patch-2.4.4 linux/arch/s390/math-emu/qrnnd.S

Next file: linux/arch/s390/math-emu/sfp-util.h
Previous file: linux/arch/s390/math-emu/math.c
Back to the patch index
Back to the overall index

diff -u --recursive --new-file v2.4.3/linux/arch/s390/math-emu/qrnnd.S linux/arch/s390/math-emu/qrnnd.S
@@ -0,0 +1,77 @@
+# S/390 __udiv_qrnnd
+
+# r2 : &__r
+# r3 : upper half of 64 bit word n
+# r4 : lower half of 64 bit word n
+# r5 : divisor d
+# the reminder r of the division is to be stored to &__r and
+# the quotient q is to be returned
+
+        .text
+        .globl __udiv_qrnnd
+__udiv_qrnnd:
+        st    %r2,24(%r15)        # store pointer to reminder for later
+        lr    %r0,%r3             # reload n
+	lr    %r1,%r4
+	ltr   %r2,%r5             # reload and test divisor
+        jp    5f
+        # divisor >= 0x80000000
+	srdl  %r0,2               # n/4
+        srl   %r2,1               # d/2
+        slr   %r1,%r2             # special case if last bit of d is set
+        brc   3,0f                #  (n/4) div (n/2) can overflow by 1
+        ahi   %r0,-1              #  trick: subtract n/2, then divide
+0:      dr    %r0,%r2             # signed division
+        ahi   %r1,1               #  trick part 2: add 1 to the quotient
+        # now (n >> 2) = (d >> 1) * %r1 + %r0
+	lhi   %r3,1
+        nr    %r3,%r1             # test last bit of q
+        jz    1f
+        alr   %r0,%r2             # add (d>>1) to r
+1:      srl   %r1,1               # q >>= 1
+        # now (n >> 2) = (d&-2) * %r1 + %r0
+        lhi   %r3,1
+        nr    %r3,%r5             # test last bit of d
+        jz    2f
+        slr   %r0,%r1             # r -= q
+	brc   3,2f                # borrow ?
+	alr   %r0,%r5             # r += d
+	ahi   %r1,-1
+2:      # now (n >> 2) = d * %r1 + %r0
+        alr   %r1,%r1             # q <<= 1
+        alr   %r0,%r0             # r <<= 1
+        brc   12,3f               # overflow on r ?
+        slr   %r0,%r5             # r -= d
+        ahi   %r1,1               # q += 1
+3:      lhi   %r3,2
+        nr    %r3,%r4             # test next to last bit of n
+        jz    4f
+        ahi   %r0,1               # r += 1
+4:      clr   %r0,%r5             # r >= d ?
+        jl    6f
+        slr   %r0,%r5             # r -= d
+        ahi   %r1,1               # q += 1
+        # now (n >> 1) = d * %r1 + %r0
+        j     6f
+5:      # divisor < 0x80000000
+	srdl  %r0,1
+        dr    %r0,%r2             # signed division
+        # now (n >> 1) = d * %r1 + %r0
+6:      alr   %r1,%r1             # q <<= 1
+        alr   %r0,%r0             # r <<= 1
+        brc   12,7f               # overflow on r ?
+        slr   %r0,%r5             # r -= d
+        ahi   %r1,1               # q += 1
+7:      lhi   %r3,1
+        nr    %r3,%r4             # isolate last bit of n
+        alr   %r0,%r3             # r += (n & 1)
+        clr   %r0,%r5             # r >= d ?
+        jl    8f
+        slr   %r0,%r5             # r -= d
+        ahi   %r1,1               # q += 1
+8:      # now n = d * %r1 + %r0
+	l     %r2,24(%r15)
+        st    %r0,0(%r2)
+        lr    %r2,%r1
+        br    %r14
+	.end	__udiv_qrnnd

FUNET's LINUX-ADM group, linux-adm@nic.funet.fi
TCL-scripts by Sam Shen (who was at: slshen@lbl.gov)