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
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2023 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; REQUIRES: regkeys
;
; RUN: igc_opt --ocl --platformpvc --igc-private-mem-resolution --regkey EnablePrivMemNewSOATranspose=2 -S %s | FileCheck %s
;
; CHECK-LABEL: @test
;;
;; prolog in entry block. Get buffer's perThreadOffset
;;
; CHECK: [[T00:%.*]] = call i16 @llvm.genx.GenISA.simdLaneId()
; CHECK: [[simdLaneId:%.*]] = zext i16 [[T00]] to i32
; CHECK: [[simdSize:%.*]] = call i32 @llvm.genx.GenISA.simdSize()
; CHECK: [[T01:%.*]] = call i32 @llvm.genx.GenISA.hw.thread.id.alloca.i32()
; CHECK: [[T02:%.*]] = mul i32 [[simdSize]], 4096
; CHECK: [[perThreadOffset:%.*]] = mul i32 [[T01]], [[T02]]
;
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024-n8:16:32"
target triple = "spir64-unknown-unknown"
; Function Attrs: nofree nosync nounwind
define spir_kernel void @test(i32 addrspace(1)* nocapture writeonly %d, <8 x i32> %r0, <8 x i32> %payloadHeader, <3 x i32> %enqueuedLocalSize, i16 %localIdX, i16 %localIdY, i16 %localIdZ, i8* nocapture readnone %privateBase) {
entry:
%payloadHeader.scalar = extractelement <8 x i32> %payloadHeader, i64 0
%enqueuedLocalSize.scalar = extractelement <3 x i32> %enqueuedLocalSize, i64 0
%r0.scalar18 = extractelement <8 x i32> %r0, i64 1
%pb = alloca [1024 x i32], align 4
%tmp0 = mul i32 %enqueuedLocalSize.scalar, %r0.scalar18
%localIdX3 = zext i16 %localIdX to i32
%tmp1 = add i32 %tmp0, %localIdX3
%ix = add i32 %tmp1, %payloadHeader.scalar
;;
;; common code for each buffer
;;
; CHECK: %idx = zext i32 %ix to i64
; CHECK: [[T10:%.*]] = mul i32 [[simdSize]], 0
; CHECK: [[T11:%.*]] = mul i32 [[simdLaneId]], 4
; CHECK: [[T12:%.*]] = add i32 [[T10]], [[T11]]
; CHECK: [[bufferOffset:%.*]] = add {{.*}} i32 [[perThreadOffset]], [[T12]]
; CHECK: [[T13:%.*]] = ptrtoint i8* %privateBase to i64
; CHECK: [[T14:%.*]] = zext i32 [[bufferOffset]] to i64
; CHECK: [[bufbaseoff:%.*]] = add {{.*}} i64 [[T13]], [[T14]]
;;
;; first store
;;
; CHECK: [[T20:%.*]] = trunc i64 %idx to i32
; CHECK: [[T21:%.*]] = mul nsw i32 [[T20]], 4
; CHECK: [[T22:%.*]] = lshr i32 [[T21]], 2
; CHECK: [[T23:%.*]] = mul i32 [[T22]], 4
; CHECK: [[T24:%.*]] = mul i32 [[simdSize]], [[T23]]
; CHECK: [[T25:%.*]] = zext i32 [[T24]] to i64
; CHECK: [[T26:%.*]] = add {{.*}} i64 [[bufbaseoff]], [[T25]]
; CHECK: [[GEPBase:%.*]] = inttoptr i64 [[T26]] to i32*
; CHECK: [[staddr0:%.*]] = getelementptr i32, i32* [[GEPBase]], i32 0
; CHECK: store i32 %ix, i32* [[staddr0]]
;;
%idx = zext i32 %ix to i64
%tmp2 = bitcast [1024 x i32]* %pb to i8*
call void @llvm.lifetime.start.p0i8(i64 4096, i8* nonnull %tmp2)
%staddr0 = getelementptr inbounds [1024 x i32], [1024 x i32]* %pb, i64 0, i64 %idx
store i32 %ix, i32* %staddr0
;;
;; second store
;;
; CHECK: [[T31:%.*]] = mul i32 [[simdSize]], 128
;;
%staddr1 = getelementptr inbounds [1024 x i32], [1024 x i32]* %pb, i64 0, i64 32
%ixx2 = shl i32 %ix, 2
store i32 %ixx2, i32* %staddr1, align 4
%tmp3 = add i32 %ix, 1
%tmp4 = and i32 %tmp3, 1023
%idx1 = zext i32 %tmp4 to i64
;;
;; first load
;;
; CHECK: [[T41:%.*]] = mul i32 [[simdSize]],
;
%ldaddr0 = getelementptr inbounds [1024 x i32], [1024 x i32]* %pb, i64 0, i64 %idx1
%v0 = load i32, i32* %ldaddr0, align 4
;;
;; second load
;;
; CHECK: [[T51:%.*]] = mul i32 [[simdSize]], 128
;
%ldaddr1 = getelementptr inbounds [1024 x i32], [1024 x i32]* %pb, i64 0, i64 32
%v1 = load i32, i32* %ldaddr1, align 4
%v2 = add nsw i32 %v0, %v1
%arrayidx = getelementptr inbounds i32, i32 addrspace(1)* %d, i64 %idx
store i32 %v2, i32 addrspace(1)* %arrayidx, align 4
call void @llvm.lifetime.end.p0i8(i64 4096, i8* nonnull %tmp2)
;;
; CHECK: ret
ret void
}
; Function Attrs: argmemonly mustprogress nofree nosync nounwind willreturn
declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture)
; Function Attrs: argmemonly mustprogress nofree nosync nounwind willreturn
declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture)
!IGCMetadata = !{!0}
!igc.functions = !{!6}
!0 = !{!"ModuleMD", !1, !3}
!1 = !{!"compOpt", !2}
!2 = !{!"UseScratchSpacePrivateMemory", i1 true}
!3 = !{!"FuncMD", !4, !5}
!4 = !{!"FuncMDMap[1]", void (i32 addrspace(1)*, <8 x i32>, <8 x i32>, <3 x i32>, i16, i16, i16, i8*)* @test}
!5 = !{!"FuncMDValue[1]", !2}
!6 = !{void (i32 addrspace(1)*, <8 x i32>, <8 x i32>, <3 x i32>, i16, i16, i16, i8*)* @test, !408}
!408 = !{!409, !410}
!409 = !{!"function_type", i32 0}
!410 = !{!"implicit_arg_desc", !411, !412, !413, !414, !415, !416, !417}
!411 = !{i32 0}
!412 = !{i32 1}
!413 = !{i32 6}
!414 = !{i32 7}
!415 = !{i32 8}
!416 = !{i32 9}
!417 = !{i32 12}
|
