msvcrt: Import fma implementation from musl.

configure

@@ -19618,7 +19618,6 @@ fi
 for ac_func in \
 	exp2 \
 	exp2f \
-	fma \
 	fmaf \
 	lgamma \
 	lgammaf \

configure.ac

@@ -2658,7 +2658,6 @@ fi
 AC_CHECK_FUNCS(\
 	exp2 \
 	exp2f \
-	fma \
 	fmaf \
 	lgamma \
 	lgammaf \

dlls/msvcrt/math.c

@@ -3258,14 +3258,214 @@ double CDECL floor( double x )
 
 /*********************************************************************
  *      fma (MSVCRT.@)
+ *
+ * Copied from musl: src/math/fma.c
  */
+struct fma_num
+{
+    UINT64 m;
+    int e;
+    int sign;
+};
+
+static struct fma_num normalize(double x)
+{
+    UINT64 ix = *(UINT64*)&x;
+    int e = ix >> 52;
+    int sign = e & 0x800;
+    struct fma_num ret;
+
+    e &= 0x7ff;
+    if (!e) {
+        x *= 0x1p63;
+        ix = *(UINT64*)&x;
+        e = ix >> 52 & 0x7ff;
+        e = e ? e - 63 : 0x800;
+    }
+    ix &= (1ull << 52) - 1;
+    ix |= 1ull << 52;
+    ix <<= 1;
+    e -= 0x3ff + 52 + 1;
+
+    ret.m = ix;
+    ret.e = e;
+    ret.sign = sign;
+    return ret;
+}
+
+static void mul(UINT64 *hi, UINT64 *lo, UINT64 x, UINT64 y)
+{
+    UINT64 t1, t2, t3;
+    UINT64 xlo = (UINT32)x, xhi = x >> 32;
+    UINT64 ylo = (UINT32)y, yhi = y >> 32;
+
+    t1 = xlo * ylo;
+    t2 = xlo * yhi + xhi * ylo;
+    t3 = xhi * yhi;
+    *lo = t1 + (t2 << 32);
+    *hi = t3 + (t2 >> 32) + (t1 > *lo);
+}
+
 double CDECL fma( double x, double y, double z )
 {
-  double w = unix_funcs->fma(x, y, z);
-  if ((isinf(x) && y == 0) || (x == 0 && isinf(y))) *_errno() = EDOM;
-  else if (isinf(x) && isinf(z) && x != z) *_errno() = EDOM;
-  else if (isinf(y) && isinf(z) && y != z) *_errno() = EDOM;
-  return w;
+    int e, d, sign, samesign, nonzero;
+    UINT64 rhi, rlo, zhi, zlo;
+    struct fma_num nx, ny, nz;
+    double r;
+    INT64 i;
+
+    /* normalize so top 10bits and last bit are 0 */
+    nx = normalize(x);
+    ny = normalize(y);
+    nz = normalize(z);
+
+    if (nx.e >= 0x7ff - 0x3ff - 52 - 1 || ny.e >= 0x7ff - 0x3ff - 52 - 1) {
+        r = x * y + z;
+        if (!isnan(x) && !isnan(y) && !isnan(z) && isnan(r)) *_errno() = EDOM;
+        return r;
+    }
+    if (nz.e >= 0x7ff - 0x3ff - 52 - 1) {
+        if (nz.e > 0x7ff - 0x3ff - 52 - 1) {/* z==0 */
+            r = x * y + z;
+            if (!isnan(x) && !isnan(y) && isnan(r)) *_errno() = EDOM;
+            return r;
+        }
+        return z;
+    }
+
+    /* mul: r = x*y */
+    mul(&rhi, &rlo, nx.m, ny.m);
+    /* either top 20 or 21 bits of rhi and last 2 bits of rlo are 0 */
+
+    /* align exponents */
+    e = nx.e + ny.e;
+    d = nz.e - e;
+    /* shift bits z<<=kz, r>>=kr, so kz+kr == d, set e = e+kr (== ez-kz) */
+    if (d > 0) {
+        if (d < 64) {
+            zlo = nz.m << d;
+            zhi = nz.m >> (64 - d);
+        } else {
+            zlo = 0;
+            zhi = nz.m;
+            e = nz.e - 64;
+            d -= 64;
+            if (d < 64 && d) {
+                rlo = rhi << (64 - d) | rlo >> d | !!(rlo << (64 - d));
+                rhi = rhi >> d;
+            } else if (d) {
+                rlo = 1;
+                rhi = 0;
+            }
+        }
+    } else {
+        zhi = 0;
+        d = -d;
+        if (d == 0) {
+            zlo = nz.m;
+        } else if (d < 64) {
+            zlo = nz.m >> d | !!(nz.m << (64 - d));
+        } else {
+            zlo = 1;
+        }
+    }
+
+    /* add */
+    sign = nx.sign ^ ny.sign;
+    samesign = !(sign ^ nz.sign);
+    nonzero = 1;
+    if (samesign) {
+        /* r += z */
+        rlo += zlo;
+        rhi += zhi + (rlo < zlo);
+    } else {
+        /* r -= z */
+        UINT64 t = rlo;
+        rlo -= zlo;
+        rhi = rhi - zhi - (t < rlo);
+        if (rhi >> 63) {
+            rlo = -rlo;
+            rhi = -rhi - !!rlo;
+            sign = !sign;
+        }
+        nonzero = !!rhi;
+    }
+
+    /* set rhi to top 63bit of the result (last bit is sticky) */
+    if (nonzero) {
+        e += 64;
+        if (rhi >> 32) {
+            BitScanReverse((DWORD*)&d, rhi >> 32);
+            d = 31 - d - 1;
+        } else {
+            BitScanReverse((DWORD*)&d, rhi);
+            d = 63 - d - 1;
+        }
+        /* note: d > 0 */
+        rhi = rhi << d | rlo >> (64 - d) | !!(rlo << d);
+    } else if (rlo) {
+        if (rlo >> 32) {
+            BitScanReverse((DWORD*)&d, rlo >> 32);
+            d = 31 - d - 1;
+        } else {
+            BitScanReverse((DWORD*)&d, rlo);
+            d = 63 - d - 1;
+        }
+        if (d < 0)
+            rhi = rlo >> 1 | (rlo & 1);
+        else
+            rhi = rlo << d;
+    } else {
+        /* exact +-0 */
+        return x * y + z;
+    }
+    e -= d;
+
+    /* convert to double */
+    i = rhi; /* i is in [1<<62,(1<<63)-1] */
+    if (sign)
+        i = -i;
+    r = i; /* |r| is in [0x1p62,0x1p63] */
+
+    if (e < -1022 - 62) {
+        /* result is subnormal before rounding */
+        if (e == -1022 - 63) {
+            double c = 0x1p63;
+            if (sign)
+                c = -c;
+            if (r == c) {
+                /* min normal after rounding, underflow depends
+                   on arch behaviour which can be imitated by
+                   a double to float conversion */
+                float fltmin = 0x0.ffffff8p-63 * FLT_MIN * r;
+                return DBL_MIN / FLT_MIN * fltmin;
+            }
+            /* one bit is lost when scaled, add another top bit to
+               only round once at conversion if it is inexact */
+            if (rhi << 53) {
+                double tiny;
+
+                i = rhi >> 1 | (rhi & 1) | 1ull << 62;
+                if (sign)
+                    i = -i;
+                r = i;
+                r = 2 * r - c; /* remove top bit */
+
+                /* raise underflow portably, such that it
+                   cannot be optimized away */
+                tiny = DBL_MIN / FLT_MIN * r;
+                r += (double)(tiny * tiny) * (r - r);
+            }
+        } else {
+            /* only round once when scaled */
+            d = 10;
+            i = (rhi >> d | !!(rhi << (64 - d))) << d;
+            if (sign)
+                i = -i;
+            r = i;
+        }
+    }
+    return __scalbn(r, e);
 }
 
 /*********************************************************************

dlls/msvcrt/unixlib.c

@@ -83,18 +83,6 @@ static float CDECL unix_exp2f( float x )
 }
 
 /*********************************************************************
- *      fma
- */
-static double CDECL unix_fma( double x, double y, double z )
-{
-#ifdef HAVE_FMA
-    return fma(x, y, z);
-#else
-    return x * y + z;
-#endif
-}
-
-/*********************************************************************
  *      fmaf
  */
 static float CDECL unix_fmaf( float x, float y, float z )
@@ -292,7 +280,6 @@ static const struct unix_funcs funcs =
     unix_expf,
     unix_exp2,
     unix_exp2f,
-    unix_fma,
     unix_fmaf,
     unix_frexp,
     unix_frexpf,

dlls/msvcrt/unixlib.h

@@ -27,7 +27,6 @@ struct unix_funcs
     float           (CDECL *expf)(float x);
     double          (CDECL *exp2)(double x);
     float           (CDECL *exp2f)(float x);
-    double          (CDECL *fma)(double x, double y, double z);
     float           (CDECL *fmaf)(float x, float y, float z);
     double          (CDECL *frexp)(double x, int *exp);
     float           (CDECL *frexpf)(float x, int *exp);

include/config.h.in

@@ -120,9 +120,6 @@
 /* Define to 1 if you have the <float.h> header file. */
 #undef HAVE_FLOAT_H
 
-/* Define to 1 if you have the `fma' function. */
-#undef HAVE_FMA
-
 /* Define to 1 if you have the `fmaf' function. */
 #undef HAVE_FMAF