From f111d72596c4071ad38a2062699f17702bbd9c6d Mon Sep 17 00:00:00 2001 From: Elie Tournier Date: Tue, 8 Aug 2017 15:39:58 +0100 Subject: [PATCH] glsl: Add "built-in" functions to do add(fp64, fp64) v2: use mix and findMSB to optimise. v3: [Sagar] Fix zFrac0 == 0u case in __normalizeRoundAndPackFloat64 Signed-off-by: Elie Tournier --- src/compiler/glsl/float64.glsl | 433 +++++++++++++++++++++++++++++++++ 1 file changed, 433 insertions(+) diff --git a/src/compiler/glsl/float64.glsl b/src/compiler/glsl/float64.glsl index a2642e9b34e..858e2f3b87d 100644 --- a/src/compiler/glsl/float64.glsl +++ b/src/compiler/glsl/float64.glsl @@ -44,6 +44,7 @@ #extension GL_ARB_gpu_shader_int64 : enable #extension GL_ARB_shader_bit_encoding : enable #extension GL_EXT_shader_integer_mix : enable +#extension GL_MESA_shader_integer_functions : enable #pragma warning(off) @@ -216,3 +217,435 @@ __fge64(uint64_t a, uint64_t b) return !__flt64_nonnan(a, b); } + +/* Adds the 64-bit value formed by concatenating `a0' and `a1' to the 64-bit + * value formed by concatenating `b0' and `b1'. Addition is modulo 2^64, so + * any carry out is lost. The result is broken into two 32-bit pieces which + * are stored at the locations pointed to by `z0Ptr' and `z1Ptr'. + */ +void +__add64(uint a0, uint a1, uint b0, uint b1, + out uint z0Ptr, + out uint z1Ptr) +{ + uint z1 = a1 + b1; + z1Ptr = z1; + z0Ptr = a0 + b0 + uint(z1 < a1); +} + + +/* Subtracts the 64-bit value formed by concatenating `b0' and `b1' from the + * 64-bit value formed by concatenating `a0' and `a1'. Subtraction is modulo + * 2^64, so any borrow out (carry out) is lost. The result is broken into two + * 32-bit pieces which are stored at the locations pointed to by `z0Ptr' and + * `z1Ptr'. + */ +void +__sub64(uint a0, uint a1, uint b0, uint b1, + out uint z0Ptr, + out uint z1Ptr) +{ + z1Ptr = a1 - b1; + z0Ptr = a0 - b0 - uint(a1 < b1); +} + +/* Shifts the 64-bit value formed by concatenating `a0' and `a1' right by the + * number of bits given in `count'. If any nonzero bits are shifted off, they + * are "jammed" into the least significant bit of the result by setting the + * least significant bit to 1. The value of `count' can be arbitrarily large; + * in particular, if `count' is greater than 64, the result will be either 0 + * or 1, depending on whether the concatenation of `a0' and `a1' is zero or + * nonzero. The result is broken into two 32-bit pieces which are stored at + * the locations pointed to by `z0Ptr' and `z1Ptr'. + */ +void +__shift64RightJamming(uint a0, + uint a1, + int count, + out uint z0Ptr, + out uint z1Ptr) +{ + uint z0; + uint z1; + int negCount = (-count) & 31; + + z0 = mix(0u, a0, count == 0); + z0 = mix(z0, (a0 >> count), count < 32); + + z1 = uint((a0 | a1) != 0u); /* count >= 64 */ + uint z1_lt64 = (a0>>(count & 31)) | uint(((a0<>count) | uint ((a1<> (count & 31)), count < 64); + z1 = mix(z1, (a0<>count), count < 32); + + a2 = mix(a2 | a1, a2, count < 32); + z0 = mix(z0, a0 >> count, count < 32); + z2 |= uint(a2 != 0u); + + z0 = mix(z0, 0u, (count == 32)); + z1 = mix(z1, a0, (count == 32)); + z2 = mix(z2, a1, (count == 32)); + z0 = mix(z0, a0, (count == 0)); + z1 = mix(z1, a1, (count == 0)); + z2 = mix(z2, a2, (count == 0)); + z2Ptr = z2; + z1Ptr = z1; + z0Ptr = z0; +} + +/* Shifts the 64-bit value formed by concatenating `a0' and `a1' left by the + * number of bits given in `count'. Any bits shifted off are lost. The value + * of `count' must be less than 32. The result is broken into two 32-bit + * pieces which are stored at the locations pointed to by `z0Ptr' and `z1Ptr'. + */ +void +__shortShift64Left(uint a0, uint a1, + int count, + out uint z0Ptr, + out uint z1Ptr) +{ + z1Ptr = a1<> ((-count) & 31))), a0, count == 0); +} + +/* Packs the sign `zSign', the exponent `zExp', and the significand formed by + * the concatenation of `zFrac0' and `zFrac1' into a double-precision floating- + * point value, returning the result. After being shifted into the proper + * positions, the three fields `zSign', `zExp', and `zFrac0' are simply added + * together to form the most significant 32 bits of the result. This means + * that any integer portion of `zFrac0' will be added into the exponent. Since + * a properly normalized significand will have an integer portion equal to 1, + * the `zExp' input should be 1 less than the desired result exponent whenever + * `zFrac0' and `zFrac1' concatenated form a complete, normalized significand. + */ +uint64_t +__packFloat64(uint zSign, int zExp, uint zFrac0, uint zFrac1) +{ + uvec2 z; + + z.y = (zSign << 31) + (uint(zExp) << 20) + zFrac0; + z.x = zFrac1; + return packUint2x32(z); +} + +/* Takes an abstract floating-point value having sign `zSign', exponent `zExp', + * and extended significand formed by the concatenation of `zFrac0', `zFrac1', + * and `zFrac2', and returns the proper double-precision floating-point value + * corresponding to the abstract input. Ordinarily, the abstract value is + * simply rounded and packed into the double-precision format, with the inexact + * exception raised if the abstract input cannot be represented exactly. + * However, if the abstract value is too large, the overflow and inexact + * exceptions are raised and an infinity or maximal finite value is returned. + * If the abstract value is too small, the input value is rounded to a + * subnormal number, and the underflow and inexact exceptions are raised if the + * abstract input cannot be represented exactly as a subnormal double-precision + * floating-point number. + * The input significand must be normalized or smaller. If the input + * significand is not normalized, `zExp' must be 0; in that case, the result + * returned is a subnormal number, and it must not require rounding. In the + * usual case that the input significand is normalized, `zExp' must be 1 less + * than the "true" floating-point exponent. The handling of underflow and + * overflow follows the IEEE Standard for Floating-Point Arithmetic. + */ +uint64_t +__roundAndPackFloat64(uint zSign, + int zExp, + uint zFrac0, + uint zFrac1, + uint zFrac2) +{ + bool roundNearestEven; + bool increment; + + roundNearestEven = FLOAT_ROUNDING_MODE == FLOAT_ROUND_NEAREST_EVEN; + increment = int(zFrac2) < 0; + if (!roundNearestEven) { + if (FLOAT_ROUNDING_MODE == FLOAT_ROUND_TO_ZERO) { + increment = false; + } else { + if (zSign != 0u) { + increment = (FLOAT_ROUNDING_MODE == FLOAT_ROUND_DOWN) && + (zFrac2 != 0u); + } else { + increment = (FLOAT_ROUNDING_MODE == FLOAT_ROUND_UP) && + (zFrac2 != 0u); + } + } + } + if (0x7FD <= zExp) { + if ((0x7FD < zExp) || + ((zExp == 0x7FD) && + (0x001FFFFFu == zFrac0 && 0xFFFFFFFFu == zFrac1) && + increment)) { + if ((FLOAT_ROUNDING_MODE == FLOAT_ROUND_TO_ZERO) || + ((zSign != 0u) && (FLOAT_ROUNDING_MODE == FLOAT_ROUND_UP)) || + ((zSign == 0u) && (FLOAT_ROUNDING_MODE == FLOAT_ROUND_DOWN))) { + return __packFloat64(zSign, 0x7FE, 0x000FFFFFu, 0xFFFFFFFFu); + } + return __packFloat64(zSign, 0x7FF, 0u, 0u); + } + if (zExp < 0) { + __shift64ExtraRightJamming( + zFrac0, zFrac1, zFrac2, -zExp, zFrac0, zFrac1, zFrac2); + zExp = 0; + if (roundNearestEven) { + increment = zFrac2 < 0u; + } else { + if (zSign != 0u) { + increment = (FLOAT_ROUNDING_MODE == FLOAT_ROUND_DOWN) && + (zFrac2 != 0u); + } else { + increment = (FLOAT_ROUNDING_MODE == FLOAT_ROUND_UP) && + (zFrac2 != 0u); + } + } + } + } + if (increment) { + __add64(zFrac0, zFrac1, 0u, 1u, zFrac0, zFrac1); + zFrac1 &= ~((zFrac2 + uint(zFrac2 == 0u)) & uint(roundNearestEven)); + } else { + zExp = mix(zExp, 0, (zFrac0 | zFrac1) == 0u); + } + return __packFloat64(zSign, zExp, zFrac0, zFrac1); +} + +/* Returns the number of leading 0 bits before the most-significant 1 bit of + * `a'. If `a' is zero, 32 is returned. + */ +int +__countLeadingZeros32(uint a) +{ + int shiftCount; + shiftCount = mix(31 - findMSB(a), 32, a == 0u); + return shiftCount; +} + +/* Takes an abstract floating-point value having sign `zSign', exponent `zExp', + * and significand formed by the concatenation of `zSig0' and `zSig1', and + * returns the proper double-precision floating-point value corresponding + * to the abstract input. This routine is just like `__roundAndPackFloat64' + * except that the input significand has fewer bits and does not have to be + * normalized. In all cases, `zExp' must be 1 less than the "true" floating- + * point exponent. + */ +uint64_t +__normalizeRoundAndPackFloat64(uint zSign, + int zExp, + uint zFrac0, + uint zFrac1) +{ + int shiftCount; + uint zFrac2; + + if (zFrac0 == 0u) { + zExp -= 32; + zFrac0 = zFrac1; + zFrac1 = 0u; + } + + shiftCount = __countLeadingZeros32(zFrac0) - 11; + if (0 <= shiftCount) { + zFrac2 = 0u; + __shortShift64Left(zFrac0, zFrac1, shiftCount, zFrac0, zFrac1); + } else { + __shift64ExtraRightJamming( + zFrac0, zFrac1, 0u, -shiftCount, zFrac0, zFrac1, zFrac2); + } + zExp -= shiftCount; + return __roundAndPackFloat64(zSign, zExp, zFrac0, zFrac1, zFrac2); +} + +/* Takes two double-precision floating-point values `a' and `b', one of which + * is a NaN, and returns the appropriate NaN result. + */ +uint64_t +__propagateFloat64NaN(uint64_t __a, uint64_t __b) +{ + bool aIsNaN = __is_nan(__a); + bool bIsNaN = __is_nan(__b); + uvec2 a = unpackUint2x32(__a); + uvec2 b = unpackUint2x32(__b); + a.y |= 0x00080000u; + b.y |= 0x00080000u; + + return packUint2x32(mix(b, mix(a, b, bvec2(bIsNaN, bIsNaN)), bvec2(aIsNaN, aIsNaN))); +} + +/* Returns the result of adding the double-precision floating-point values + * `a' and `b'. The operation is performed according to the IEEE Standard for + * Floating-Point Arithmetic. + */ +uint64_t +__fadd64(uint64_t a, uint64_t b) +{ + uint aSign = __extractFloat64Sign(a); + uint bSign = __extractFloat64Sign(b); + uint aFracLo = __extractFloat64FracLo(a); + uint aFracHi = __extractFloat64FracHi(a); + uint bFracLo = __extractFloat64FracLo(b); + uint bFracHi = __extractFloat64FracHi(b); + int aExp = __extractFloat64Exp(a); + int bExp = __extractFloat64Exp(b); + uint zFrac0 = 0u; + uint zFrac1 = 0u; + int expDiff = aExp - bExp; + if (aSign == bSign) { + uint zFrac2 = 0u; + int zExp; + bool orig_exp_diff_is_zero = (expDiff == 0); + + if (orig_exp_diff_is_zero) { + if (aExp == 0x7FF) { + bool propagate = (aFracHi | aFracLo | bFracHi | bFracLo) != 0u; + return mix(a, __propagateFloat64NaN(a, b), propagate); + } + __add64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1); + if (aExp == 0) + return __packFloat64(aSign, 0, zFrac0, zFrac1); + zFrac2 = 0u; + zFrac0 |= 0x00200000u; + zExp = aExp; + __shift64ExtraRightJamming( + zFrac0, zFrac1, zFrac2, 1, zFrac0, zFrac1, zFrac2); + } else if (0 < expDiff) { + if (aExp == 0x7FF) { + bool propagate = (aFracHi | aFracLo) != 0u; + return mix(a, __propagateFloat64NaN(a, b), propagate); + } + + expDiff = mix(expDiff, expDiff - 1, bExp == 0); + bFracHi = mix(bFracHi | 0x00100000u, bFracHi, bExp == 0); + __shift64ExtraRightJamming( + bFracHi, bFracLo, 0u, expDiff, bFracHi, bFracLo, zFrac2); + zExp = aExp; + } else if (expDiff < 0) { + if (bExp == 0x7FF) { + bool propagate = (bFracHi | bFracLo) != 0u; + return mix(__packFloat64(aSign, 0x7ff, 0u, 0u), __propagateFloat64NaN(a, b), propagate); + } + expDiff = mix(expDiff, expDiff + 1, aExp == 0); + aFracHi = mix(aFracHi | 0x00100000u, aFracHi, aExp == 0); + __shift64ExtraRightJamming( + aFracHi, aFracLo, 0u, - expDiff, aFracHi, aFracLo, zFrac2); + zExp = bExp; + } + if (!orig_exp_diff_is_zero) { + aFracHi |= 0x00100000u; + __add64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1); + --zExp; + if (!(zFrac0 < 0x00200000u)) { + __shift64ExtraRightJamming(zFrac0, zFrac1, zFrac2, 1, zFrac0, zFrac1, zFrac2); + ++zExp; + } + } + return __roundAndPackFloat64(aSign, zExp, zFrac0, zFrac1, zFrac2); + + } else { + int zExp; + + __shortShift64Left(aFracHi, aFracLo, 10, aFracHi, aFracLo); + __shortShift64Left(bFracHi, bFracLo, 10, bFracHi, bFracLo); + if (0 < expDiff) { + if (aExp == 0x7FF) { + bool propagate = (aFracHi | aFracLo) != 0u; + return mix(a, __propagateFloat64NaN(a, b), propagate); + } + expDiff = mix(expDiff, expDiff - 1, bExp == 0); + bFracHi = mix(bFracHi | 0x40000000u, bFracHi, bExp == 0); + __shift64RightJamming(bFracHi, bFracLo, expDiff, bFracHi, bFracLo); + aFracHi |= 0x40000000u; + __sub64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1); + zExp = aExp; + --zExp; + return __normalizeRoundAndPackFloat64(aSign, zExp - 10, zFrac0, zFrac1); + } + if (expDiff < 0) { + if (bExp == 0x7FF) { + bool propagate = (bFracHi | bFracLo) != 0u; + return mix(__packFloat64(aSign ^ 1u, 0x7ff, 0u, 0u), __propagateFloat64NaN(a, b), propagate); + } + expDiff = mix(expDiff, expDiff + 1, aExp == 0); + aFracHi = mix(aFracHi | 0x40000000u, aFracHi, aExp == 0); + __shift64RightJamming(aFracHi, aFracLo, - expDiff, aFracHi, aFracLo); + bFracHi |= 0x40000000u; + __sub64(bFracHi, bFracLo, aFracHi, aFracLo, zFrac0, zFrac1); + zExp = bExp; + aSign ^= 1u; + --zExp; + return __normalizeRoundAndPackFloat64(aSign, zExp - 10, zFrac0, zFrac1); + } + if (aExp == 0x7FF) { + bool propagate = (aFracHi | aFracLo | bFracHi | bFracLo) != 0u; + return mix(0xFFFFFFFFFFFFFFFFUL, __propagateFloat64NaN(a, b), propagate); + } + bExp = mix(bExp, 1, aExp == 0); + aExp = mix(aExp, 1, aExp == 0); + bool zexp_normal = false; + bool blta = true; + if (bFracHi < aFracHi) { + __sub64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1); + zexp_normal = true; + } + else if (aFracHi < bFracHi) { + __sub64(bFracHi, bFracLo, aFracHi, aFracLo, zFrac0, zFrac1); + blta = false; + zexp_normal = true; + } + else if (bFracLo < aFracLo) { + __sub64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1); + zexp_normal = true; + } + else if (aFracLo < bFracLo) { + __sub64(bFracHi, bFracLo, aFracHi, aFracLo, zFrac0, zFrac1); + blta = false; + zexp_normal = true; + } + zExp = mix(bExp, aExp, blta); + aSign = mix(aSign ^ 1u, aSign, blta); + uint64_t retval_0 = __packFloat64(uint(FLOAT_ROUNDING_MODE == FLOAT_ROUND_DOWN), 0, 0u, 0u); + uint64_t retval_1 = __normalizeRoundAndPackFloat64(aSign, zExp - 11, zFrac0, zFrac1); + return mix(retval_0, retval_1, zexp_normal); + } +} -- 2.30.2