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
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2025 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
;
; REQUIRES: llvm-16-plus
; RUN: igc_opt --typed-pointers -platformpvc -igc-joint-matrix-resolution -S %s 2>&1 | FileCheck %s
; ------------------------------------------------
; JointMatrixFuncsResolutionPass
; ------------------------------------------------
%jm = type { target("spirv.JointMatrixINTEL", float, 16, 16, 3, 3, 2) }
; Test case 1:
; function itself does not have arguments to resolve
; there are instructions before calls of functions, which require resolving
; function parameters needs resolving and are using values from these instructions
; there are other instructions after calls of resolved functions which are using result of resolved function
define spir_kernel void @main(i32 %a, i32 %b, float addrspace(1)* %dst0) {
; CHECK: define spir_kernel void @main(i32 %a, i32 %b, float addrspace(1)* %dst0) {
; CHECK-NEXT: [[TMP1:%.*]] = alloca <16 x float>
; CHECK-NEXT: [[TC1:%.*]] = alloca [[JM_RESOLVED:%.*]]
; CHECK-NEXT: [[TC23:%.*]] = alloca [[JM_RESOLVED]]
; CHECK-NEXT: [[JUSTINT:%.*]] = alloca i32
; CHECK-NEXT: [[JUSTINT2:%.*]] = alloca i32
; CHECK-NEXT: call spir_func void @test_resolved(%jm.resolved* [[TC1]])
; CHECK-NEXT: call spir_func void @test2_resolved(%jm.resolved* [[TC1]], %jm.resolved* [[TC23]])
; CHECK-NEXT: call spir_func void @test3_resolved(i32* [[JUSTINT]], i32* [[JUSTINT2]], %jm.resolved* [[TC23]])
; CHECK-NEXT: call spir_func void @test3_resolved(i32* [[JUSTINT]], i32* [[JUSTINT2]], %jm.resolved* [[TC23]])
; CHECK-NEXT: [[R5:%.*]] = call spir_func %jm.resolved* @test4_resolved(%jm.resolved* [[TC1]], %jm.resolved* [[TC23]], i32 [[A:%.*]], i32 [[B:%.*]])
; CHECK-NEXT: [[R17:%.*]] = bitcast %jm.resolved* [[R5]] to <16 x float>*
; CHECK-NEXT: [[R29:%.*]] = load <16 x float>, <16 x float>* [[R17]]
; CHECK-NEXT: store <16 x float> [[R29]], <16 x float>* [[TMP1]]
; CHECK-NEXT: [[TMP2:%.*]] = bitcast <16 x float>* [[TMP1]] to i8*
; CHECK-NEXT: call void @__builtin_spriv_OpJointMatrixStoreINTEL_Accumulator_RowMajor_SG16_16x16_i32_16_global_pi64_v8i8(float addrspace(1)* [[DST0:%.*]], i8* [[TMP2]], i64 16, i32 0)
; CHECK-NEXT: [[TC311:%.*]] = alloca %jm.resolved*
; CHECK-NEXT: store %jm.resolved* [[R5]], %jm.resolved** [[TC311]]
; CHECK-NEXT: ret void
;
%tC = alloca %jm, align 8
%tC2 = alloca %jm, align 8
%justInt = alloca i32, align 8
%justInt2 = alloca i32, align 8
call spir_func void @test(%jm* %tC)
call spir_func void @test2(%jm* %tC, %jm* %tC2)
call spir_func void @test3(i32* %justInt, i32* %justInt2, %jm* %tC2)
call spir_func void @test3(i32* %justInt, i32* %justInt2, %jm* %tC2)
%r = call spir_func %jm* @test4(%jm* %tC, %jm* %tC2, i32 %a, i32 %b)
%r1 = bitcast %jm* %r to target("spirv.JointMatrixINTEL", float, 16, 16, 3, 3, 2)*
%r2 = load target("spirv.JointMatrixINTEL", float, 16, 16, 3, 3, 2), target("spirv.JointMatrixINTEL", float, 16, 16, 3, 3, 2)* %r1
call spir_func void @_Z29__spirv_JointMatrixStoreINTELPU3AS1fPU3AS143__spirv_JointMatrixINTEL__float_16_16_3_3_2liii(float addrspace(1)* %dst0, target("spirv.JointMatrixINTEL", float, 16, 16, 3, 3, 2) %r2, i64 16, i32 0, i32 3, i32 0)
%tC3 = alloca %jm*
store %jm* %r, %jm** %tC3
ret void
}
; Test case 2:
; test that calls inside were resolved correctly when signature was resolved
define spir_func void @test(%jm* %this) {
; CHECK: define spir_func void @test_resolved(%jm.resolved* %this) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = alloca <16 x float>
; CHECK-NEXT: [[JM_PTR2:%.*]] = alloca %jm.resolved*
; CHECK-NEXT: [[INDEX3:%.*]] = getelementptr %jm.resolved*, %jm.resolved** [[JM_PTR2]], i32 5
; CHECK-NEXT: store %jm.resolved* [[THIS:%.*]], %jm.resolved** [[INDEX3]]
; CHECK-NEXT: [[DST:%.*]] = alloca float
; CHECK-NEXT: [[DST0:%.*]] = addrspacecast float* [[DST]] to float addrspace(1)*
; CHECK-NEXT: store <16 x float> <float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00, float 5.000000e+00>, <16 x float>* [[TMP0]]
; CHECK-NEXT: [[TMP1:%.*]] = bitcast <16 x float>* [[TMP0]] to i8*
; CHECK-NEXT: call void @__builtin_spriv_OpJointMatrixStoreINTEL_Accumulator_RowMajor_SG16_16x16_i32_16_global_pi64_v8i8(float addrspace(1)* [[DST0]], i8* [[TMP1]], i64 16, i32 0)
; CHECK-NEXT: ret void
;
entry:
%jm.ptr = alloca %jm*
%index = getelementptr %jm*, %jm** %jm.ptr, i32 5
store %jm* %this, %jm** %index
%0 = call spir_func target("spirv.JointMatrixINTEL", float, 16, 16, 3, 3, 2) @_Z26__spirv_CompositeConstructf(float 5.000000e+00)
%dst = alloca float
%dst0 = addrspacecast float* %dst to float addrspace(1)*
call spir_func void @_Z29__spirv_JointMatrixStoreINTELPU3AS1fPU3AS143__spirv_JointMatrixINTEL__float_16_16_3_3_2liii(float addrspace(1)* %dst0, target("spirv.JointMatrixINTEL", float, 16, 16, 3, 3, 2) %0, i64 16, i32 0, i32 3, i32 0)
ret void
}
; Test case 3:
; tests that function with resolved signature which calls function with resolved signature
; is resolved correctly
define spir_func void @test2(%jm* %this, %jm* %this2) {
; CHECK: define spir_func void @test2_resolved(%jm.resolved* %this, %jm.resolved* %this2) {
; CHECK-NEXT: entry:
; CHECK-NEXT: call spir_func void @test_resolved(%jm.resolved* [[THIS:%.*]])
; CHECK-NEXT: [[TC:%.*]] = alloca %jm.resolved
; CHECK-NEXT: call spir_func void @test_resolved(%jm.resolved* [[TC]])
; CHECK-NEXT: ret void
;
entry:
call spir_func void @test(%jm* %this)
%tC = alloca %jm
call spir_func void @test(%jm* %tC)
ret void
}
; CHECK: define spir_func void @test3_resolved(i32* %this, i32* %this2, %jm.resolved* %this3) {
define spir_func void @test3(i32* %this, i32* %this2, %jm* %this3) {
entry:
ret void
}
define spir_func %jm* @test4(%jm* %this1, %jm* %this2, i32 %a, i32 %b) {
; CHECK: define spir_func %jm.resolved* @test4_resolved(%jm.resolved* %this1, %jm.resolved* %this2, i32 %a, i32 %b) {
; CHECK: end:
; CHECK-NEXT: [[RETVAL1:%.*]] = phi %jm.resolved* [ [[THIS1:%.*]], %btrue ], [ [[THIS2:%.*]], %bfalse ]
; CHECK-NEXT: ret %jm.resolved* [[RETVAL1]]
;
entry:
%0 = icmp sgt i32 %a, %b
br i1 %0, label %btrue, label %bfalse
btrue: ; preds = %2
br label %end
bfalse: ; preds = %2
br label %end
end: ; preds = %btrue, %bfalse
%retval = phi %jm* [%this1, %btrue], [%this2, %bfalse]
ret %jm* %retval
}
declare spir_func target("spirv.JointMatrixINTEL", float, 16, 16, 3, 3, 2) @_Z26__spirv_CompositeConstructf(float)
declare spir_func void @_Z29__spirv_JointMatrixStoreINTELPU3AS1fPU3AS143__spirv_JointMatrixINTEL__float_16_16_3_3_2liii(float addrspace(1)*, target("spirv.JointMatrixINTEL", float, 16, 16, 3, 3, 2), i64, i32, i32, i32)
!igc.functions = !{!0}
!0 = !{void (i32 , i32 , float addrspace(1)* )* @main, !1}
!1 = !{!2, !3}
!2 = !{!"function_type", i32 0}
!3 = !{!"sub_group_size", i32 16}
|
