Displaying 2 results from an estimated 2 matches for "switchdest13".
2018 Dec 05
2
Strange regalloc behaviour: one more available register causes much worse allocation
...lloc.txt, the same log but with the patch applied.
Note that the patch is not correct, but it happens to be a useful way of
provoking the problem.
Without the patch generating assembly with llc -mcpu=cortex-a57 everything looks
fine, but with the patch we get this (which comes from the block
bb.17.switchdest13):
.LBB0_16:
mov x29, x24
mov w24, w20
mov w20, w19
mov w19, w7
mov w7, w6
mov w6, w5
mov w5, w2
mov x2, x18
mov w18, w15
orr w15, wzr, #0x1c
str w15, [x8, #8]
mov w0, wzr
mov w15, w18
mov x18, x2
mov w2, w5
mov w5, w6
mov w6, w7
mov w7, w19
mov w19, w20
mov w20, w24
mov x24, x...
2018 Dec 05
3
Strange regalloc behaviour: one more available register causes much worse allocation
...lloc.txt, the same log but with the patch applied.
Note that the patch is not correct, but it happens to be a useful way of
provoking the problem.
Without the patch generating assembly with llc -mcpu=cortex-a57 everything looks
fine, but with the patch we get this (which comes from the block
bb.17.switchdest13):
.LBB0_16:
mov x29, x24
mov w24, w20
mov w20, w19
mov w19, w7
mov w7, w6
mov w6, w5
mov w5, w2
mov x2, x18
mov w18, w15
orr w15, wzr, #0x1c
str w15, [x8, #8]...