Re: [PATCH] libgcc/m68k: More fixes for soft float

[Keith Packard <keithp@keithp.com>]

[-- Attachment #1.1: Type: text/plain, Size: 386 bytes --]


I thought I had sent all of my m68k soft float fixes along last year,
but it looks like I managed to leave some out. I had been building
my toolchain from a private repo (oops), which masked my mistake. We're
fixing that by automatically building toolchains from upstream source
with a list of patches now, so this "shouldn't" happen again. Now to
drive that list of patches to zero.


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1.2: 0001-libgcc-m68k-Fixes-for-soft-float.patch --]
[-- Type: text/x-diff, Size: 7515 bytes --]

From 1a3e403603f5a687724c7e2f6625debc3d23d934 Mon Sep 17 00:00:00 2001
From: Keith Packard <keithp@keithp.com>
Date: Sun, 29 Dec 2024 01:42:29 -0800
Subject: [PATCH] libgcc/m68k: Fixes for soft float

Fix __extenddfxf2:

  * Remove bogus denorm handling block which would never execute --
    the converted exp value is always positive as EXCESSX > EXCESSD.

  * Compute the whole significand in dl instead of doing part of it in
    ldl.

    * Mask off exponent from dl.l.upper so the denorm shift test
      works.

    * Insert the hidden one bit into dl.l.upper as needed.

Fix __truncxfdf2 denorm handling. All that is required is to shift the
significand right by the correct amount; it already has all of the
necessary bits set including the explicit one. Compute the shift
amount, then perform the wide shift across both elements of the
significand.

Fix __fixxfsi:

  * The value  was off by a factor of two as the significand contains
    32 bits, not 31 so we need to shift by one more than the equivalent
    code in __fixdfsi.

  * Simplify the code having realized that the lower 32 bits of the
    significand can never appear in the results.

Return positive qNaN instead of negative. For floats, qNaN is 0x7fff_ffff. For
doubles, qNaN is 0x7fff_ffff_ffff_ffff.

Return correctly signed zero on float and double divide underflow. This means
that Ld$underflow now expects d7 to contain the sign bit, just like the other
return paths.

Signed-off-by: Keith Packard <keithp@keithp.com>
---
 libgcc/config/m68k/fpgnulib.c | 78 ++++++++++++++++-------------------
 libgcc/config/m68k/lb1sf68.S  | 13 ++++--
 2 files changed, 45 insertions(+), 46 deletions(-)

diff --git a/libgcc/config/m68k/fpgnulib.c b/libgcc/config/m68k/fpgnulib.c
index 70bfd442d75..a7d4258dff0 100644
--- a/libgcc/config/m68k/fpgnulib.c
+++ b/libgcc/config/m68k/fpgnulib.c
@@ -449,34 +449,37 @@ __extenddfxf2 (double d)
     }
 
   exp = EXPD (dl) - EXCESSD + EXCESSX;
-  /* Check for underflow and denormals. */
-  if (exp < 0)
+
+  dl.l.upper &= MANTDMASK;
+
+  /* Recover from a denorm. */
+  if (exp == -EXCESSD + EXCESSX)
     {
-      if (exp < -53)
-	{
-	  ldl.l.middle = 0;
-	  ldl.l.lower = 0;
-	}
-      else if (exp < -30)
-	{
-	  ldl.l.lower = (ldl.l.middle & MANTXMASK) >> ((1 - exp) - 32);
-	  ldl.l.middle &= ~MANTXMASK;
-	}
-      else
+      exp++;
+      while ((dl.l.upper & HIDDEND) == 0)
 	{
-	  ldl.l.lower >>= 1 - exp;
-	  ldl.l.lower |= (ldl.l.middle & MANTXMASK) << (32 - (1 - exp));
-	  ldl.l.middle = (ldl.l.middle & ~MANTXMASK) | (ldl.l.middle & MANTXMASK >> (1 - exp));
+	  exp--;
+	  dl.l.upper = (dl.l.upper << 1) | (dl.l.lower >> 31);
+	  dl.l.lower = dl.l.lower << 1;
 	}
-      exp = 0;
     }
+
   /* Handle inf and NaN */
-  if (exp == EXPDMASK - EXCESSD + EXCESSX)
-    exp = EXPXMASK;
+  else if (exp == EXPDMASK - EXCESSD + EXCESSX)
+    {
+      exp = EXPXMASK;
+      /* Add hidden one bit for NaN */
+      if (dl.l.upper != 0 || dl.l.lower != 0)
+        dl.l.upper |= HIDDEND;
+    }
+  else
+    {
+      dl.l.upper |= HIDDEND;
+    }
+
   ldl.l.upper |= exp << 16;
-  ldl.l.middle = HIDDENX;
   /* 31-20: # mantissa bits in ldl.l.middle - # mantissa bits in dl.l.upper */
-  ldl.l.middle |= (dl.l.upper & MANTDMASK) << (31 - 20);
+  ldl.l.middle = dl.l.upper << (31 - 20);
   /* 1+20: explicit-integer-bit + # mantissa bits in dl.l.upper */
   ldl.l.middle |= dl.l.lower >> (1 + 20);
   /* 32 - 21: # bits of dl.l.lower in ldl.l.middle */
@@ -508,21 +511,21 @@ __truncxfdf2 (long double ld)
   /* Check for underflow and denormals. */
   if (exp <= 0)
     {
-      if (exp < -53)
+      long shift = 1 - exp;
+      if (shift > 52)
 	{
 	  ldl.l.middle = 0;
 	  ldl.l.lower = 0;
 	}
-      else if (exp < -30)
+      else if (shift >= 32)
 	{
-	  ldl.l.lower = (ldl.l.middle & MANTXMASK) >> ((1 - exp) - 32);
-	  ldl.l.middle &= ~MANTXMASK;
+	  ldl.l.lower = (ldl.l.middle) >> (shift - 32);
+          ldl.l.middle = 0;
 	}
       else
 	{
-	  ldl.l.lower >>= 1 - exp;
-	  ldl.l.lower |= (ldl.l.middle & MANTXMASK) << (32 - (1 - exp));
-	  ldl.l.middle = (ldl.l.middle & ~MANTXMASK) | (ldl.l.middle & MANTXMASK >> (1 - exp));
+	  ldl.l.lower = (ldl.l.middle << (32 - shift)) | (ldl.l.lower >> shift);
+          ldl.l.middle = ldl.l.middle >> shift;
 	}
       exp = 0;
     }
@@ -585,7 +588,6 @@ __fixxfsi (long double a)
 {
   union long_double_long ldl;
   long exp;
-  long l;
 
   ldl.ld = a;
 
@@ -593,28 +595,20 @@ __fixxfsi (long double a)
   if (exp == 0 && ldl.l.middle == 0 && ldl.l.lower == 0)
     return 0;
 
-  exp = exp - EXCESSX - 63;
+  exp = exp - EXCESSX - 32;
 
-  if (exp > 0)
+  if (exp >= 0)
     {
       /* Return largest integer.  */
       return SIGNX (ldl) ? 0x80000000L : 0x7fffffffL;
     }
 
-  if (exp <= -64)
+  if (exp <= -32)
     return 0;
 
-  if (exp <= -32)
-    {
-      ldl.l.lower = ldl.l.middle >> (-exp - 32);
-    }
-  else if (exp < 0)
-    {
-      ldl.l.lower = ldl.l.lower >> -exp;
-      ldl.l.lower |= ldl.l.middle << (32 + exp);
-    }
+  ldl.l.middle >>= -exp;
 
-  return SIGNX (ldl) ? -ldl.l.lower : ldl.l.lower;
+  return SIGNX (ldl) ? -ldl.l.middle : ldl.l.middle;
 }
 
 /* The remaining provide crude math support by working in double precision.  */
diff --git a/libgcc/config/m68k/lb1sf68.S b/libgcc/config/m68k/lb1sf68.S
index 22f2772520c..ebec095c6e7 100644
--- a/libgcc/config/m68k/lb1sf68.S
+++ b/libgcc/config/m68k/lb1sf68.S
@@ -635,7 +635,7 @@ SYM (__modsi3):
 	.globl	SYM (_fpCCR)
 	.globl  $_exception_handler
 
-QUIET_NaN      = 0xffffffff
+QUIET_NaN      = 0x7fffffff
 
 D_MAX_EXP      = 0x07ff
 D_BIAS         = 1022
@@ -700,9 +700,10 @@ Ld$overflow:
 	PICJUMP	$_exception_handler
 
 Ld$underflow:
-| Return 0 and set the exception flags 
+| Return a properly signed 0 and set the exception flags 
 	movel	IMM (0),d0
 	movel	d0,d1
+	orl	d7,d0
 	movew	IMM (INEXACT_RESULT+UNDERFLOW),d7
 	moveq	IMM (DOUBLE_FLOAT),d6
 	PICJUMP	$_exception_handler
@@ -711,6 +712,7 @@ Ld$inop:
 | Return a quiet NaN and set the exception flags
 	movel	IMM (QUIET_NaN),d0
 	movel	d0,d1
+	bset	IMM (31),d1
 	movew	IMM (INEXACT_RESULT+INVALID_OPERATION),d7
 	moveq	IMM (DOUBLE_FLOAT),d6
 	PICJUMP	$_exception_handler
@@ -2082,6 +2084,7 @@ Ldivdf$b$nf:
 | If d2 == 0x7ff00000 we have to check d3.
 	tstl	d3		|
 	bne	Ld$inop		| if d3 <> 0, b is NaN
+	movel	a0,d7		| put a's sign
 	bra	Ld$underflow	| else b is +/-INFINITY, so signal underflow
 
 Ldivdf$a$nf:
@@ -2187,6 +2190,7 @@ Lround$exit:
 #endif
 	beq	2f		| if not loop back
 	bra	1b              |
+	movel	a0,d7		| get back sign bit into d7
 	bra	Ld$underflow	| safety check, shouldn't execute '
 2:	orl	d6,d2		| this is a trick so we don't lose  '
 	orl	d7,d3		| the bits which were flushed right
@@ -2549,7 +2553,7 @@ Lround$to$minus:
 	.globl	SYM (_fpCCR)
 	.globl  $_exception_handler
 
-QUIET_NaN    = 0xffffffff
+QUIET_NaN    = 0x7fffffff
 SIGNL_NaN    = 0x7f800001
 INFINITY     = 0x7f800000
 
@@ -2615,8 +2619,9 @@ Lf$overflow:
 	PICJUMP	$_exception_handler
 
 Lf$underflow:
-| Return 0 and set the exception flags 
+| Return a properly signed 0 and set the exception flags 
 	moveq	IMM (0),d0
+	orl	d7,d0
 	moveq	IMM (INEXACT_RESULT+UNDERFLOW),d7
 	moveq	IMM (SINGLE_FLOAT),d6
 	PICJUMP	$_exception_handler
-- 
2.45.2


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]