mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-10-30 23:42:52 +01:00
b3ecd3b03e
System z branches have a mask to select which of the 4 CC values should cause the branch to be taken. We can invert a branch by inverting the mask. However, not all instructions can produce all 4 CC values, so inverting the branch like this can lead to some oddities. For example, integer comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater). If an integer EQ is reversed to NE before instruction selection, the branch will test for 1 or 2. If instead the branch is reversed after instruction selection (by inverting the mask), it will test for 1, 2 or 3. Both are correct, but the second isn't really canonical. This patch therefore keeps track of which CC values are possible and uses this when inverting a mask. Although this is mostly cosmestic, it fixes undefined behavior for the CIJNLH in branch-08.ll. Another fix would have been to mask out bit 0 when generating the fused compare and branch, but the point of this patch is that we shouldn't need to do that in the first place. The patch also makes it easier to reuse CC results from other instructions. llvm-svn: 187495
214 lines
5.7 KiB
LLVM
214 lines
5.7 KiB
LLVM
; Test f64 conditional stores that are presented as selects.
|
|
;
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
|
|
|
|
declare void @foo(double *)
|
|
|
|
; Test with the loaded value first.
|
|
define void @f1(double *%ptr, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f1:
|
|
; CHECK-NOT: %r2
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r2
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: br %r14
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; ...and with the loaded value second
|
|
define void @f2(double *%ptr, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f2:
|
|
; CHECK-NOT: %r2
|
|
; CHECK: jhe [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r2
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: br %r14
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %alt, double %orig
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; Check the high end of the aligned STD range.
|
|
define void @f3(double *%base, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f3:
|
|
; CHECK-NOT: %r2
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r2
|
|
; CHECK: std %f0, 4088(%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double *%base, i64 511
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; Check the next doubleword up, which should use STDY instead of STD.
|
|
define void @f4(double *%base, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f4:
|
|
; CHECK-NOT: %r2
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r2
|
|
; CHECK: stdy %f0, 4096(%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double *%base, i64 512
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; Check the high end of the aligned STDY range.
|
|
define void @f5(double *%base, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f5:
|
|
; CHECK-NOT: %r2
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r2
|
|
; CHECK: stdy %f0, 524280(%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double *%base, i64 65535
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; Check the next doubleword up, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define void @f6(double *%base, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f6:
|
|
; CHECK-NOT: %r2
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r2
|
|
; CHECK: agfi %r2, 524288
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double *%base, i64 65536
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; Check the low end of the STDY range.
|
|
define void @f7(double *%base, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f7:
|
|
; CHECK-NOT: %r2
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r2
|
|
; CHECK: stdy %f0, -524288(%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double *%base, i64 -65536
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; Check the next doubleword down, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define void @f8(double *%base, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f8:
|
|
; CHECK-NOT: %r2
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r2
|
|
; CHECK: agfi %r2, -524296
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double *%base, i64 -65537
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; Check that STDY allows an index.
|
|
define void @f9(i64 %base, i64 %index, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f9:
|
|
; CHECK-NOT: %r2
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r2
|
|
; CHECK: stdy %f0, 524287(%r3,%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: br %r14
|
|
%add1 = add i64 %base, %index
|
|
%add2 = add i64 %add1, 524287
|
|
%ptr = inttoptr i64 %add2 to double *
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; Check that volatile loads are not matched.
|
|
define void @f10(double *%ptr, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f10:
|
|
; CHECK: ld {{%f[0-5]}}, 0(%r2)
|
|
; CHECK: {{jl|jnl}} [[LABEL:[^ ]*]]
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: std {{%f[0-5]}}, 0(%r2)
|
|
; CHECK: br %r14
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load volatile double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; ...likewise stores. In this case we should have a conditional load into %f0.
|
|
define void @f11(double *%ptr, double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f11:
|
|
; CHECK: jhe [[LABEL:[^ ]*]]
|
|
; CHECK: ld %f0, 0(%r2)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: br %r14
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store volatile double %res, double *%ptr
|
|
ret void
|
|
}
|
|
|
|
; Try a frame index base.
|
|
define void @f12(double %alt, i32 %limit) {
|
|
; CHECK-LABEL: f12:
|
|
; CHECK: brasl %r14, foo@PLT
|
|
; CHECK-NOT: %r15
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
; CHECK-NOT: %r15
|
|
; CHECK: std {{%f[0-9]+}}, {{[0-9]+}}(%r15)
|
|
; CHECK: [[LABEL]]:
|
|
; CHECK: brasl %r14, foo@PLT
|
|
; CHECK: br %r14
|
|
%ptr = alloca double
|
|
call void @foo(double *%ptr)
|
|
%cond = icmp ult i32 %limit, 42
|
|
%orig = load double *%ptr
|
|
%res = select i1 %cond, double %orig, double %alt
|
|
store double %res, double *%ptr
|
|
call void @foo(double *%ptr)
|
|
ret void
|
|
}
|