Merge changes between r21775 and r21911 from /fsf/trunk.

git-svn-id: svn://svn.eglibc.org/trunk@21912 7b3dc134-2b1b-0410-93df-9e9f96275f8d

2 files changed, 71 insertions, 36 deletions

diff --git a/libc/sysdeps/i386/fpu/e_powl.S b/libc/sysdeps/i386/fpu/e_powl.S
index ac4842cf6..7e297756f 100644
--- a/libc/sysdeps/i386/fpu/e_powl.S
+++ b/libc/sysdeps/i386/fpu/e_powl.S
@@ -26,9 +26,9 @@
 	.type one,@object
 one:	.double 1.0
 	ASM_SIZE_DIRECTIVE(one)
-	.type limit,@object
-limit:	.double 0.29
-	ASM_SIZE_DIRECTIVE(limit)
+	.type p3,@object
+p3:	.byte 0, 0, 0, 0, 0, 0, 0x20, 0x40
+	ASM_SIZE_DIRECTIVE(p3)
 	.type p63,@object
 p63:	.byte 0, 0, 0, 0, 0, 0, 0xe0, 0x43
 	ASM_SIZE_DIRECTIVE(p63)
@@ -141,7 +141,15 @@ ENTRY(__ieee754_powl)
 	fchs			// -0x1p-79 : x
 	jmp	3f
 
-9:	/* OK, we have an integer value for y.  */
+9:	/* OK, we have an integer value for y.  Unless very small
+	   (we use < 8), use the algorithm for real exponent to avoid
+	   accumulation of errors.  */
+	fld	%st		// y : y : x
+	fabs			// |y| : y : x
+	fcompl	MO(p3)		// y : x
+	fnstsw
+	sahf
+	jnc	2f
 	popl	%eax
 	cfi_adjust_cfa_offset (-4)
 	popl	%edx
@@ -182,7 +190,7 @@ ENTRY(__ieee754_powl)
 
 	cfi_adjust_cfa_offset (8)
 	.align ALIGNARG(4)
-2:	// y is a large integer (absolute value at least 1L<<63), but
+2:	// y is a large integer (absolute value at least 8), but
 	// may be odd unless at least 1L<<64.  So it may be necessary
 	// to adjust the sign of a negative result afterwards.
 	fxch			// x : y
@@ -205,34 +213,21 @@ ENTRY(__ieee754_powl)
 	fchs			// -(1L<<78) : |x|
 	.align ALIGNARG(4)
 3:	/* y is a real number.  */
-	fxch			// x : y
-	fldl	MO(one)		// 1.0 : x : y
-	fldl	MO(limit)	// 0.29 : 1.0 : x : y
-	fld	%st(2)		// x : 0.29 : 1.0 : x : y
-	fsub	%st(2)		// x-1 : 0.29 : 1.0 : x : y
-	fabs			// |x-1| : 0.29 : 1.0 : x : y
-	fucompp			// 1.0 : x : y
-	fnstsw
-	fxch			// x : 1.0 : y
-	sahf
-	ja	7f
-	fsub	%st(1)		// x-1 : 1.0 : y
-	fyl2xp1			// log2(x) : y
-	jmp	8f
-
-7:	fyl2x			// log2(x) : y
-8:	fmul	%st(1)		// y*log2(x) : y
-	fst	%st(1)		// y*log2(x) : y*log2(x)
-	frndint			// int(y*log2(x)) : y*log2(x)
-	fsubr	%st, %st(1)	// int(y*log2(x)) : fract(y*log2(x))
-	fxch			// fract(y*log2(x)) : int(y*log2(x))
-	f2xm1			// 2^fract(y*log2(x))-1 : int(y*log2(x))
-	faddl	MO(one)		// 2^fract(y*log2(x)) : int(y*log2(x))
-	fscale			// 2^fract(y*log2(x))*2^int(y*log2(x)) : int(y*log2(x))
-	fstp	%st(1)		// 2^fract(y*log2(x))*2^int(y*log2(x))
+	subl	$28, %esp
+	cfi_adjust_cfa_offset (28)
+	fstpt	12(%esp)	// x
+	fstpt	(%esp)		// <empty>
+	mov	%edx, 24(%esp)
+	call	HIDDEN_JUMPTARGET (__powl_helper)	// <result>
+	mov	24(%esp), %edx
+	addl	$28, %esp
+	cfi_adjust_cfa_offset (-28)
 	testb	$2, %dh
 	jz	292f
 	// x is negative.  If y is an odd integer, negate the result.
+#ifdef	PIC
+	LOAD_PIC_REG (cx)
+#endif
 	fldt	24(%esp)	// y : abs(result)
 	fld	%st		// y : y : abs(result)
 	fabs			// |y| : y : abs(result)
diff --git a/libc/sysdeps/i386/fpu/libm-test-ulps b/libc/sysdeps/i386/fpu/libm-test-ulps
index 239da44a7..5b595bc56 100644
--- a/libc/sysdeps/i386/fpu/libm-test-ulps
+++ b/libc/sysdeps/i386/fpu/libm-test-ulps
@@ -1339,6 +1339,9 @@ idouble: 1
 ifloat: 1
 
 # cos
+Test "cos (0x1p+50) == 8.68095904660550604334592502063501320395739e-01":
+float: 1
+ifloat: 1
 Test "cos (M_PI_6l * 2.0) == 0.5":
 double: 1
 float: 1
@@ -1817,10 +1820,10 @@ ldouble: 1
 
 # ctan
 Test "Real part of: ctan (-2 - 3 i) == 0.376402564150424829275122113032269084e-2 - 1.00323862735360980144635859782192726 i":
-float: 1
-ifloat: 1
 double: 1
+float: 1
 idouble: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 Test "Imaginary part of: ctan (-2 - 3 i) == 0.376402564150424829275122113032269084e-2 - 1.00323862735360980144635859782192726 i":
@@ -2477,6 +2480,11 @@ ifloat: 1
 ildouble: 1
 ldouble: 1
 
+# pow
+Test "pow (0x0.ffffffp0, -0x1p24) == 2.7182819094701610539628664526874952929416":
+ildouble: 1
+ldouble: 1
+
 # pow_downward
 Test "pow_downward (1.0625, 1.125) == 1.070582293028761362162622578677070098674":
 double: 1
@@ -2683,6 +2691,9 @@ Test "sincos (0x1.fffff8p+127, &sin_res, &cos_res) puts 4.8578606313048733970111
 float: 1
 Test "sincos (0x1p+127, &sin_res, &cos_res) puts 6.23385512955870240370428801097126489001833e-01 in sin_res":
 float: 1
+Test "sincos (0x1p+50, &sin_res, &cos_res) puts 8.68095904660550604334592502063501320395739e-01 in cos_res":
+float: 1
+ifloat: 1
 Test "sincos (0x1p65, &sin_res, &cos_res) puts -0.047183876212354673805106149805700013943218 in sin_res":
 float: 1
 ifloat: 1
@@ -3015,6 +3026,28 @@ ldouble: 1
 Test "y0 (0x1.ff00000000002p+840) == 1.846591691699331493194965158699937660696e-127":
 double: 1
 idouble: 1
+Test "y0 (0x1p-10) == -4.4865150767109739412411806297168793661098":
+ildouble: 1
+ldouble: 1
+Test "y0 (0x1p-30) == -1.3311940304267782826037118027401817264906e+1":
+double: 1
+idouble: 1
+ildouble: 1
+ldouble: 1
+Test "y0 (0x1p-40) == -1.7724652307320814696990854700366226762563e+1":
+double: 1
+float: 1
+idouble: 1
+ifloat: 1
+Test "y0 (0x1p-60) == -2.6550076313426878432849115782108205929120e+1":
+double: 1
+idouble: 1
+Test "y0 (0x1p-70) == -3.0962788316479910300778244424468159753887e+1":
+double: 1
+idouble: 1
+Test "y0 (0x1p-80) == -3.5375500319532942168707373066828113573541e+1":
+double: 1
+idouble: 1
 Test "y0 (1.0) == 0.0882569642156769579829267660235151628":
 double: 2
 float: 1
@@ -3056,6 +3089,9 @@ idouble: 1
 ifloat: 2
 ildouble: 1
 ldouble: 1
+Test "y1 (0x1p-10) == -6.5190099301063115047395187618929589514382e+02":
+float: 1
+ifloat: 1
 Test "y1 (1.0) == -0.781212821300288716547150000047964821":
 double: 1
 idouble: 1
@@ -3562,10 +3598,10 @@ ildouble: 1
 ldouble: 1
 
 Function: Real part of "ctan_towardzero":
-float: 1
-ifloat: 1
 double: 1
+float: 1
 idouble: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 
@@ -3636,10 +3672,10 @@ ildouble: 4
 ldouble: 4
 
 Function: Imaginary part of "ctanh_towardzero":
-float: 1
-ifloat: 1
 double: 1
+float: 1
 idouble: 1
+ifloat: 1
 ildouble: 1
 ldouble: 1
 
@@ -3751,6 +3787,10 @@ ifloat: 1
 ildouble: 1
 ldouble: 1
 
+Function: "pow":
+ildouble: 1
+ldouble: 1
+
 Function: "pow_downward":
 double: 1
 float: 1