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
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2025 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; REQUIRES: regkeys
; RUN: igc_opt --opaque-pointers --regkey DisableCodeScheduling=0 --regkey EnableCodeSchedulingIfNoSpills=1 \
; RUN: --regkey PrintToConsole=1 --regkey DumpCodeScheduling=1 --regkey CodeSchedulingForceMWOnly=1 \
; RUN: --igc-code-scheduling -S %s 2>&1 | FileCheck %s
; Check that the lifetime end intrinsic is not moved before the last use
; and lifetime start is not moved after the first use of the alloca
define spir_kernel void @test_lifetime() {
; CHECK-LABEL: @test_lifetime(
; CHECK: entry:
; CHECK: [[MAIN_ALLOCA:%.*]] = alloca [16 x i32], align 16
; CHECK: [[BC:%.*]] = bitcast ptr [[MAIN_ALLOCA]] to ptr
; CHECK: call void @llvm.lifetime.start.p0(i64 64, ptr [[BC]])
; CHECK: [[BC32:%.*]] = bitcast ptr [[MAIN_ALLOCA]] to ptr
; CHECK: [[BC32_INT:%.*]] = ptrtoint ptr [[BC32]] to i64
; CHECK: [[PTR3_INT:%.*]] = add i64 [[BC32_INT]], [[OFFSET3:%.*]]
; CHECK: [[PTR3:%.*]] = inttoptr i64 [[PTR3_INT]] to ptr
; CHECK: [[PTR3_I16:%.*]] = bitcast ptr [[PTR3]] to ptr
; CHECK: [[LOAD:%.*]] = load [16 x i32], ptr [[MAIN_ALLOCA]], align 16
; CHECK: store i32 42, ptr [[PTR3]], align 4
; CHECK: [[VAL1:%.*]] = load i16, ptr [[PTR3_I16]], align 2
; CHECK: [[PTR5_INT:%.*]] = add i64 [[BC32_INT]], [[OFFSET5:%.*]]
; CHECK: [[PTR5:%.*]] = inttoptr i64 [[PTR5_INT]] to ptr
; CHECK: store i16 [[VAL1]], ptr [[PTR5]], align 2
; CHECK: [[VAL2:%.*]] = load i16, ptr [[PTR5]], align 2
; CHECK: store i16 [[A5:%.*]], ptr [[PTR5]], align 2
; CHECK: call void @llvm.lifetime.end.p0(i64 64, ptr [[BC]])
; CHECK: ret void
;
entry:
%main_alloca = alloca [16 x i32], align 16
%some_arith = add i32 1, 2
%bc = bitcast [16 x i32]* %main_alloca to i8*
call void @llvm.lifetime.start.p0i8(i64 64, i8* %bc)
%load = load [16 x i32], [16 x i32]* %main_alloca, align 16
; Lowered GEP for index 3
%bc32 = bitcast [16 x i32]* %main_alloca to i32*
%bc32_int = ptrtoint i32* %bc32 to i64
%offset3 = mul i64 3, 4
%ptr3_int = add i64 %bc32_int, %offset3
%ptr3 = inttoptr i64 %ptr3_int to i32*
store i32 42, i32* %ptr3, align 4
%ptr3_i16 = bitcast i32* %ptr3 to i16*
%val1 = load i16, i16* %ptr3_i16, align 2
; Random arithmetic, independent of %dpas_res
%a1 = add i16 %val1, 123
; DPAS call with new signature (removed i1 argument)
%vec = insertelement <8 x i16> undef, i16 %val1, i32 0
%dpas_res = call <8 x float> @llvm.genx.GenISA.sub.group.dpas.v8f32.v8f32.v8i16.v8i32(
<8 x float> zeroinitializer, <8 x i16> %vec, <8 x i32> zeroinitializer, i32 1, i32 2, i32 3, i32 4, i1 false)
%offset5 = mul i64 5, 4
%ptr5_int = add i64 %bc32_int, %offset5
%ptr5 = inttoptr i64 %ptr5_int to i16*
store i16 %val1, i16* %ptr5, align 2
%val2 = load i16, i16* %ptr5, align 2
%a2 = mul i16 %val2, 7
%a3 = xor i16 %a1, %a2
%a4 = sub i16 %a3, 42
%a5 = or i16 %a4, %val1
; store a5
store i16 %a5, i16* %ptr5, align 2
call void @llvm.lifetime.end.p0i8(i64 64, i8* %bc)
ret void
}
declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture)
declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture)
declare <8 x float> @llvm.genx.GenISA.sub.group.dpas.v8f32.v8f32.v8i16.v8i32(
<8 x float>, <8 x i16>, <8 x i32>, i32, i32, i32, i32, i1)
|
