diff options
Diffstat (limited to 'volk/spu_lib/spu_32fc_pointwise_multiply_unaligned.c')
| -rw-r--r-- | volk/spu_lib/spu_32fc_pointwise_multiply_unaligned.c | 94 |
1 files changed, 47 insertions, 47 deletions
diff --git a/volk/spu_lib/spu_32fc_pointwise_multiply_unaligned.c b/volk/spu_lib/spu_32fc_pointwise_multiply_unaligned.c index 58fd4aa0c..d1c960488 100644 --- a/volk/spu_lib/spu_32fc_pointwise_multiply_unaligned.c +++ b/volk/spu_lib/spu_32fc_pointwise_multiply_unaligned.c @@ -7,15 +7,15 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi //loop iterator i int i = 0; void* retval = target; - + //put the target and source addresses into qwords vector unsigned int address_counter_tgt = {(unsigned int)target, 0, 0, 0}; vector unsigned int address_counter_src0 = {(unsigned int)src0, 0, 0 ,0}; vector unsigned int address_counter_src1 = {(unsigned int)src1, 0, 0, 0}; - + //create shuffle masks - + //shuffle mask building blocks: //all from the first vector vector unsigned char oneup = {0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, @@ -23,9 +23,9 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi //all from the second vector vector unsigned char second_oneup = {0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f}; - - + + //gamma: second half of the second, first half of the first, break at (unsigned int)src0%16 vector unsigned char src_cmp = spu_splats((unsigned char)((unsigned int)src0%16)); vector unsigned char gt_res = spu_cmpgt(oneup, src_cmp); @@ -33,7 +33,7 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi vector unsigned char cmp_res = spu_or(gt_res, eq_res); vector unsigned char sixteen_uchar = spu_splats((unsigned char)16); vector unsigned char phase_change = spu_and(sixteen_uchar, cmp_res); - vector unsigned int shuffle_mask_gamma = spu_add((vector unsigned int)phase_change, + vector unsigned int shuffle_mask_gamma = spu_add((vector unsigned int)phase_change, (vector unsigned int)oneup); shuffle_mask_gamma = spu_rlqwbyte(shuffle_mask_gamma, (unsigned int)src0%16); @@ -44,17 +44,17 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi cmp_res = spu_or(gt_res, eq_res); sixteen_uchar = spu_splats((unsigned char)16); phase_change = spu_and(sixteen_uchar, cmp_res); - vector unsigned int shuffle_mask_eta = spu_add((vector unsigned int)phase_change, + vector unsigned int shuffle_mask_eta = spu_add((vector unsigned int)phase_change, (vector unsigned int)oneup); shuffle_mask_eta = spu_rlqwbyte(shuffle_mask_eta, (unsigned int)src1%16); - - - + + + vector unsigned char tgt_second = spu_rlqwbyte(second_oneup, -((unsigned int)target%16)); vector unsigned char tgt_first = spu_rlqwbyte(oneup, -((unsigned int)target%16)); - + //alpha: first half of first, second half of second, break at (unsigned int)target%16 src_cmp = spu_splats((unsigned char)((unsigned int)target%16)); gt_res = spu_cmpgt(oneup, src_cmp); @@ -63,13 +63,13 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi phase_change = spu_and(sixteen_uchar, cmp_res); vector unsigned int shuffle_mask_alpha = spu_add((vector unsigned int)phase_change, (vector unsigned int)oneup); - + //delta: first half of first, first half of second, break at (unsigned int)target%16 vector unsigned char shuffle_mask_delta = spu_shuffle(oneup, tgt_second, (vector unsigned char)shuffle_mask_alpha); //epsilon: second half of second, second half of first, break at (unsigned int)target%16 vector unsigned char shuffle_mask_epsilon = spu_shuffle(tgt_second, oneup, (vector unsigned char)shuffle_mask_alpha); //zeta: second half of second, first half of first, break at 16 - (unsigned int)target%16 - vector unsigned int shuffle_mask_zeta = spu_rlqwbyte(shuffle_mask_alpha, (unsigned int)target%16); + vector unsigned int shuffle_mask_zeta = spu_rlqwbyte(shuffle_mask_alpha, (unsigned int)target%16); //beta: first half of first, second half of second, break at num_bytes%16 src_cmp = spu_splats((unsigned char)(num_bytes%16)); @@ -79,19 +79,19 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi phase_change = spu_and(sixteen_uchar, cmp_res); vector unsigned int shuffle_mask_beta = spu_add((vector unsigned int)phase_change, (vector unsigned int)oneup); - - - - + + + + qword src0_past; qword src0_present; qword src1_past; qword src1_present; qword tgt_past; qword tgt_present; - + qword in_temp0; qword in_temp1; qword out_temp0; @@ -101,7 +101,7 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi src0_past = si_lqd((qword)address_counter_src0, 0); src1_past = si_lqd((qword)address_counter_src1, 0); tgt_past = si_lqd((qword)address_counter_tgt, 0); - + vector unsigned char shuffle_mask_complexprod0 = {0x04, 0x05, 0x06, 0x07, 0x00, 0x01, 0x02, 0x03, 0x0c, 0x0d, 0x0e, 0x0f, 0x08, 0x09, 0x0a, 0x0b}; vector unsigned char shuffle_mask_complexprod1 = {0x00, 0x01, 0x02, 0x03, 0x10, 0x11, 0x12, 0x13, @@ -110,7 +110,7 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi 0x0c, 0x0d, 0x0e, 0x0f, 0x1c, 0x1d, 0x1e, 0x1f}; vector unsigned char sign_changer = {0x00, 0x00, 0x00, 0x00, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, 0x00, 0x00}; - + vector float prod0; qword shuf0; vector float prod1; @@ -118,54 +118,54 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi qword summand0; qword summand1; vector float sum; - + for(i = 0; i < num_bytes/16; ++i) { - + src0_present = si_lqd((qword)address_counter_src0, 16); src1_present = si_lqd((qword)address_counter_src1, 16); tgt_present = si_lqd((qword)address_counter_tgt, 16); - + in_temp0 = spu_shuffle(src0_present, src0_past, (vector unsigned char)shuffle_mask_gamma); in_temp1 = spu_shuffle(src1_present, src1_past, (vector unsigned char)shuffle_mask_eta); - + prod0 = spu_mul((vector float)in_temp0, (vector float)in_temp1); shuf0 = spu_shuffle((qword)in_temp1, (qword)in_temp1, shuffle_mask_complexprod0); prod1 = spu_mul((vector float)in_temp0, (vector float)shuf0); sign_change = spu_xor(prod0, (vector float)sign_changer); - + summand0 = spu_shuffle((qword)sign_change, (qword)prod1, shuffle_mask_complexprod1); - + summand1 = spu_shuffle((qword)sign_change, (qword)prod1, shuffle_mask_complexprod2); - + sum = spu_add((vector float)summand0, (vector float)summand1); - + out_temp0 = spu_shuffle(tgt_past, (qword)sum, shuffle_mask_delta); out_temp1 = spu_shuffle(tgt_present, (qword)sum, shuffle_mask_epsilon); - + si_stqd(out_temp0, (qword)address_counter_tgt, 0); si_stqd(out_temp1, (qword)address_counter_tgt, 16); - + tgt_past = out_temp1; src0_past = src0_present; src1_past = src1_present; address_counter_src0 = spu_add(address_counter_src0, 16); address_counter_src1 = spu_add(address_counter_src1, 16); address_counter_tgt = spu_add(address_counter_tgt, 16); - - + + } - + src0_present = si_lqd((qword)address_counter_src0, 16); src1_present = si_lqd((qword)address_counter_src1, 16); tgt_present = si_lqd((qword)address_counter_tgt, 16); - - + + in_temp0 = spu_shuffle(src0_present, src0_past, (vector unsigned char) shuffle_mask_gamma); in_temp1 = spu_shuffle(src1_present, src1_past, (vector unsigned char) shuffle_mask_eta); - - + + prod0 = spu_mul((vector float)in_temp0, (vector float)in_temp1); shuf0 = spu_shuffle((qword)in_temp1, (qword)in_temp1, shuffle_mask_complexprod0); prod1 = spu_mul(prod0, (vector float)shuf0); @@ -173,20 +173,20 @@ void* libvector_pointwise_multiply_32fc_unaligned(void* target, void* src0, voi summand0 = spu_shuffle((qword)sign_change, (qword)prod1, shuffle_mask_complexprod1); summand1 = spu_shuffle((qword)sign_change, (qword)prod1, shuffle_mask_complexprod2); sum = spu_add((vector float)summand0, (vector float)summand1); - - - + + + qword target_temp = spu_shuffle(tgt_present, tgt_past, (vector unsigned char) shuffle_mask_zeta); qword meld = spu_shuffle((qword)sum, target_temp, (vector unsigned char)shuffle_mask_beta); - - - + + + out_temp0 = spu_shuffle(tgt_past, meld, shuffle_mask_delta); out_temp1 = spu_shuffle(tgt_present, meld, shuffle_mask_epsilon); - + si_stqd(out_temp0, (qword)address_counter_tgt, 0); si_stqd(out_temp1, (qword)address_counter_tgt, 16); - + return retval; } @@ -209,14 +209,14 @@ int main(){ vector_product_complex(res, pooh, bear, 48*sizeof(float)); - + for(i = 0; i < 48; ++i) { printf("%f, ", res[i]); } printf("\n"); - + } */ |