mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-25 04:02:41 +01:00
f10d5c00fc
float t1(int argc) {
return (argc == 1123) ? 1.234f : 2.38213f;
}
We would generate truly awful code on ARM (those with a weak stomach should look
away):
_t1:
movw r1, #1123
movs r2, #1
movs r3, #0
cmp r0, r1
mov.w r0, #0
it eq
moveq r0, r2
movs r1, #4
cmp r0, #0
it ne
movne r3, r1
adr r0, #LCPI1_0
ldr r0, [r0, r3]
bx lr
The problem was that legalization was creating a cascade of SELECT_CC nodes, for
for the comparison of "argc == 1123" which was fed into a SELECT node for the ?:
statement which was itself converted to a SELECT_CC node. This is because the
ARM back-end doesn't have custom lowering for SELECT nodes, so it used the
default "Expand".
I added a fairly simple "LowerSELECT" to the ARM back-end. It takes care of this
testcase, but can obviously be expanded to include more cases.
Now we generate this, which looks optimal to me:
_t1:
movw r1, #1123
movs r2, #0
cmp r0, r1
adr r0, #LCPI0_0
it eq
moveq r2, #4
ldr r0, [r0, r2]
bx lr
.align 2
LCPI0_0:
.long 1075344593 @ float 2.382130e+00
.long 1067316150 @ float 1.234000e+00
llvm-svn: 110799
93 lines
2.4 KiB
LLVM
93 lines
2.4 KiB
LLVM
; RUN: llc < %s -march=arm | FileCheck %s
|
|
; RUN: llc < %s -march=arm -mattr=+vfp2 | FileCheck %s --check-prefix=CHECK-VFP
|
|
; RUN: llc < %s -mattr=+neon,+thumb2 -mtriple=thumbv7-apple-darwin | FileCheck %s --check-prefix=CHECK-NEON
|
|
|
|
define i32 @f1(i32 %a.s) {
|
|
;CHECK: f1:
|
|
;CHECK: moveq
|
|
entry:
|
|
%tmp = icmp eq i32 %a.s, 4
|
|
%tmp1.s = select i1 %tmp, i32 2, i32 3
|
|
ret i32 %tmp1.s
|
|
}
|
|
|
|
define i32 @f2(i32 %a.s) {
|
|
;CHECK: f2:
|
|
;CHECK: movgt
|
|
entry:
|
|
%tmp = icmp sgt i32 %a.s, 4
|
|
%tmp1.s = select i1 %tmp, i32 2, i32 3
|
|
ret i32 %tmp1.s
|
|
}
|
|
|
|
define i32 @f3(i32 %a.s, i32 %b.s) {
|
|
;CHECK: f3:
|
|
;CHECK: movlt
|
|
entry:
|
|
%tmp = icmp slt i32 %a.s, %b.s
|
|
%tmp1.s = select i1 %tmp, i32 2, i32 3
|
|
ret i32 %tmp1.s
|
|
}
|
|
|
|
define i32 @f4(i32 %a.s, i32 %b.s) {
|
|
;CHECK: f4:
|
|
;CHECK: movle
|
|
entry:
|
|
%tmp = icmp sle i32 %a.s, %b.s
|
|
%tmp1.s = select i1 %tmp, i32 2, i32 3
|
|
ret i32 %tmp1.s
|
|
}
|
|
|
|
define i32 @f5(i32 %a.u, i32 %b.u) {
|
|
;CHECK: f5:
|
|
;CHECK: movls
|
|
entry:
|
|
%tmp = icmp ule i32 %a.u, %b.u
|
|
%tmp1.s = select i1 %tmp, i32 2, i32 3
|
|
ret i32 %tmp1.s
|
|
}
|
|
|
|
define i32 @f6(i32 %a.u, i32 %b.u) {
|
|
;CHECK: f6:
|
|
;CHECK: movhi
|
|
entry:
|
|
%tmp = icmp ugt i32 %a.u, %b.u
|
|
%tmp1.s = select i1 %tmp, i32 2, i32 3
|
|
ret i32 %tmp1.s
|
|
}
|
|
|
|
define double @f7(double %a, double %b) {
|
|
;CHECK: f7:
|
|
;CHECK: movlt
|
|
;CHECK: movlt
|
|
;CHECK-VFP: f7:
|
|
;CHECK-VFP: vmovmi
|
|
%tmp = fcmp olt double %a, 1.234e+00
|
|
%tmp1 = select i1 %tmp, double -1.000e+00, double %b
|
|
ret double %tmp1
|
|
}
|
|
|
|
; <rdar://problem/7260094>
|
|
;
|
|
; We used to generate really horrible code for this function. The main cause was
|
|
; a lack of a custom lowering routine for an ISD::SELECT. This would result in
|
|
; two "it" blocks in the code: one for the "icmp" and another to move the index
|
|
; into the constant pool based on the value of the "icmp". If we have one "it"
|
|
; block generated, odds are good that we have close to the ideal code for this:
|
|
;
|
|
; CHECK-NEON: _f8:
|
|
; CHECK-NEON: movw [[REGISTER_1:r[0-9]+]], #1123
|
|
; CHECK-NEON-NEXT: movs [[REGISTER_2:r[0-9]+]], #0
|
|
; CHECK-NEON-NEXT: cmp r0, [[REGISTER_1]]
|
|
; CHECK-NEON-NEXT: adr [[REGISTER_3:r[0-9]+]], #LCPI
|
|
; CHECK-NEON-NEXT: it eq
|
|
; CHECK-NEON-NEXT: moveq [[REGISTER_2]], #4
|
|
; CHECK-NEON-NEXT: ldr
|
|
; CHECK-NEON: bx
|
|
|
|
define arm_apcscc float @f8(i32 %a) nounwind {
|
|
%tmp = icmp eq i32 %a, 1123
|
|
%tmp1 = select i1 %tmp, float 0x3FF3BE76C0000000, float 0x40030E9A20000000
|
|
ret float %tmp1
|
|
}
|