The attached test case contains several continuous bit fields assignment,
compile it with options -mthumb -Os, gcc generates
.fnstart
.LFB1:
ldrb r3, [r0]
mov r2, #7
bic r3, r3, r2
strb r3, [r0]
ldrb r3, [r0]
mov r2, #56
bic r3, r3, r2
mov r2, #8
orr r3, r3, r2
strb r3, [r0]
ldrb r3, [r0]
mov r2, #64
bic r3, r3, r2
strb r3, [r0]
ldrb r2, [r0]
mov r3, #127
and r3, r3, r2
strb r3, [r0]
@ sp needed for prologue
bx lr
The 4 fields are contained in one word, for each field assignment the code
loads the word, changes the field, then write the word back. A better code
sequence should load the word once, change all 4 fields, then write back the
changed word.
ldrb r3, [r0]
mov r2, #255 // bit mask
bic r3, r3, r2
mov r2, #8 // the new value of all 4 fields
orr r3, r3, r2
strb r3, [r0]
@ sp needed for prologue
bx lr
or more aggressively if the word contains only these four fields
mov r3, #8
strb r3, [r0]
@ sp needed for prologue
bx lr
--
Summary: inefficient bit fields assignments
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: carrot at google dot com
GCC build triplet: i686-linux
GCC host triplet: i686-linux
GCC target triplet: arm-eabi
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42172
