diff options
author | Frank Chang <frank.chang@sifive.com> | 2021-12-10 15:56:56 +0800 |
---|---|---|
committer | Alistair Francis <alistair.francis@wdc.com> | 2021-12-20 14:53:31 +1000 |
commit | 55c35407c3706148761373aaa1d3350b57e86e8d (patch) | |
tree | 8b7ccb071c8f7fe1064e9830f3df13aa66ecfbb7 /target/riscv/vector_helper.c | |
parent | e848a1e5632518647ac146d75b2fe006050ffb82 (diff) |
target/riscv: rvv-1.0: floating-point reciprocal estimate instruction
Implement the floating-point reciprocal estimate to 7 bits instruction.
Signed-off-by: Frank Chang <frank.chang@sifive.com>
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20211210075704.23951-71-frank.chang@sifive.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Diffstat (limited to 'target/riscv/vector_helper.c')
-rw-r--r-- | target/riscv/vector_helper.c | 191 |
1 files changed, 191 insertions, 0 deletions
diff --git a/target/riscv/vector_helper.c b/target/riscv/vector_helper.c index d5f3229bcb..946dca53ff 100644 --- a/target/riscv/vector_helper.c +++ b/target/riscv/vector_helper.c @@ -3587,6 +3587,197 @@ GEN_VEXT_V_ENV(vfrsqrt7_v_h, 2, 2) GEN_VEXT_V_ENV(vfrsqrt7_v_w, 4, 4) GEN_VEXT_V_ENV(vfrsqrt7_v_d, 8, 8) +/* + * Vector Floating-Point Reciprocal Estimate Instruction + * + * Adapted from riscv-v-spec recip.c: + * https://github.com/riscv/riscv-v-spec/blob/master/recip.c + */ +static uint64_t frec7(uint64_t f, int exp_size, int frac_size, + float_status *s) +{ + uint64_t sign = extract64(f, frac_size + exp_size, 1); + uint64_t exp = extract64(f, frac_size, exp_size); + uint64_t frac = extract64(f, 0, frac_size); + + const uint8_t lookup_table[] = { + 127, 125, 123, 121, 119, 117, 116, 114, + 112, 110, 109, 107, 105, 104, 102, 100, + 99, 97, 96, 94, 93, 91, 90, 88, + 87, 85, 84, 83, 81, 80, 79, 77, + 76, 75, 74, 72, 71, 70, 69, 68, + 66, 65, 64, 63, 62, 61, 60, 59, + 58, 57, 56, 55, 54, 53, 52, 51, + 50, 49, 48, 47, 46, 45, 44, 43, + 42, 41, 40, 40, 39, 38, 37, 36, + 35, 35, 34, 33, 32, 31, 31, 30, + 29, 28, 28, 27, 26, 25, 25, 24, + 23, 23, 22, 21, 21, 20, 19, 19, + 18, 17, 17, 16, 15, 15, 14, 14, + 13, 12, 12, 11, 11, 10, 9, 9, + 8, 8, 7, 7, 6, 5, 5, 4, + 4, 3, 3, 2, 2, 1, 1, 0 + }; + const int precision = 7; + + if (exp == 0 && frac != 0) { /* subnormal */ + /* Normalize the subnormal. */ + while (extract64(frac, frac_size - 1, 1) == 0) { + exp--; + frac <<= 1; + } + + frac = (frac << 1) & MAKE_64BIT_MASK(0, frac_size); + + if (exp != 0 && exp != UINT64_MAX) { + /* + * Overflow to inf or max value of same sign, + * depending on sign and rounding mode. + */ + s->float_exception_flags |= (float_flag_inexact | + float_flag_overflow); + + if ((s->float_rounding_mode == float_round_to_zero) || + ((s->float_rounding_mode == float_round_down) && !sign) || + ((s->float_rounding_mode == float_round_up) && sign)) { + /* Return greatest/negative finite value. */ + return (sign << (exp_size + frac_size)) | + (MAKE_64BIT_MASK(frac_size, exp_size) - 1); + } else { + /* Return +-inf. */ + return (sign << (exp_size + frac_size)) | + MAKE_64BIT_MASK(frac_size, exp_size); + } + } + } + + int idx = frac >> (frac_size - precision); + uint64_t out_frac = (uint64_t)(lookup_table[idx]) << + (frac_size - precision); + uint64_t out_exp = 2 * MAKE_64BIT_MASK(0, exp_size - 1) + ~exp; + + if (out_exp == 0 || out_exp == UINT64_MAX) { + /* + * The result is subnormal, but don't raise the underflow exception, + * because there's no additional loss of precision. + */ + out_frac = (out_frac >> 1) | MAKE_64BIT_MASK(frac_size - 1, 1); + if (out_exp == UINT64_MAX) { + out_frac >>= 1; + out_exp = 0; + } + } + + uint64_t val = 0; + val = deposit64(val, 0, frac_size, out_frac); + val = deposit64(val, frac_size, exp_size, out_exp); + val = deposit64(val, frac_size + exp_size, 1, sign); + return val; +} + +static float16 frec7_h(float16 f, float_status *s) +{ + int exp_size = 5, frac_size = 10; + bool sign = float16_is_neg(f); + + /* frec7(+-inf) = +-0 */ + if (float16_is_infinity(f)) { + return float16_set_sign(float16_zero, sign); + } + + /* frec7(+-0) = +-inf */ + if (float16_is_zero(f)) { + s->float_exception_flags |= float_flag_divbyzero; + return float16_set_sign(float16_infinity, sign); + } + + /* frec7(sNaN) = canonical NaN */ + if (float16_is_signaling_nan(f, s)) { + s->float_exception_flags |= float_flag_invalid; + return float16_default_nan(s); + } + + /* frec7(qNaN) = canonical NaN */ + if (float16_is_quiet_nan(f, s)) { + return float16_default_nan(s); + } + + /* +-normal, +-subnormal */ + uint64_t val = frec7(f, exp_size, frac_size, s); + return make_float16(val); +} + +static float32 frec7_s(float32 f, float_status *s) +{ + int exp_size = 8, frac_size = 23; + bool sign = float32_is_neg(f); + + /* frec7(+-inf) = +-0 */ + if (float32_is_infinity(f)) { + return float32_set_sign(float32_zero, sign); + } + + /* frec7(+-0) = +-inf */ + if (float32_is_zero(f)) { + s->float_exception_flags |= float_flag_divbyzero; + return float32_set_sign(float32_infinity, sign); + } + + /* frec7(sNaN) = canonical NaN */ + if (float32_is_signaling_nan(f, s)) { + s->float_exception_flags |= float_flag_invalid; + return float32_default_nan(s); + } + + /* frec7(qNaN) = canonical NaN */ + if (float32_is_quiet_nan(f, s)) { + return float32_default_nan(s); + } + + /* +-normal, +-subnormal */ + uint64_t val = frec7(f, exp_size, frac_size, s); + return make_float32(val); +} + +static float64 frec7_d(float64 f, float_status *s) +{ + int exp_size = 11, frac_size = 52; + bool sign = float64_is_neg(f); + + /* frec7(+-inf) = +-0 */ + if (float64_is_infinity(f)) { + return float64_set_sign(float64_zero, sign); + } + + /* frec7(+-0) = +-inf */ + if (float64_is_zero(f)) { + s->float_exception_flags |= float_flag_divbyzero; + return float64_set_sign(float64_infinity, sign); + } + + /* frec7(sNaN) = canonical NaN */ + if (float64_is_signaling_nan(f, s)) { + s->float_exception_flags |= float_flag_invalid; + return float64_default_nan(s); + } + + /* frec7(qNaN) = canonical NaN */ + if (float64_is_quiet_nan(f, s)) { + return float64_default_nan(s); + } + + /* +-normal, +-subnormal */ + uint64_t val = frec7(f, exp_size, frac_size, s); + return make_float64(val); +} + +RVVCALL(OPFVV1, vfrec7_v_h, OP_UU_H, H2, H2, frec7_h) +RVVCALL(OPFVV1, vfrec7_v_w, OP_UU_W, H4, H4, frec7_s) +RVVCALL(OPFVV1, vfrec7_v_d, OP_UU_D, H8, H8, frec7_d) +GEN_VEXT_V_ENV(vfrec7_v_h, 2, 2) +GEN_VEXT_V_ENV(vfrec7_v_w, 4, 4) +GEN_VEXT_V_ENV(vfrec7_v_d, 8, 8) + /* Vector Floating-Point MIN/MAX Instructions */ RVVCALL(OPFVV2, vfmin_vv_h, OP_UUU_H, H2, H2, H2, float16_minimum_number) RVVCALL(OPFVV2, vfmin_vv_w, OP_UUU_W, H4, H4, H4, float32_minimum_number) |