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
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2022-2024 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
;
; RUN: igc_opt -enable-debugify -igc-joint-matrix-resolution -dce -S --platformdg2 2>&1 < %s | FileCheck %s
; ------------------------------------------------
; JointMatrixFuncsResolutionPass
; ------------------------------------------------
; Debug-info related check
; CHECK-COUNT-1: WARNING
; CHECK-SAME: Missing line 7
; CHECK: CheckModuleDebugify: PASS
define spir_kernel void @test_jm(
i32 %t1_a, i8* %t1_dst1, i32* %t1_dst2,
i8* %t2_a, i8* %t2_dst,
float addrspace(1)* %t3_a, float addrspace(1)* %t3_dst) {
call void @fill_length(i32 %t1_a, i8* %t1_dst1, i32* %t1_dst2)
call void @load_store_legacy(i8* %t2_a, i8* %t2_dst)
call void @load_store_acc_transpose(float addrspace(1)* %t3_a, float addrspace(1)* %t3_dst)
ret void
}
%intel.joint_matrix_packedA_8x32_i8_ = type opaque
define void @fill_length(i32 %a, i8* %dst, i32* %dst2) {
; CHECK-LABEL: define void @fill_length(
; CHECK: [[PTR:%.*]] = alloca <8 x i32>
; CHECK: [[TMP1:%.*]] = insertelement <8 x i32> undef, i32 [[A:%.*]], i64 0
; CHECK: [[TMP2:%.*]] = insertelement <8 x i32> [[TMP1]], i32 [[A]], i64 1
; CHECK: [[TMP3:%.*]] = insertelement <8 x i32> [[TMP2]], i32 [[A]], i64 2
; CHECK: [[TMP4:%.*]] = insertelement <8 x i32> [[TMP3]], i32 [[A]], i64 3
; CHECK: [[TMP5:%.*]] = insertelement <8 x i32> [[TMP4]], i32 [[A]], i64 4
; CHECK: [[TMP6:%.*]] = insertelement <8 x i32> [[TMP5]], i32 [[A]], i64 5
; CHECK: [[TMP7:%.*]] = insertelement <8 x i32> [[TMP6]], i32 [[A]], i64 6
; CHECK: [[TMP8:%.*]] = insertelement <8 x i32> [[TMP7]], i32 [[A]], i64 7
; CHECK: store <8 x i32> [[TMP8]], <8 x i32>* [[PTR:%.*]]
; CHECK: [[MATPTR:%.*]] = bitcast <8 x i32>* [[PTR:%.*]] to i8*, !dbg [[DBG1:![0-9]*]]
; CHECK: call void @__builtin_spriv_OpJointMatrixStoreINTEL_PackedA_RowMajor_8x32_i8_8_generic_pi64_v8i8(i8* %dst, i8* [[MATPTR]], i32 8, i32 0), !dbg [[DBG1]]
; CHECK: ret void
; CHECK-NOT: error
;
%1 = call spir_func %intel.joint_matrix_packedA_8x32_i8_* @__builtin_spirv_OpCompositeConstructJointMatrixINTEL(i32 %a)
call spir_func void @__builtin_spirv_OpJointMatrixStoreINTEL(i8* %dst, %intel.joint_matrix_packedA_8x32_i8_* %1, i32 8, i32 0)
%2 = call spir_func i32 @__builtin_spirv_OpJointMatrixWorkItemLengthINTEL(%intel.joint_matrix_packedA_8x32_i8_* %1)
store i32 %2, i32* %dst2, align 4
ret void
}
declare spir_func %intel.joint_matrix_packedA_8x32_i8_* @__builtin_spirv_OpCompositeConstructJointMatrixINTEL(i32)
declare spir_func i32 @__builtin_spirv_OpJointMatrixWorkItemLengthINTEL(%intel.joint_matrix_packedA_8x32_i8_*)
declare spir_func void @__builtin_spirv_OpJointMatrixStoreINTEL(i8*, %intel.joint_matrix_packedA_8x32_i8_*, i32, i32)
%intel.joint_matrix_packedA_8x16_i16_ = type opaque
define void @load_store_legacy(i8* %a, i8* %dst) {
; CHECK-LABEL: define void @load_store_legacy(
; CHECK: [[TMP4:%.*]] = alloca <8 x i32>
; CHECK: [[PTR:%.*]] = alloca <8 x i32>
; CHECK: [[MATPTR:%.*]] = bitcast <8 x i32>* [[PTR]] to i8*
; CHECK: call void @__builtin_spriv_OpJointMatrixLoadINTEL_PackedA_RowMajor_8x16_i16_8_generic_v8i8_pi32_i32(i8* [[MATPTR]], i8* %a, i32 16, i32 0), !dbg [[DBG2:![0-9]*]]
; CHECK: [[MATRIX:%.*]] = load <8 x i32>, <8 x i32>* [[PTR]]
; CHECK: store <8 x i32> [[MATRIX]], <8 x i32>* [[TMP4]]
; CHECK: [[TMP5:%.*]] = bitcast <8 x i32>* [[TMP4]] to i8*
; CHECK: call void @__builtin_spriv_OpJointMatrixStoreINTEL_PackedA_RowMajor_8x16_i16_8_generic_pi64_v8i8(i8* %dst, i8* [[TMP5]], i32 8, i32 0), !dbg [[DBG3:![0-9]*]]
; CHECK: ret void
; CHECK-NOT: error
;
%1 = call spir_func %intel.joint_matrix_packedA_8x16_i16_* @__builtin_spirv_OpJointMatrixLoadINTEL(i8* %a, i32 16, i32 0)
call spir_func void @__builtin_spirv_OpJointMatrixStoreINTEL.8x16(i8* %dst, %intel.joint_matrix_packedA_8x16_i16_* %1, i32 8, i32 0)
ret void
}
declare spir_func %intel.joint_matrix_packedA_8x16_i16_* @__builtin_spirv_OpJointMatrixLoadINTEL(i8*, i32, i32)
declare spir_func void @__builtin_spirv_OpJointMatrixStoreINTEL.8x16(i8*, %intel.joint_matrix_packedA_8x16_i16_*, i32, i32)
%spirv.JointMatrixINTEL._float_8_8_3_3_2 = type opaque
define void @load_store_acc_transpose(float addrspace(1)* %a, float addrspace(1)* %dst) {
; CHECK-LABEL: define void @load_store_acc_transpose(
; CHECK: [[TMP4:%.*]] = alloca <8 x float>
; CHECK: [[PTR:%.*]] = alloca <8 x float>
; CHECK: [[MATPTR:%.*]] = bitcast <8 x float>* [[PTR]] to i8*
; CHECK: call void @__builtin_spriv_OpJointMatrixLoadINTEL_Accumulator_ColumnMajor_8x8_i32_8_global_v8i8_pi32_i32(i8* [[MATPTR]], float addrspace(1)* %a, i64 64, i32 0), !dbg [[DBG2:![0-9]*]]
; CHECK: [[MATRIX:%.*]] = load <8 x float>, <8 x float>* [[PTR]]
; CHECK: store <8 x float> [[MATRIX]], <8 x float>* [[TMP4]]
; CHECK: [[TMP5:%.*]] = bitcast <8 x float>* [[TMP4]] to i8*
; CHECK: call void @__builtin_spriv_OpJointMatrixStoreINTEL_Accumulator_ColumnMajor_8x8_i32_8_global_pi64_v8i8(float addrspace(1)* %dst, i8* [[TMP5]], i64 64, i32 0), !dbg [[DBG3:![0-9]*]]
; CHECK: ret void
; CHECK-NOT: error
; CHECK-NOT: @_Z26__spirv_CompositeConstructf(float 1
;
%1 = call spir_func %spirv.JointMatrixINTEL._float_8_8_3_3_2 addrspace(1)* @_Z80__spirv_JointMatrixLoadINTEL_RPU3AS142__spirv_JointMatrixINTEL__float_8_8_3_3_2PU3AS1fliii(float addrspace(1)* %a, i64 64, i32 1, i32 3, i32 0) #0
%dead_call = call spir_func %spirv.JointMatrixINTEL._float_8_8_3_3_2 addrspace(1)* @_Z26__spirv_CompositeConstructf(float 1.000000e+00)
call spir_func void @_Z29__spirv_JointMatrixStoreINTELPU3AS1fPU3AS142__spirv_JointMatrixINTEL__float_8_8_3_3_2liii(float addrspace(1)* %dst, %spirv.JointMatrixINTEL._float_8_8_3_3_2 addrspace(1)* %1, i64 64, i32 1, i32 3, i32 0) #0
ret void
}
declare spir_func %spirv.JointMatrixINTEL._float_8_8_3_3_2 addrspace(1)* @_Z80__spirv_JointMatrixLoadINTEL_RPU3AS142__spirv_JointMatrixINTEL__float_8_8_3_3_2PU3AS1fliii(float addrspace(1)*, i64, i32, i32, i32) #0
declare spir_func void @_Z29__spirv_JointMatrixStoreINTELPU3AS1fPU3AS142__spirv_JointMatrixINTEL__float_8_8_3_3_2liii(float addrspace(1)*, %spirv.JointMatrixINTEL._float_8_8_3_3_2 addrspace(1)*, i64, i32, i32, i32) #0
declare spir_func %spirv.JointMatrixINTEL._float_8_8_3_3_2 addrspace(1)* @_Z26__spirv_CompositeConstructf(float) #0
!igc.functions = !{!0}
!0 = !{void (i32, i8*, i32*, i8*, i8*, float addrspace(1)*, float addrspace(1)*)* @test_jm, !1}
!1 = !{!2, !3}
!2 = !{!"function_type", i32 0}
!3 = !{!"sub_group_size", i32 8}
|
