llvm-project/libclc/riscv32/lib/compiler-rt/truncdfsf2.cl

131 lines
3.9 KiB
C

//===-- lib/truncdfsf2.c - double -> single conversion ------------*- C -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// Referenced from truncdfsf2.c, but implemented in OpenCL language
//
//===----------------------------------------------------------------------===//
#ifdef cl_khr_fp64
#pragma OPENCL EXTENSION cl_khr_fp64 : enable
#include "types.h"
typedef double src_t;
typedef uint64_t src_rep_t;
static const int srcSigBits = 52;
typedef float dst_t;
typedef uint32_t dst_rep_t;
static const int dstSigBits = 23;
static __inline src_rep_t srcToRep(src_t x) {
const union {
src_t f;
src_rep_t i;
} rep = {.f = x};
return rep.i;
}
static __inline dst_t dstFromRep(dst_rep_t x) {
const union {
dst_t f;
dst_rep_t i;
} rep = {.i = x};
return rep.f;
}
static __inline dst_t __truncXfYf2__(src_t a) {
const int srcBits = sizeof(src_t) * 8;
const int srcExpBits = srcBits - srcSigBits - 1;
const int srcInfExp = (1 << srcExpBits) - 1;
const int srcExpBias = srcInfExp >> 1;
const src_rep_t srcMinNormal = 1UL << srcSigBits;
const src_rep_t srcSignificandMask = srcMinNormal - 1;
const src_rep_t srcInfinity = (src_rep_t)srcInfExp << srcSigBits;
const src_rep_t srcSignMask = 1UL << (srcSigBits + srcExpBits);
const src_rep_t srcAbsMask = srcSignMask - 1;
const src_rep_t roundMask = (1UL << (srcSigBits - dstSigBits)) - 1;
const src_rep_t halfway = 1UL << (srcSigBits - dstSigBits - 1);
const src_rep_t srcQNaN = 1UL << (srcSigBits - 1);
const src_rep_t srcNaNCode = srcQNaN - 1;
const int dstBits = sizeof(dst_t) * 8;
const int dstExpBits = dstBits - dstSigBits - 1;
const int dstInfExp = (1 << dstExpBits) - 1;
const int dstExpBias = dstInfExp >> 1;
const int underflowExponent = srcExpBias + 1 - dstExpBias;
const int overflowExponent = srcExpBias + dstInfExp - dstExpBias;
const src_rep_t underflow = (src_rep_t)underflowExponent << srcSigBits;
const src_rep_t overflow = (src_rep_t)overflowExponent << srcSigBits;
const dst_rep_t dstQNaN = 1U << (dstSigBits - 1);
const dst_rep_t dstNaNCode = dstQNaN - 1;
const src_rep_t aRep = srcToRep(a);
const src_rep_t aAbs = aRep & srcAbsMask;
const src_rep_t sign = aRep & srcSignMask;
dst_rep_t absResult;
if (aAbs - underflow < aAbs - overflow) {
absResult = aAbs >> (srcSigBits - dstSigBits);
absResult -= (dst_rep_t)(srcExpBias - dstExpBias) << dstSigBits;
const src_rep_t roundBits = aAbs & roundMask;
if (roundBits > halfway)
absResult++;
else if (roundBits == halfway)
absResult += absResult & 1;
} else if (aAbs > srcInfinity) {
absResult = (dst_rep_t)dstInfExp << dstSigBits;
absResult |= dstQNaN;
absResult |=
((aAbs & srcNaNCode) >> (srcSigBits - dstSigBits)) & dstNaNCode;
} else if (aAbs >= overflow) {
absResult = (dst_rep_t)dstInfExp << dstSigBits;
} else {
const int aExp = aAbs >> srcSigBits;
const int shift = srcExpBias - dstExpBias - aExp + 1;
const src_rep_t significand = (aRep & srcSignificandMask) | srcMinNormal;
if (shift > srcSigBits) {
absResult = 0;
} else {
const _Bool sticky = (significand << (srcBits - shift)) != 0;
src_rep_t denormalizedSignificand = significand >> shift | sticky;
absResult = denormalizedSignificand >> (srcSigBits - dstSigBits);
const src_rep_t roundBits = denormalizedSignificand & roundMask;
if (roundBits > halfway)
absResult++;
else if (roundBits == halfway)
absResult += absResult & 1;
}
}
const dst_rep_t result = absResult | sign >> (srcBits - dstBits);
return dstFromRep(result);
}
float __truncdfsf2(double a) { return __truncXfYf2__(a); }
#endif