Temporarily revert to the old NRM2 kernels for ThunderX2/3 and NeoverseN1tags/v0.3.14^2
| @@ -91,10 +91,10 @@ IDAMAXKERNEL = iamax_thunderx2t99.c | |||||
| ICAMAXKERNEL = izamax_thunderx2t99.c | ICAMAXKERNEL = izamax_thunderx2t99.c | ||||
| IZAMAXKERNEL = izamax_thunderx2t99.c | IZAMAXKERNEL = izamax_thunderx2t99.c | ||||
| SNRM2KERNEL = scnrm2_thunderx2t99.c | |||||
| DNRM2KERNEL = dznrm2_thunderx2t99.c | |||||
| CNRM2KERNEL = scnrm2_thunderx2t99.c | |||||
| ZNRM2KERNEL = dznrm2_thunderx2t99.c | |||||
| SNRM2KERNEL = nrm2.S | |||||
| DNRM2KERNEL = nrm2.S | |||||
| CNRM2KERNEL = znrm2.S | |||||
| ZNRM2KERNEL = znrm2.S | |||||
| DDOTKERNEL = dot_thunderx2t99.c | DDOTKERNEL = dot_thunderx2t99.c | ||||
| SDOTKERNEL = dot_thunderx2t99.c | SDOTKERNEL = dot_thunderx2t99.c | ||||
| @@ -153,12 +153,12 @@ IDAMAXKERNEL = iamax_thunderx2t99.c | |||||
| ICAMAXKERNEL = izamax_thunderx2t99.c | ICAMAXKERNEL = izamax_thunderx2t99.c | ||||
| IZAMAXKERNEL = izamax_thunderx2t99.c | IZAMAXKERNEL = izamax_thunderx2t99.c | ||||
| SNRM2KERNEL = scnrm2_thunderx2t99.c | |||||
| CNRM2KERNEL = scnrm2_thunderx2t99.c | |||||
| SNRM2KERNEL = nrm2.S | |||||
| CNRM2KERNEL = nrm2.S | |||||
| #DNRM2KERNEL = dznrm2_thunderx2t99_fast.c | #DNRM2KERNEL = dznrm2_thunderx2t99_fast.c | ||||
| #ZNRM2KERNEL = dznrm2_thunderx2t99_fast.c | #ZNRM2KERNEL = dznrm2_thunderx2t99_fast.c | ||||
| DNRM2KERNEL = dznrm2_thunderx2t99.c | |||||
| ZNRM2KERNEL = dznrm2_thunderx2t99.c | |||||
| DNRM2KERNEL = znrm2.S | |||||
| ZNRM2KERNEL = znrm2.S | |||||
| DDOTKERNEL = dot_thunderx2t99.c | DDOTKERNEL = dot_thunderx2t99.c | ||||
| @@ -153,13 +153,16 @@ IDAMAXKERNEL = iamax_thunderx2t99.c | |||||
| ICAMAXKERNEL = izamax_thunderx2t99.c | ICAMAXKERNEL = izamax_thunderx2t99.c | ||||
| IZAMAXKERNEL = izamax_thunderx2t99.c | IZAMAXKERNEL = izamax_thunderx2t99.c | ||||
| SNRM2KERNEL = scnrm2_thunderx2t99.c | |||||
| CNRM2KERNEL = scnrm2_thunderx2t99.c | |||||
| #DNRM2KERNEL = dznrm2_thunderx2t99_fast.c | |||||
| #ZNRM2KERNEL = dznrm2_thunderx2t99_fast.c | |||||
| DNRM2KERNEL = dznrm2_thunderx2t99.c | |||||
| ZNRM2KERNEL = dznrm2_thunderx2t99.c | |||||
| #SNRM2KERNEL = scnrm2_thunderx2t99.c | |||||
| #CNRM2KERNEL = scnrm2_thunderx2t99.c | |||||
| ##DNRM2KERNEL = dznrm2_thunderx2t99_fast.c | |||||
| ##ZNRM2KERNEL = dznrm2_thunderx2t99_fast.c | |||||
| #DNRM2KERNEL = dznrm2_thunderx2t99.c | |||||
| #ZNRM2KERNEL = dznrm2_thunderx2t99.c | |||||
| SNRM2KERNEL = nrm2.S | |||||
| DNRM2KERNEL = nrm2.S | |||||
| CNRM2KERNEL = znrm2.S | |||||
| ZNRM2KERNEL = znrm2.S | |||||
| DDOTKERNEL = dot_thunderx2t99.c | DDOTKERNEL = dot_thunderx2t99.c | ||||
| SDOTKERNEL = dot_thunderx2t99.c | SDOTKERNEL = dot_thunderx2t99.c | ||||
| @@ -13,7 +13,7 @@ static void srot_kernel(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT c, FLOAT s) | |||||
| { | { | ||||
| BLASLONG i = 0; | BLASLONG i = 0; | ||||
| #if V_SIMD && (defined(HAVE_FMA3) || V_SIMD > 128) | |||||
| #if V_SIMD && !defined(C_PGI) && (defined(HAVE_FMA3) || V_SIMD > 128) | |||||
| const int vstep = v_nlanes_f32; | const int vstep = v_nlanes_f32; | ||||
| const int unrollx4 = n & (-vstep * 4); | const int unrollx4 = n & (-vstep * 4); | ||||
| const int unrollx = n & -vstep; | const int unrollx = n & -vstep; | ||||