mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-10-30 15:32:52 +01:00
c7495a0fca
Add a mapping from register-based <INSN>R instructions to the corresponding memory-based <INSN>. Use it to cut down on the number of spill loads. Some instructions extend their operands from smaller fields, so this required a new TSFlags field to say how big the unextended operand is. This optimisation doesn't trigger for C(G)R and CL(G)R because in practice we always combine those instructions with a branch. Adding a test for every other case probably seems excessive, but it did catch a missed optimisation for DSGF (fixed in r185435). llvm-svn: 185529
175 lines
4.1 KiB
LLVM
175 lines
4.1 KiB
LLVM
; Test 32-bit ANDs in which the second operand is variable.
|
|
;
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
|
|
|
|
declare i32 @foo()
|
|
|
|
; Check NR.
|
|
define i32 @f1(i32 %a, i32 %b) {
|
|
; CHECK: f1:
|
|
; CHECK: nr %r2, %r3
|
|
; CHECK: br %r14
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check the low end of the N range.
|
|
define i32 @f2(i32 %a, i32 *%src) {
|
|
; CHECK: f2:
|
|
; CHECK: n %r2, 0(%r3)
|
|
; CHECK: br %r14
|
|
%b = load i32 *%src
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check the high end of the aligned N range.
|
|
define i32 @f3(i32 %a, i32 *%src) {
|
|
; CHECK: f3:
|
|
; CHECK: n %r2, 4092(%r3)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i32 *%src, i64 1023
|
|
%b = load i32 *%ptr
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check the next word up, which should use NY instead of N.
|
|
define i32 @f4(i32 %a, i32 *%src) {
|
|
; CHECK: f4:
|
|
; CHECK: ny %r2, 4096(%r3)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i32 *%src, i64 1024
|
|
%b = load i32 *%ptr
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check the high end of the aligned NY range.
|
|
define i32 @f5(i32 %a, i32 *%src) {
|
|
; CHECK: f5:
|
|
; CHECK: ny %r2, 524284(%r3)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i32 *%src, i64 131071
|
|
%b = load i32 *%ptr
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check the next word up, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define i32 @f6(i32 %a, i32 *%src) {
|
|
; CHECK: f6:
|
|
; CHECK: agfi %r3, 524288
|
|
; CHECK: n %r2, 0(%r3)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i32 *%src, i64 131072
|
|
%b = load i32 *%ptr
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check the high end of the negative aligned NY range.
|
|
define i32 @f7(i32 %a, i32 *%src) {
|
|
; CHECK: f7:
|
|
; CHECK: ny %r2, -4(%r3)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i32 *%src, i64 -1
|
|
%b = load i32 *%ptr
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check the low end of the NY range.
|
|
define i32 @f8(i32 %a, i32 *%src) {
|
|
; CHECK: f8:
|
|
; CHECK: ny %r2, -524288(%r3)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i32 *%src, i64 -131072
|
|
%b = load i32 *%ptr
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check the next word down, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define i32 @f9(i32 %a, i32 *%src) {
|
|
; CHECK: f9:
|
|
; CHECK: agfi %r3, -524292
|
|
; CHECK: n %r2, 0(%r3)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i32 *%src, i64 -131073
|
|
%b = load i32 *%ptr
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check that N allows an index.
|
|
define i32 @f10(i32 %a, i64 %src, i64 %index) {
|
|
; CHECK: f10:
|
|
; CHECK: n %r2, 4092({{%r4,%r3|%r3,%r4}})
|
|
; CHECK: br %r14
|
|
%add1 = add i64 %src, %index
|
|
%add2 = add i64 %add1, 4092
|
|
%ptr = inttoptr i64 %add2 to i32 *
|
|
%b = load i32 *%ptr
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check that NY allows an index.
|
|
define i32 @f11(i32 %a, i64 %src, i64 %index) {
|
|
; CHECK: f11:
|
|
; CHECK: ny %r2, 4096({{%r4,%r3|%r3,%r4}})
|
|
; CHECK: br %r14
|
|
%add1 = add i64 %src, %index
|
|
%add2 = add i64 %add1, 4096
|
|
%ptr = inttoptr i64 %add2 to i32 *
|
|
%b = load i32 *%ptr
|
|
%and = and i32 %a, %b
|
|
ret i32 %and
|
|
}
|
|
|
|
; Check that ANDs of spilled values can use N rather than NR.
|
|
define i32 @f12(i32 *%ptr0) {
|
|
; CHECK: f12:
|
|
; CHECK: brasl %r14, foo@PLT
|
|
; CHECK: n %r2, 16{{[04]}}(%r15)
|
|
; CHECK: br %r14
|
|
%ptr1 = getelementptr i32 *%ptr0, i64 2
|
|
%ptr2 = getelementptr i32 *%ptr0, i64 4
|
|
%ptr3 = getelementptr i32 *%ptr0, i64 6
|
|
%ptr4 = getelementptr i32 *%ptr0, i64 8
|
|
%ptr5 = getelementptr i32 *%ptr0, i64 10
|
|
%ptr6 = getelementptr i32 *%ptr0, i64 12
|
|
%ptr7 = getelementptr i32 *%ptr0, i64 14
|
|
%ptr8 = getelementptr i32 *%ptr0, i64 16
|
|
%ptr9 = getelementptr i32 *%ptr0, i64 18
|
|
|
|
%val0 = load i32 *%ptr0
|
|
%val1 = load i32 *%ptr1
|
|
%val2 = load i32 *%ptr2
|
|
%val3 = load i32 *%ptr3
|
|
%val4 = load i32 *%ptr4
|
|
%val5 = load i32 *%ptr5
|
|
%val6 = load i32 *%ptr6
|
|
%val7 = load i32 *%ptr7
|
|
%val8 = load i32 *%ptr8
|
|
%val9 = load i32 *%ptr9
|
|
|
|
%ret = call i32 @foo()
|
|
|
|
%and0 = and i32 %ret, %val0
|
|
%and1 = and i32 %and0, %val1
|
|
%and2 = and i32 %and1, %val2
|
|
%and3 = and i32 %and2, %val3
|
|
%and4 = and i32 %and3, %val4
|
|
%and5 = and i32 %and4, %val5
|
|
%and6 = and i32 %and5, %val6
|
|
%and7 = and i32 %and6, %val7
|
|
%and8 = and i32 %and7, %val8
|
|
%and9 = and i32 %and8, %val9
|
|
|
|
ret i32 %and9
|
|
}
|