[illumos-gate merge]

commit 5a172a1e0f9068ec8db35fa123baecf7bd540eb5
    5224 snprintf rounding under [default] FE_TONEAREST

6 files changed, 247 insertions, 108 deletions

diff --git a/usr/src/lib/libc/amd64/Makefile b/usr/src/lib/libc/amd64/Makefile
index a7cfc36ddb..9a496b0d7e 100644
--- a/usr/src/lib/libc/amd64/Makefile
+++ b/usr/src/lib/libc/amd64/Makefile
@@ -47,12 +47,12 @@ CRTOBJS=			\
 DYNOBJS=
 
 FPOBJS=				\
-	_base_il.o		\
 	fpgetmask.o		\
 	fpgetround.o		\
 	fpsetmask.o		\
 	fpsetround.o		\
-	fpstart.o
+	fpstart.o		\
+	ieee.o
 
 I386FPOBJS=			\
 	_D_cplx_div.o		\
@@ -81,6 +81,7 @@ FPASMOBJS=			\
 	__xgetRD.o		\
 	_xtoll.o		\
 	_xtoull.o		\
+	_base_il.o		\
 	fpcw.o			\
 	fpgetsticky.o		\
 	fpsetsticky.o
diff --git a/usr/src/lib/libc/amd64/fp/_base_il.s b/usr/src/lib/libc/amd64/fp/_base_il.s
new file mode 100644
index 0000000000..221c848538
--- /dev/null
+++ b/usr/src/lib/libc/amd64/fp/_base_il.s
@@ -0,0 +1,121 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2014 OmniTI Computer Consulting, Inc. All rights reserved.
+ */
+
+	.file	"_base_il.s"
+
+/*
+ * These files are in assembly because some compilers will mistakenly reorder
+ * multiplications or divisions wrapped in _putsw() and _getsw().  They are
+ * proper subroutines for now, but should be considered candidates for
+ * inlining eventually.
+ *
+ * The original C sources are included for readability in the pre-function
+ * comment blocks.
+ */
+
+#include <SYS.h>
+
+/*
+ * Multiplies two normal or subnormal doubles, returns result and exceptions.
+ *
+
+double
+__mul_set(double x, double y, int *pe) {
+	extern void _putmxcsr(), _getmxcsr();
+	int csr;
+	double z;
+
+	_putmxcsr(CSR_DEFAULT);
+	z = x * y;
+	_getmxcsr(&csr);
+	if ((csr & 0x3f) == 0) {
+		*pe = 0;
+	} else {
+		*pe = 1;
+	}
+	return (z);
+}
+
+ */
+	ENTRY(__mul_set)
+	xorl	%eax, %eax	/* Zero-out eax for later... */
+	subq	$0x4, %rsp
+	movl	$0x1f80, (%rsp)	/* 0x1f80 == CSR_DEFAULT. */
+	/* Set the MXCSR to its default (i.e. No FP exceptions). */
+	ldmxcsr	(%rsp)	/* Essentially _putmxcsr(CSR_DEFAULT); */
+	
+	mulsd	%xmm1, %xmm0	/* Do the multiply. */
+	
+	/* Check to see if the multiply caused any exceptions. */
+	stmxcsr	(%rsp)	/* Essentially do _getmxcsr(). */
+	andl	$0x3f, (%rsp)	/* Check it. */
+	setne	%al		/* Boolean FP exception indicator for *pe. */
+	movl	%eax, (%rdi)
+	addq	$0x4, %rsp
+	ret
+	SET_SIZE(__mul_set)
+
+/*
+ * Divides two normal or subnormal doubles x/y, returns result and exceptions.
+ *
+
+double
+__div_set(double x, double y, int *pe) {
+	extern void _putmxcsr(), _getmxcsr();
+	int csr;
+	double z;
+
+	_putmxcsr(CSR_DEFAULT);
+	z = x / y;
+	_getmxcsr(&csr);
+	if ((csr & 0x3f) == 0) {
+		*pe = 0;
+	} else {
+		*pe = 1;
+	}
+	return (z);
+}
+
+ */
+	
+	ENTRY(__div_set)
+	xorl	%eax, %eax	/* Zero-out eax for later... */
+	subq	$0x4, %rsp
+	movl	$0x1f80, (%rsp)	/* 0x1f80 == CSR_DEFAULT. */
+	/* Set the MXCSR to its default (i.e. No FP exceptions). */
+	ldmxcsr	(%rsp)	/* Essentially _putmxcsr(CSR_DEFAULT); */
+	
+	divsd	%xmm1, %xmm0	/* Do the divide. */
+	
+	/* Check to see if the divide caused any exceptions. */
+	stmxcsr	(%rsp)	/* Essentially do _getmxcsr(). */
+	andl	$0x3f, (%rsp)	/* Check it. */
+	setne	%al		/* Boolean FP exception indicator for *pe. */
+	movl	%eax, (%rdi)
+	addq	$0x4, %rsp
+	ret
+	SET_SIZE(__div_set)
+
+/* double __dabs(double *d) - Get the abs. value of *d.  Straightforward. */
+
+	ENTRY(__dabs)
+	subq	$0x8, %rsp
+	movq	(%rdi), %rax	/* Zero the sign bit of the 64-bit double. */
+	btrq	$63, %rax
+	movq	%rax, (%rsp)	/* Get it into %xmm0... */
+	movsd   (%rsp), %xmm0	/* ....for an amd64 "double" return. */
+	addq	$0x8, %rsp
+	ret
+	SET_SIZE(__dabs)
diff --git a/usr/src/lib/libc/amd64/fp/_base_il.c b/usr/src/lib/libc/amd64/fp/ieee.c
index da4e3ea506..8327d64611 100644
--- a/usr/src/lib/libc/amd64/fp/_base_il.c
+++ b/usr/src/lib/libc/amd64/fp/ieee.c
@@ -24,64 +24,12 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include "lint.h"
 #include "base_conversion.h"
 #include <sys/isa_defs.h>
 
 #define	CSR_DEFAULT 0x1f80
 
-/* The following should be coded as inline expansion templates.	 */
-
-/*
- * Multiplies two normal or subnormal doubles, returns result and exceptions.
- */
-double
-__mul_set(double x, double y, int *pe) {
-	extern void _putmxcsr(), _getmxcsr();
-	int csr;
-	double z;
-
-	_putmxcsr(CSR_DEFAULT);
-	z = x * y;
-	_getmxcsr(&csr);
-	if ((csr & 0x3f) == 0) {
-		*pe = 0;
-	} else {
-		/* Result may not be exact. */
-		*pe = 1;
-	}
-	return (z);
-}
-
-/*
- * Divides two normal or subnormal doubles x/y, returns result and exceptions.
- */
-double
-__div_set(double x, double y, int *pe) {
-	extern void _putmxcsr(), _getmxcsr();
-	int csr;
-	double z;
-
-	_putmxcsr(CSR_DEFAULT);
-	z = x / y;
-	_getmxcsr(&csr);
-	if ((csr & 0x3f) == 0) {
-		*pe = 0;
-	} else {
-		*pe = 1;
-	}
-	return (z);
-}
-
-double
-__dabs(double *d)
-{
-	/* should use hardware fabs instruction */
-	return ((*d < 0.0) ? -*d : *d);
-}
-
 /*
  * Returns IEEE mode/status and
  * sets up standard environment for base conversion.
diff --git a/usr/src/lib/libc/i386/Makefile.com b/usr/src/lib/libc/i386/Makefile.com
index e8b72336f3..01f71c334c 100644
--- a/usr/src/lib/libc/i386/Makefile.com
+++ b/usr/src/lib/libc/i386/Makefile.com
@@ -70,17 +70,18 @@ FPOBJS=				\
 	_X_cplx_lr_div_ix.o	\
 	_X_cplx_lr_div_rx.o	\
 	_X_cplx_mul.o		\
-	_base_il.o		\
 	fpgetmask.o		\
 	fpgetround.o		\
 	fpgetsticky.o		\
 	fpsetmask.o		\
 	fpsetround.o		\
 	fpsetsticky.o		\
-	fpstart.o
+	fpstart.o		\
+	ieee.o
 
 FPASMOBJS=			\
 	__xgetRD.o		\
+	_base_il.o		\
 	_xtoll.o		\
 	_xtoull.o		\
 	fpcw.o
diff --git a/usr/src/lib/libc/i386/fp/_base_il.s b/usr/src/lib/libc/i386/fp/_base_il.s
new file mode 100644
index 0000000000..157e977922
--- /dev/null
+++ b/usr/src/lib/libc/i386/fp/_base_il.s
@@ -0,0 +1,120 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2014 OmniTI Computer Consulting, Inc. All rights reserved.
+ */
+
+	.file	"_base_il.s"
+
+/*
+ * These files are in assembly because some compilers will mistakenly reorder
+ * multiplications or divisions wrapped in _putsw() and _getsw().  They are
+ * proper subroutines for now, but should be considered candidates for
+ * inlining eventually.
+ *
+ * The original C sources are included for readability in the pre-function
+ * comment blocks.
+ */
+
+#include <SYS.h>
+
+/*
+ * Multiplies two normal or subnormal doubles, returns result and exceptions.
+ *
+
+double
+__mul_set(double x, double y, int *pe) {
+	extern void _putsw(), _getsw();
+	int sw;
+	double z;
+
+	_putsw(0);
+	z = x * y;
+	_getsw(&sw);
+	if ((sw & 0x3f) == 0) {
+		*pe = 0;
+	} else {
+		*pe = 1;
+	}
+	return (z);
+}
+
+ */
+	ENTRY(__mul_set)
+	subl	$0x8, %esp	/* Give us an extra 8 bytes to play with. */
+	/* Clear the floating point exception register. */
+	fnclex			/* Equivalent of _putsw(0); */
+	
+	fldl	0xc(%esp)	/* Load up x */
+	fmull	0x14(%esp)	/* And multiply! */
+	
+	/* Check to see if the multiply caused any exceptions. */
+	fstsw	(%esp)		/* Equivalent of... */
+	xorl	%edx, %edx
+	andl	$0x3f, (%esp)	/* If the status word (low bits) are zero... */
+	setne	%dl		/* ... set *pe (aka. (%eax)) accordingly. */
+	movl	0x1c(%esp), %eax/* Get pe. */
+	movl	%edx, (%eax)	/* And set it.  (True == FP exception). */
+	addl	$0x8, %esp	/* Release the 8 play bytes. */
+	ret
+	SET_SIZE(__mul_set)
+
+/*
+ * Divides two normal or subnormal doubles x/y, returns result and exceptions.
+ *
+
+double
+__div_set(double x, double y, int *pe) {
+	extern void _putsw(), _getsw();
+	int sw;
+	double z;
+
+	_putsw(0);
+	z = x / y;
+	_getsw(&sw);
+	if ((sw & 0x3f) == 0) {
+		*pe = 0;
+	} else {
+		*pe = 1;
+	}
+	return (z);
+}
+
+ */
+	
+	ENTRY(__div_set)
+	subl	$0x8, %esp	/* Give us an extra 8 bytes to play with. */
+	/* Clear the floating point exception register. */
+	fnclex			/* Equivalent of _putsw(0); */
+	
+	fldl	0xc(%esp)	/* Load up x */
+	fdivl	0x14(%esp)	/* And divide! */
+	
+	/* Check to see if the divide caused any exceptions. */
+	fstsw	(%esp)		/* Equivalent of... */
+	xorl	%edx, %edx
+	andl	$0x3f, (%esp)	/* If the status word (low bits) are zero... */
+	setne	%dl		/* ... set *pe (aka. (%eax)) accordingly. */
+	movl	0x1c(%esp), %eax/* Get pe. */
+	movl	%edx, (%eax)	/* And set it.  (True == FP exception). */
+	addl	$0x8, %esp	/* Release the 8 play bytes. */
+	ret
+	SET_SIZE(__div_set)
+
+/* double __dabs(double *d) - Get the abs. value of *d.  Straightforward. */
+
+	ENTRY(__dabs)
+	movl	0x4(%esp), %eax
+	fldl	(%eax)
+	fabs			/* Just let the FPU do its thing. */
+	ret
+	SET_SIZE(__dabs)
diff --git a/usr/src/lib/libc/i386/fp/_base_il.c b/usr/src/lib/libc/i386/fp/ieee.c
index a821932d41..3e2f4c950c 100644
--- a/usr/src/lib/libc/i386/fp/_base_il.c
+++ b/usr/src/lib/libc/i386/fp/ieee.c
@@ -24,61 +24,9 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include "lint.h"
 #include "base_conversion.h"
 
-/* The following should be coded as inline expansion templates.	 */
-
-/*
- * Multiplies two normal or subnormal doubles, returns result and exceptions.
- */
-double
-__mul_set(double x, double y, int *pe) {
-	extern void _putsw(), _getsw();
-	int sw;
-	double z;
-
-	_putsw(0);
-	z = x * y;
-	_getsw(&sw);
-	if ((sw & 0x3f) == 0) {
-		*pe = 0;
-	} else {
-		/* Result may not be exact. */
-		*pe = 1;
-	}
-	return (z);
-}
-
-/*
- * Divides two normal or subnormal doubles x/y, returns result and exceptions.
- */
-double
-__div_set(double x, double y, int *pe) {
-	extern void _putsw(), _getsw();
-	int sw;
-	double z;
-
-	_putsw(0);
-	z = x / y;
-	_getsw(&sw);
-	if ((sw & 0x3f) == 0) {
-		*pe = 0;
-	} else {
-		*pe = 1;
-	}
-	return (z);
-}
-
-double
-__dabs(double *d)
-{
-	/* should use hardware fabs instruction */
-	return ((*d < 0.0) ? -*d : *d);
-}
-
 /*
  * Returns IEEE mode/status and
  * sets up standard environment for base conversion.