diff --git a/src/math/erfc_s390x.s b/src/math/erfc_s390x.s index 1a867ad9f4..57710b254b 100644 --- a/src/math/erfc_s390x.s +++ b/src/math/erfc_s390x.s @@ -4,7 +4,8 @@ #include "textflag.h" -#define NegInf 0xFFF0000000000000 +#define Neg2p11 0xC000E147AE147AE1 +#define Pos15 0x402E // Minimax polynomial coefficients and other constants DATA ·erfcrodataL38<> + 0(SB)/8, $.234875460637085087E-01 @@ -136,21 +137,24 @@ GLOBL ·erfctab2069<> + 0(SB), RODATA, $128 // Erfc(NaN) = NaN // The algorithm used is minimax polynomial approximation // with coefficients determined with a Remez exchange algorithm. +// This assembly implementation handles inputs in the range [-2.11, +15]. +// For all other inputs we call the generic Go implementation. -TEXT ·erfcAsm(SB), NOSPLIT, $0-16 - //special case Erfc(+Inf) = 0 +TEXT ·erfcAsm(SB), NOSPLIT|NOFRAME, $0-16 MOVD x+0(FP), R1 - MOVD $NegInf, R2 - CMPUBEQ R1, R2, erfcIsPosInf + MOVD $Neg2p11, R2 + CMPUBGT R1, R2, usego FMOVD x+0(FP), F0 MOVD $·erfcrodataL38<>+0(SB), R9 - WORD $0xB3CD0010 //lgdr %r1, %f0 FMOVD F0, F2 SRAD $48, R1 - MOVH $0x3FFF, R3 MOVH R1, R2 ANDW $0x7FFF, R1 + MOVH $Pos15, R3 + CMPW R1, R3 + BGT usego + MOVH $0x3FFF, R3 MOVW R1, R6 MOVW R3, R7 CMPBGT R6, R7, L2 @@ -523,8 +527,5 @@ L18: L25: FMOVD 568(R9), F2 BR L1 -erfcIsPosInf: - FMOVD $(2.0), F1 - FMOVD F1, ret+8(FP) - RET - +usego: + BR ·erfc(SB)