mirror of
https://github.com/RPCSX/llvm.git
synced 2024-12-03 17:31:50 +00:00
cf0fa9b9dd
The z13 vector facility includes some instructions that operate only on the high f64 in a v2f64, effectively extending the FP register set from 16 to 32 registers. It's still better to use the old instructions if the operands happen to fit though, since the older instructions have a shorter encoding. Based on a patch by Richard Sandiford. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236524 91177308-0d34-0410-b5e6-96231b3b80d8
112 lines
2.9 KiB
LLVM
112 lines
2.9 KiB
LLVM
; Test 64-bit floating-point loads.
|
|
;
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | FileCheck %s
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
|
|
|
|
; Test the low end of the LD range.
|
|
define double @f1(double *%src) {
|
|
; CHECK-LABEL: f1:
|
|
; CHECK: ld %f0, 0(%r2)
|
|
; CHECK: br %r14
|
|
%val = load double , double *%src
|
|
ret double %val
|
|
}
|
|
|
|
; Test the high end of the LD range.
|
|
define double @f2(double *%src) {
|
|
; CHECK-LABEL: f2:
|
|
; CHECK: ld %f0, 4088(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double, double *%src, i64 511
|
|
%val = load double , double *%ptr
|
|
ret double %val
|
|
}
|
|
|
|
; Check the next doubleword up, which should use LDY instead of LD.
|
|
define double @f3(double *%src) {
|
|
; CHECK-LABEL: f3:
|
|
; CHECK: ldy %f0, 4096(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double, double *%src, i64 512
|
|
%val = load double , double *%ptr
|
|
ret double %val
|
|
}
|
|
|
|
; Check the high end of the aligned LDY range.
|
|
define double @f4(double *%src) {
|
|
; CHECK-LABEL: f4:
|
|
; CHECK: ldy %f0, 524280(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double, double *%src, i64 65535
|
|
%val = load double , double *%ptr
|
|
ret double %val
|
|
}
|
|
|
|
; Check the next doubleword up, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define double @f5(double *%src) {
|
|
; CHECK-LABEL: f5:
|
|
; CHECK: agfi %r2, 524288
|
|
; CHECK: ld %f0, 0(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double, double *%src, i64 65536
|
|
%val = load double , double *%ptr
|
|
ret double %val
|
|
}
|
|
|
|
; Check the high end of the negative aligned LDY range.
|
|
define double @f6(double *%src) {
|
|
; CHECK-LABEL: f6:
|
|
; CHECK: ldy %f0, -8(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double, double *%src, i64 -1
|
|
%val = load double , double *%ptr
|
|
ret double %val
|
|
}
|
|
|
|
; Check the low end of the LDY range.
|
|
define double @f7(double *%src) {
|
|
; CHECK-LABEL: f7:
|
|
; CHECK: ldy %f0, -524288(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double, double *%src, i64 -65536
|
|
%val = load double , double *%ptr
|
|
ret double %val
|
|
}
|
|
|
|
; Check the next doubleword down, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define double @f8(double *%src) {
|
|
; CHECK-LABEL: f8:
|
|
; CHECK: agfi %r2, -524296
|
|
; CHECK: ld %f0, 0(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr double, double *%src, i64 -65537
|
|
%val = load double , double *%ptr
|
|
ret double %val
|
|
}
|
|
|
|
; Check that LD allows an index.
|
|
define double @f9(i64 %src, i64 %index) {
|
|
; CHECK-LABEL: f9:
|
|
; CHECK: ld %f0, 4095({{%r3,%r2|%r2,%r3}})
|
|
; CHECK: br %r14
|
|
%add1 = add i64 %src, %index
|
|
%add2 = add i64 %add1, 4095
|
|
%ptr = inttoptr i64 %add2 to double *
|
|
%val = load double , double *%ptr
|
|
ret double %val
|
|
}
|
|
|
|
; Check that LDY allows an index.
|
|
define double @f10(i64 %src, i64 %index) {
|
|
; CHECK-LABEL: f10:
|
|
; CHECK: ldy %f0, 4096({{%r3,%r2|%r2,%r3}})
|
|
; CHECK: br %r14
|
|
%add1 = add i64 %src, %index
|
|
%add2 = add i64 %add1, 4096
|
|
%ptr = inttoptr i64 %add2 to double *
|
|
%val = load double , double *%ptr
|
|
ret double %val
|
|
}
|