diff options
author | xleroy <xleroy@fca1b0fc-160b-0410-b1d3-a4f43f01ea2e> | 2013-05-05 16:29:20 +0000 |
---|---|---|
committer | xleroy <xleroy@fca1b0fc-160b-0410-b1d3-a4f43f01ea2e> | 2013-05-05 16:29:20 +0000 |
commit | b257a6d283f6f5784cb351856b5dbe8c645a1f6f (patch) | |
tree | 3e134e46c53a5e75a3b6c78878d07a554b396367 /runtime/arm/i64_dtos.S | |
parent | bc24cf49659f91245d8f42ca06fbe7d21a5c06cd (diff) | |
download | compcert-kvx-b257a6d283f6f5784cb351856b5dbe8c645a1f6f.tar.gz compcert-kvx-b257a6d283f6f5784cb351856b5dbe8c645a1f6f.zip |
ia32/i64_dtou: wrong play on rounding mode
arm, powerpc: expand shifts inline in dtos and dtou
arm: branchless code for shl and shr
test: more tests for double -> long long conversions.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2234 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
Diffstat (limited to 'runtime/arm/i64_dtos.S')
-rw-r--r-- | runtime/arm/i64_dtos.S | 33 |
1 files changed, 19 insertions, 14 deletions
diff --git a/runtime/arm/i64_dtos.S b/runtime/arm/i64_dtos.S index d0ef0371..43748683 100644 --- a/runtime/arm/i64_dtos.S +++ b/runtime/arm/i64_dtos.S @@ -40,8 +40,7 @@ .global __i64_dtos __i64_dtos: - push {r4, lr} - mov r4, r1, asr #31 @ save sign in r4 + mov r12, r1, asr #31 @ save sign of result in r12 @ extract unbiased exponent ((HI & 0x7FF00000) >> 20) - (1023 + 52) in r2 @ note: 1023 + 52 = 1075 = 1024 + 51 @ note: (HI & 0x7FF00000) >> 20 = (HI << 1) >> 21 @@ -61,31 +60,37 @@ __i64_dtos: @ shift it appropriately cmp r2, #0 blt 3f - bl __i64_shl @ if EXP >= 0, shift left by EXP - b 4f -3: rsb r2, r2, #0 - bl __i64_shr @ otherwise, shift right by -EXP + @ EXP >= 0: shift left by EXP. Note that EXP < 12 + rsb r3, r2, #32 @ r3 = 32 - amount + mov r1, r1, lsl r2 + orr r1, r1, r0, lsr r3 + mov r0, r0, lsl r2 + b 4f + @ EXP < 0: shift right by -EXP. Note that -EXP <= 52 but can be >= 32 +3: rsb r2, r2, #0 @ r2 = -EXP - shift amount + rsb r3, r2, #32 @ r3 = 32 - amount + mov r0, r0, lsr r2 + orr r0, r0, r1, lsl r3 + sub r3, r2, #32 @ r3 = amount - 32 (see i64_shr.s) + orr r0, r0, r1, lsr r3 + mov r1, r1, lsr r2 @ apply sign to result -4: eor r0, r0, r4 - eor r1, r1, r4 - subs r0, r0, r4 - sbc r1, r1, r4 - pop {r4, lr} +4: eor r0, r0, r12 + eor r1, r1, r12 + subs r0, r0, r12 + sbc r1, r1, r12 bx lr @ special cases 1: mov r0, #0 @ result is 0 mov r1, #0 - pop {r4, lr} bx lr 2: cmp r4, #0 blt 6f mvn r0, #0 @ result is 0x7F....FF (MAX_SINT) mov r1, r0, lsr #1 - pop {r4, lr} bx lr 6: mov r0, #0 @ result is 0x80....00 (MIN_SINT) mov r1, #0x80000000 - pop {r4, lr} bx lr .type __i64_dtos, %function .size __i64_dtos, . - __i64_dtos |