mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-23 19:23:23 +01:00
8564d2aa2a
This implements a small enhancement to https://reviews.llvm.org/D55506 Specifically, while we were able to match strict FP nodes for floating-point extend operations with a register as source, this did not work for operations with memory as source. That is because from regular operations, this is represented as a combined "extload" node (which is a variant of a load SD node); but there is no equivalent using a strict FP operation. However, it turns out that even in the absence of an extload node, we can still just match the operations explicitly, e.g. (strict_fpextend (f32 (load node:$ptr)) This patch implements that method to match the LDEB/LXEB/LXDB SystemZ instructions even when the extend uses a strict-FP node. llvm-svn: 364450
99 lines
3.1 KiB
LLVM
99 lines
3.1 KiB
LLVM
; Test strict extensions of f32 to f128.
|
|
;
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
|
|
|
|
declare fp128 @llvm.experimental.constrained.fpext.f128.f32(float, metadata)
|
|
|
|
; Check register extension.
|
|
define void @f1(fp128 *%dst, float %val) {
|
|
; CHECK-LABEL: f1:
|
|
; CHECK: lxebr %f0, %f0
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: std %f2, 8(%r2)
|
|
; CHECK: br %r14
|
|
%res = call fp128 @llvm.experimental.constrained.fpext.f128.f32(float %val,
|
|
metadata !"fpexcept.strict")
|
|
store fp128 %res, fp128 *%dst
|
|
ret void
|
|
}
|
|
|
|
; Check the low end of the LXEB range.
|
|
define void @f2(fp128 *%dst, float *%ptr) {
|
|
; CHECK-LABEL: f2:
|
|
; CHECK: lxeb %f0, 0(%r3)
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: std %f2, 8(%r2)
|
|
; CHECK: br %r14
|
|
%val = load float, float *%ptr
|
|
%res = call fp128 @llvm.experimental.constrained.fpext.f128.f32(float %val,
|
|
metadata !"fpexcept.strict")
|
|
store fp128 %res, fp128 *%dst
|
|
ret void
|
|
}
|
|
|
|
; Check the high end of the aligned LXEB range.
|
|
define void @f3(fp128 *%dst, float *%base) {
|
|
; CHECK-LABEL: f3:
|
|
; CHECK: lxeb %f0, 4092(%r3)
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: std %f2, 8(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr float, float *%base, i64 1023
|
|
%val = load float, float *%ptr
|
|
%res = call fp128 @llvm.experimental.constrained.fpext.f128.f32(float %val,
|
|
metadata !"fpexcept.strict")
|
|
store fp128 %res, fp128 *%dst
|
|
ret void
|
|
}
|
|
|
|
; Check the next word up, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define void @f4(fp128 *%dst, float *%base) {
|
|
; CHECK-LABEL: f4:
|
|
; CHECK: aghi %r3, 4096
|
|
; CHECK: lxeb %f0, 0(%r3)
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: std %f2, 8(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr float, float *%base, i64 1024
|
|
%val = load float, float *%ptr
|
|
%res = call fp128 @llvm.experimental.constrained.fpext.f128.f32(float %val,
|
|
metadata !"fpexcept.strict")
|
|
store fp128 %res, fp128 *%dst
|
|
ret void
|
|
}
|
|
|
|
; Check negative displacements, which also need separate address logic.
|
|
define void @f5(fp128 *%dst, float *%base) {
|
|
; CHECK-LABEL: f5:
|
|
; CHECK: aghi %r3, -4
|
|
; CHECK: lxeb %f0, 0(%r3)
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: std %f2, 8(%r2)
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr float, float *%base, i64 -1
|
|
%val = load float, float *%ptr
|
|
%res = call fp128 @llvm.experimental.constrained.fpext.f128.f32(float %val,
|
|
metadata !"fpexcept.strict")
|
|
store fp128 %res, fp128 *%dst
|
|
ret void
|
|
}
|
|
|
|
; Check that LXEB allows indices.
|
|
define void @f6(fp128 *%dst, float *%base, i64 %index) {
|
|
; CHECK-LABEL: f6:
|
|
; CHECK: sllg %r1, %r4, 2
|
|
; CHECK: lxeb %f0, 400(%r1,%r3)
|
|
; CHECK: std %f0, 0(%r2)
|
|
; CHECK: std %f2, 8(%r2)
|
|
; CHECK: br %r14
|
|
%ptr1 = getelementptr float, float *%base, i64 %index
|
|
%ptr2 = getelementptr float, float *%ptr1, i64 100
|
|
%val = load float, float *%ptr2
|
|
%res = call fp128 @llvm.experimental.constrained.fpext.f128.f32(float %val,
|
|
metadata !"fpexcept.strict")
|
|
store fp128 %res, fp128 *%dst
|
|
ret void
|
|
}
|
|
|