1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2022 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; This test checks that prologue optimizations in GenXFuncBaling pass does not produce GEPs when
; folding bitcasts in global store/load instructions operating with SEVs.
; RUN: %opt %use_old_pass_manager% -GenXFuncBaling -march=genx64 -mcpu=Gen9 -mtriple=spir64-unknown-unknown -S < %s | FileCheck %s
; CHECK-NOT: getelementptr
@sevgv = internal global <1 x i32> undef, align 4 #0
define dllexport spir_kernel void @test() local_unnamed_addr #1 {
entry:
%gload.i = load volatile <1 x i32>, <1 x i32>* @sevgv
%sev.cast.1 = bitcast <1 x i32> %gload.i to i32
%add.i = add i32 %sev.cast.1, 1
%sev.cast.2 = bitcast i32 %add.i to <1 x i32>
store volatile <1 x i32> %sev.cast.2, <1 x i32>* @sevgv
ret void
}
attributes #0 = { "VCByteOffset"="0" "VCGlobalVariable" "VCSingleElementVector"="0" "VCVolatile" "genx_byte_offset"="0" "genx_volatile" }
attributes #1 = { noinline nounwind "CMGenxMain" }
|
