AVCODEC_I386MMX_H */ diff --git a/drivers/staging/echo/oslec.h b/drivers/staging/echo/oslec.h index e41de61..bad8523 100644 --- a/drivers/staging/echo/oslec.h +++ b/drivers/staging/echo/oslec.h @@ -83,4 +83,4 @@ int16_t oslec_update(struct oslec_state *ec, int16_t tx, int16_t rx); */ int16_t oslec_hpf_tx(struct oslec_state *ec, int16_t tx); -#endif /* __OSLEC_H */ +#endif /* __OSLEC_H */ |