|
|
@@ -145,45 +145,16 @@
|
|
|
function fpc_mul_qword(f1,f2 : qword) : qword;[public,alias: 'FPC_MUL_QWORD']; compilerproc;
|
|
|
assembler; nostackframe;
|
|
|
asm
|
|
|
- // (r3:r4) = (r3:r4) * (r5:r6), checkoverflow is in r7
|
|
|
+ // (r3:r4) = (r3:r4) * (r5:r6)
|
|
|
// res f1 f2
|
|
|
|
|
|
or. r10,r3,r5 // are both msw's 0?
|
|
|
mulhwu r8,r4,r6 // msw of product of lsw's
|
|
|
- not r0,r7 // if no overflowcheck, r0 := $ffffffff, else r0 := 0;
|
|
|
beq .LDone // if both msw's are zero, skip cross products
|
|
|
mullw r9,r4,r5 // lsw of first cross-product
|
|
|
- cntlzw r11,r3 // count leading zeroes of msw1
|
|
|
- cntlzw r12,r5 // count leading zeroes of msw2
|
|
|
mullw r7,r3,r6 // lsw of second cross-product
|
|
|
- add r12,r11,r12 // sum of leading zeroes
|
|
|
- mr r10,r8
|
|
|
- or r0,r12,r0 // maximise sum if no overflow checking, otherwise it remains
|
|
|
add r8,r8,r9 // add
|
|
|
- cmplwi cr1,r0,64 // >= 64 leading zero bits in total? If so, no overflow
|
|
|
add r8,r8,r7 // add
|
|
|
- bge+ cr1,.LDone // if the sum of leading zero's >= 64 (or checkoverflow was 0)
|
|
|
- // there's no overflow, otherwise more thorough check
|
|
|
- add r7,r7,r9
|
|
|
- mulhwu r3,r6,r3
|
|
|
- addc r7,r7,r10 // add the msw of the product of the lsw's, record carry
|
|
|
- cntlzw r9,r5
|
|
|
- cntlzw r10,r4 // get leading zeroes count of lsw f1
|
|
|
- mulhwu r5,r4,r5
|
|
|
- addze r3,r3
|
|
|
- subfic r0,r11,31 // if msw f1 = 0, then r0 := -1, else r0 >= 0
|
|
|
- cntlzw r7,r6
|
|
|
- subfic r11,r9,31 // same for f2
|
|
|
- srawi r0,r0,31 // if msw f1 = 0, then r0 := 1, else r0 := 0
|
|
|
- srawi r11,r11,31
|
|
|
- and r10,r10,r0 // if msw f1 <> 0, the leading zero count lsw f1 := 0
|
|
|
- and r9,r7,r11 // same for f2
|
|
|
- or. r5,r5,r3
|
|
|
- add r9,r9,r10 // add leading zero counts of lsw's to sum if appropriate
|
|
|
- add r9,r9,r12
|
|
|
- cmplwi cr7,r9,64 // is the sum now >= 64?
|
|
|
- cmplwi cr1,r9,62 // or <= 62?
|
|
|
-
|
|
|
.LDone:
|
|
|
mullw r4,r4,r6 // lsw of product of lsw's
|
|
|
mr r3,r8 // get msw of product in correct register
|
|
|
@@ -193,12 +164,11 @@
|
|
|
function fpc_mul_qword_checkoverflow(f1,f2 : qword) : qword;[public,alias: 'FPC_MUL_QWORD_CHECKOVERFLOW']; compilerproc;
|
|
|
assembler; nostackframe;
|
|
|
asm
|
|
|
- // (r3:r4) = (r3:r4) * (r5:r6), checkoverflow is in r7
|
|
|
+ // (r3:r4) = (r3:r4) * (r5:r6)
|
|
|
// res f1 f2
|
|
|
|
|
|
or. r10,r3,r5 // are both msw's 0?
|
|
|
mulhwu r8,r4,r6 // msw of product of lsw's
|
|
|
- xor r0,r0,r0 // r0 := 0 for overflow checking
|
|
|
beq .LDone // if both msw's are zero, skip cross products
|
|
|
mullw r9,r4,r5 // lsw of first cross-product
|
|
|
cntlzw r11,r3 // count leading zeroes of msw1
|
|
|
@@ -206,9 +176,8 @@
|
|
|
mullw r7,r3,r6 // lsw of second cross-product
|
|
|
add r12,r11,r12 // sum of leading zeroes
|
|
|
mr r10,r8
|
|
|
- or r0,r12,r0 // maximise sum if no overflow checking, otherwise it remains
|
|
|
add r8,r8,r9 // add
|
|
|
- cmplwi cr1,r0,64 // >= 64 leading zero bits in total? If so, no overflow
|
|
|
+ cmplwi cr1,r12,64 // >= 64 leading zero bits in total? If so, no overflow
|
|
|
add r8,r8,r7 // add
|
|
|
bge+ cr1,.LDone // if the sum of leading zero's >= 64 (or checkoverflow was 0)
|
|
|
// there's no overflow, otherwise more thorough check
|
Jonas Maebe