From bae74203755e7cec96d84dc749a572d8f374ed39 Mon Sep 17 00:00:00 2001 From: kms1212 Date: Fri, 25 Oct 2024 08:53:30 +0900 Subject: m68k: add FPU instructions `FETOX`, `FETOXM1`, `FTWOTOX`, `FTENTOX` (#12903) --- 3rdparty/softfloat3/bochs_ext/f2xm1.c | 343 +++++++++++++++++++++++-- 3rdparty/softfloat3/bochs_ext/softfloat3_ext.h | 3 + src/devices/cpu/m68000/m68kfpu.cpp | 36 +++ 3 files changed, 363 insertions(+), 19 deletions(-) diff --git a/3rdparty/softfloat3/bochs_ext/f2xm1.c b/3rdparty/softfloat3/bochs_ext/f2xm1.c index a8201937660..bc8544edd63 100644 --- a/3rdparty/softfloat3/bochs_ext/f2xm1.c +++ b/3rdparty/softfloat3/bochs_ext/f2xm1.c @@ -25,27 +25,40 @@ these four paragraphs for those parts of this code that are retained. #define FLOAT128 +#include "../build/MAME/platform.h" + #include "../source/include/softfloat.h" #include "../source/include/internals.h" #include "softfloat-extra.h" #include "softfloat-helpers.h" #include "softfloat-specialize.h" +static const extFloat80_t floatx80_posone = packFloatx80(0, 0x3fff, uint64_t(0x8000000000000000)); static const extFloat80_t floatx80_negone = packFloatx80(1, 0x3fff, uint64_t(0x8000000000000000)); static const extFloat80_t floatx80_neghalf = packFloatx80(1, 0x3ffe, uint64_t(0x8000000000000000)); +static const float128_t float128_e = + packFloat128(uint64_t(0x40005BF0A8B14576), uint64_t(0x95355FB8AC404E7A)); static const float128_t float128_ln2 = packFloat128(uint64_t(0x3ffe62e42fefa39e), uint64_t(0xf35793c7673007e6)); +static const float128_t float128_ln10 = + packFloat128(uint64_t(0x400026bb1bbb5551), uint64_t(0x582dd4adac5705a6)); #ifdef BETTER_THAN_PENTIUM #define LN2_SIG_HI uint64_t(0xb17217f7d1cf79ab) #define LN2_SIG_LO uint64_t(0xc9e3b39800000000) /* 96 bit precision */ +#define LN10_SIG_HI uint64_t(0xae585ba95bb0a2aa) +#define LN10_SIG_LO uint64_t(0x763776d400000000) /* 96 bit precision */ + #else #define LN2_SIG_HI uint64_t(0xb17217f7d1cf79ab) #define LN2_SIG_LO uint64_t(0xc000000000000000) /* 67-bit precision */ +#define LN10_SIG_HI uint64_t(0xae585ba95bb0a2aa) +#define LN10_SIG_LO uint64_t(0x7000000000000000) /* 67-bit precision */ + #endif #define EXP_ARR_SIZE 15 @@ -75,25 +88,25 @@ extern float128_t EvalPoly(float128_t x, const float128_t *arr, int n); static float128_t poly_exp(float128_t x) { /* - // 2 3 4 5 6 7 8 9 - // x x x x x x x x x - // e - 1 ~ x + --- + --- + --- + --- + --- + --- + --- + --- + ... - // 2! 3! 4! 5! 6! 7! 8! 9! - // - // 2 3 4 5 6 7 8 - // x x x x x x x x - // = x [ 1 + --- + --- + --- + --- + --- + --- + --- + --- + ... ] - // 2! 3! 4! 5! 6! 7! 8! 9! - // - // 8 8 - // -- 2k -- 2k+1 - // p(x) = > C * x q(x) = > C * x - // -- 2k -- 2k+1 - // k=0 k=0 - // - // x - // e - 1 ~ x * [ p(x) + x * q(x) ] - // + 2 3 4 5 6 7 8 9 + x x x x x x x x x + e - 1 ~ x + --- + --- + --- + --- + --- + --- + --- + --- + ... + 2! 3! 4! 5! 6! 7! 8! 9! + + 2 3 4 5 6 7 8 + x x x x x x x x + = x [ 1 + --- + --- + --- + --- + --- + --- + --- + --- + ... ] + 2! 3! 4! 5! 6! 7! 8! 9! + + 8 8 + -- 2k -- 2k+1 + p(x) = > C * x q(x) = > C * x + -- 2k -- 2k+1 + k=0 k=0 + + x + e - 1 ~ x * [ p(x) + x * q(x) ] + */ float128_t t = EvalPoly(x, (const float128_t*) exp_arr, EXP_ARR_SIZE); return f128_mul(t, x); @@ -184,3 +197,295 @@ extFloat80_t extFloat80_2xm1(extFloat80_t a) return a; } } + +// ================================================= +// x +// Compute e +// ================================================= + +// +// Uses the following identities: +// +// 1. ---------------------------------------------------------- +// 2 3 4 5 n +// x x x x x x x +// e = 1 + --- + --- + --- + --- + --- + ... + --- + ... +// 1! 2! 3! 4! 5! n! +// +extFloat80_t extFloat80_etox(extFloat80_t a) +{ + uint64_t zSig0, zSig1, zSig2; + struct exp32_sig64 normExpSig; + + // handle unsupported extended double-precision floating encodings + if (extF80_isUnsupported(a)) { + softfloat_exceptionFlags |= softfloat_flag_invalid; + return floatx80_default_nan; + } + + uint64_t aSig = extF80_fraction(a); + int32_t aExp = extF80_exp(a); + int aSign = extF80_sign(a); + + if (aExp == 0x7FFF) { + if (aSig << 1) { + const uint128 nan = softfloat_propagateNaNExtF80UI(a.signExp, aSig, 0, 0); + extFloat80_t rv; + rv.signExp = nan.v64; + rv.signif = nan.v0; + return rv; + } + + return (aSign) ? floatx80_negone : a; + } + + if (! aExp) { + if (! aSig) return a; + softfloat_exceptionFlags |= softfloat_flag_inexact; // denormal also + normExpSig = softfloat_normSubnormalExtF80Sig(aSig); + aExp = normExpSig.exp + 1; + aSig = normExpSig.sig; + + mul128By64To192(0x1000000000000000, 0, aSig, &zSig0, &zSig1, &zSig2); + if (0 < (int64_t) zSig0) { + shortShift128Left(zSig0, zSig1, 1, &zSig0, &zSig1); + --aExp; + } + return softfloat_roundPackToExtF80(aSign, aExp, zSig0, zSig1, 80); + } + + softfloat_exceptionFlags |= softfloat_flag_inexact; + + const extFloat80_t base = f128_to_extF80(float128_e); + extFloat80_t val = floatx80_posone; + float128_t x; + + while (!(extF80_lt(a, floatx80_posone))) { + val = extF80_mul(val, base); + + if (extF80_exp(val) == 0x7FFF) { // inf + goto ret; + } + + a = extF80_sub(a, floatx80_posone); + } + + while (extF80_le(a, floatx80_negone)) { + val = extF80_div(val, base); + + if (!extF80_exp(val)) { // subnormal + goto ret; + } + + a = extF80_add(a, floatx80_posone); + } + +ret: + /* ******************************** */ + /* using float128 for approximation */ + /* ******************************** */ + x = extF80_to_f128(a); + x = poly_exp(x); + + val = extF80_mul(val, extF80_add(f128_to_extF80(x), floatx80_posone)); + + return val; +} + +// ================================================= +// x +// Compute 2 +// ================================================= + +// +// Uses the following identities: +// +// 1. ---------------------------------------------------------- +// x x*ln(2) +// 2 = e +// +// 2. ---------------------------------------------------------- +// 2 3 4 5 n +// x x x x x x x +// e = 1 + --- + --- + --- + --- + --- + ... + --- + ... +// 1! 2! 3! 4! 5! n! +// +extFloat80_t extFloat80_2tox(extFloat80_t a) +{ + uint64_t zSig0, zSig1, zSig2; + struct exp32_sig64 normExpSig; + + // handle unsupported extended double-precision floating encodings + if (extF80_isUnsupported(a)) { + softfloat_exceptionFlags |= softfloat_flag_invalid; + return floatx80_default_nan; + } + + uint64_t aSig = extF80_fraction(a); + int32_t aExp = extF80_exp(a); + int aSign = extF80_sign(a); + + if (aExp == 0x7FFF) { // inf or NaN + if (aSig << 1) { // NaN + const uint128 nan = softfloat_propagateNaNExtF80UI(a.signExp, aSig, 0, 0); + extFloat80_t rv; + rv.signExp = nan.v64; + rv.signif = nan.v0; + return rv; + } + + return (aSign) ? floatx80_negone : a; // -inf or +inf + } + + if (! aExp) { // subnormal + if (! aSig) return a; + softfloat_exceptionFlags |= softfloat_flag_inexact; // denormal also + normExpSig = softfloat_normSubnormalExtF80Sig(aSig); + aExp = normExpSig.exp + 1; + aSig = normExpSig.sig; + + mul128By64To192(LN2_SIG_HI, LN2_SIG_LO, aSig, &zSig0, &zSig1, &zSig2); + if (0 < (int64_t) zSig0) { + shortShift128Left(zSig0, zSig1, 1, &zSig0, &zSig1); + --aExp; + } + return softfloat_roundPackToExtF80(aSign, aExp, zSig0, zSig1, 80); + } + + softfloat_exceptionFlags |= softfloat_flag_inexact; + + const extFloat80_t base = i32_to_extF80(2); + extFloat80_t val = floatx80_posone; + float128_t x; + + while (!(extF80_lt(a, floatx80_posone))) { + val = extF80_mul(val, base); + + if (extF80_exp(val) == 0x7FFF) { // inf + goto ret; + } + + a = extF80_sub(a, floatx80_posone); + } + + while (extF80_le(a, floatx80_negone)) { + val = extF80_div(val, base); + + if (!extF80_exp(val)) { // subnormal + goto ret; + } + + a = extF80_add(a, floatx80_posone); + } + +ret: + /* ******************************** */ + /* using float128 for approximation */ + /* ******************************** */ + x = extF80_to_f128(a); + x = f128_mul(x, float128_ln2); + x = poly_exp(x); + + val = extF80_mul(val, extF80_add(f128_to_extF80(x), floatx80_posone)); + + return val; +} + +// ================================================= +// x +// Compute 10 +// ================================================= + +// +// Uses the following identities: +// +// 1. ---------------------------------------------------------- +// x x*ln(10) +// 2 = e +// +// 2. ---------------------------------------------------------- +// 2 3 4 5 n +// x x x x x x x +// e = 1 + --- + --- + --- + --- + --- + ... + --- + ... +// 1! 2! 3! 4! 5! n! +// +extFloat80_t extFloat80_10tox(extFloat80_t a) +{ + uint64_t zSig0, zSig1, zSig2; + struct exp32_sig64 normExpSig; + + // handle unsupported extended double-precision floating encodings + if (extF80_isUnsupported(a)) { + softfloat_exceptionFlags |= softfloat_flag_invalid; + return floatx80_default_nan; + } + + uint64_t aSig = extF80_fraction(a); + int32_t aExp = extF80_exp(a); + int aSign = extF80_sign(a); + + if (aExp == 0x7FFF) { + if (aSig << 1) { + const uint128 nan = softfloat_propagateNaNExtF80UI(a.signExp, aSig, 0, 0); + extFloat80_t rv; + rv.signExp = nan.v64; + rv.signif = nan.v0; + return rv; + } + + return (aSign) ? floatx80_negone : a; + } + + if (! aExp) { + if (! aSig) return a; + softfloat_exceptionFlags |= softfloat_flag_inexact; // denormal also + normExpSig = softfloat_normSubnormalExtF80Sig(aSig); + aExp = normExpSig.exp + 1; + aSig = normExpSig.sig; + + mul128By64To192(LN10_SIG_HI, LN10_SIG_LO, aSig, &zSig0, &zSig1, &zSig2); + if (0 < (int64_t) zSig0) { + shortShift128Left(zSig0, zSig1, 1, &zSig0, &zSig1); + --aExp; + } + return softfloat_roundPackToExtF80(aSign, aExp, zSig0, zSig1, 80); + } + + softfloat_exceptionFlags |= softfloat_flag_inexact; + + const extFloat80_t base = i32_to_extF80(10); + extFloat80_t val = floatx80_posone; + float128_t x; + + while (!(extF80_lt(a, floatx80_posone))) { + val = extF80_mul(val, base); + + if (extF80_exp(val) == 0x7FFF) { // inf + goto ret; + } + + a = extF80_sub(a, floatx80_posone); + } + + while (extF80_le(a, floatx80_negone)) { + val = extF80_div(val, base); + + if (!extF80_exp(val)) { // subnormal + goto ret; + } + + a = extF80_add(a, floatx80_posone); + } + +ret: + /* ******************************** */ + /* using float128 for approximation */ + /* ******************************** */ + x = extF80_to_f128(a); + x = f128_mul(x, float128_ln10); + x = poly_exp(x); + + val = extF80_mul(val, extF80_add(f128_to_extF80(x), floatx80_posone)); + + return val; +} diff --git a/3rdparty/softfloat3/bochs_ext/softfloat3_ext.h b/3rdparty/softfloat3/bochs_ext/softfloat3_ext.h index a12675901bc..03b3688c740 100644 --- a/3rdparty/softfloat3/bochs_ext/softfloat3_ext.h +++ b/3rdparty/softfloat3/bochs_ext/softfloat3_ext.h @@ -11,6 +11,9 @@ extFloat80_t extFloat80_getman(extFloat80_t a); extFloat80_t extFloat80_fyl2x(extFloat80_t a, extFloat80_t b); extFloat80_t extFloat80_fyl2xp1(extFloat80_t a, extFloat80_t b); extFloat80_t extFloat80_2xm1(extFloat80_t a); +extFloat80_t extFloat80_etox(extFloat80_t a); +extFloat80_t extFloat80_2tox(extFloat80_t a); +extFloat80_t extFloat80_10tox(extFloat80_t a); extFloat80_t extFloat80_lognp1(extFloat80_t a); extFloat80_t extFloat80_logn(extFloat80_t a); extFloat80_t extFloat80_log2(extFloat80_t a); diff --git a/src/devices/cpu/m68000/m68kfpu.cpp b/src/devices/cpu/m68000/m68kfpu.cpp index 4fb984242fe..218e383a8fa 100644 --- a/src/devices/cpu/m68000/m68kfpu.cpp +++ b/src/devices/cpu/m68000/m68kfpu.cpp @@ -1738,6 +1738,42 @@ void m68000_musashi_device::fpgen_rm_reg(u16 w2) m_icount -= 56; break; } + case 0x08: // FETOXM1 + { + m_fpr[dst] = extF80_sub(extFloat80_etox(source), i32_to_extF80(1)); + set_condition_codes(m_fpr[dst]); + sync_exception_flags(source, dstCopy, EXC_ENB_UNDFLOW); + LOGMASKED(LOG_INSTRUCTIONS_VERBOSE, "FETOXM1: e ** %f - 1 = %f\n", fx80_to_double(source), fx80_to_double(m_fpr[dst])); + m_icount -= 568; + break; + } + case 0x10: // FETOX + { + m_fpr[dst] = extFloat80_etox(source); + set_condition_codes(m_fpr[dst]); + sync_exception_flags(source, dstCopy, EXC_ENB_UNDFLOW); + LOGMASKED(LOG_INSTRUCTIONS_VERBOSE, "FETOX: e ** %f = %f\n", fx80_to_double(source), fx80_to_double(m_fpr[dst])); + m_icount -= 520; + break; + } + case 0x11: // FTWOTOX + { + m_fpr[dst] = extFloat80_2tox(source); + set_condition_codes(m_fpr[dst]); + sync_exception_flags(source, dstCopy, EXC_ENB_UNDFLOW); + printf("FTWOTOX: 2 ** %f = %f\n", fx80_to_double(source), fx80_to_double(m_fpr[dst])); + m_icount -= 590; + break; + } + case 0x12: // FTENTOX + { + m_fpr[dst] = extFloat80_10tox(source); + set_condition_codes(m_fpr[dst]); + sync_exception_flags(source, dstCopy, EXC_ENB_UNDFLOW); + LOGMASKED(LOG_INSTRUCTIONS_VERBOSE, "FTENTOX: 10 ** %f = %f\n", fx80_to_double(source), fx80_to_double(m_fpr[dst])); + m_icount -= 590; + break; + } default: fatalerror("fpgen_rm_reg: unimplemented opmode %02X at %08X\n", opmode, m_ppc); } -- cgit v1.2.3