1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+amx-int8 -mattr=+avx512f -verify-machineinstrs | FileCheck %s
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+amx-int8 -mattr=+avx512f,+egpr --show-mc-encoding -verify-machineinstrs | FileCheck %s --check-prefix=EGPR
5 define dso_local void @test1(ptr%buf) nounwind {
7 ; CHECK: # %bb.0: # %entry
8 ; CHECK-NEXT: pushq %rbp
9 ; CHECK-NEXT: pushq %r15
10 ; CHECK-NEXT: pushq %r14
11 ; CHECK-NEXT: pushq %rbx
12 ; CHECK-NEXT: subq $4056, %rsp # imm = 0xFD8
13 ; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0
14 ; CHECK-NEXT: vmovups %zmm0, {{[0-9]+}}(%rsp)
15 ; CHECK-NEXT: movb $1, {{[0-9]+}}(%rsp)
16 ; CHECK-NEXT: movb $8, {{[0-9]+}}(%rsp)
17 ; CHECK-NEXT: movw $8, {{[0-9]+}}(%rsp)
18 ; CHECK-NEXT: movb $8, {{[0-9]+}}(%rsp)
19 ; CHECK-NEXT: movw $8, {{[0-9]+}}(%rsp)
20 ; CHECK-NEXT: movb $8, {{[0-9]+}}(%rsp)
21 ; CHECK-NEXT: movw $8, {{[0-9]+}}(%rsp)
22 ; CHECK-NEXT: movb $8, {{[0-9]+}}(%rsp)
23 ; CHECK-NEXT: movw $8, {{[0-9]+}}(%rsp)
24 ; CHECK-NEXT: ldtilecfg {{[0-9]+}}(%rsp)
25 ; CHECK-NEXT: movl $64, %eax
26 ; CHECK-NEXT: movw $8, %bp
27 ; CHECK-NEXT: tileloadd (%rdi,%rax), %tmm3
28 ; CHECK-NEXT: xorl %eax, %eax
29 ; CHECK-NEXT: testb %al, %al
30 ; CHECK-NEXT: jne .LBB0_3
31 ; CHECK-NEXT: # %bb.1: # %loop.header.preheader
32 ; CHECK-NEXT: movq %rdi, %rbx
33 ; CHECK-NEXT: xorl %r14d, %r14d
34 ; CHECK-NEXT: movl $32, %r15d
35 ; CHECK-NEXT: .p2align 4
36 ; CHECK-NEXT: .LBB0_2: # %loop.header
37 ; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
38 ; CHECK-NEXT: movabsq $64, %rax
39 ; CHECK-NEXT: tilestored %tmm3, 3024(%rsp,%rax) # 1024-byte Folded Spill
40 ; CHECK-NEXT: vzeroupper
41 ; CHECK-NEXT: callq foo
42 ; CHECK-NEXT: ldtilecfg {{[0-9]+}}(%rsp)
43 ; CHECK-NEXT: movabsq $64, %rax
44 ; CHECK-NEXT: tileloadd 3024(%rsp,%rax), %tmm3 # 1024-byte Folded Reload
45 ; CHECK-NEXT: tileloadd (%rbx,%r15), %tmm0
46 ; CHECK-NEXT: tileloadd (%rbx,%r15), %tmm1
47 ; CHECK-NEXT: tilestored %tmm3, 1024(%rsp,%rax) # 1024-byte Folded Spill
48 ; CHECK-NEXT: tileloadd 1024(%rsp,%rax), %tmm2 # 1024-byte Folded Reload
49 ; CHECK-NEXT: tdpbssd %tmm1, %tmm0, %tmm2
50 ; CHECK-NEXT: tilestored %tmm2, (%rbx,%r15)
51 ; CHECK-NEXT: incl %r14d
52 ; CHECK-NEXT: cmpw $100, %r14w
53 ; CHECK-NEXT: jl .LBB0_2
54 ; CHECK-NEXT: .LBB0_3: # %exit
55 ; CHECK-NEXT: addq $4056, %rsp # imm = 0xFD8
56 ; CHECK-NEXT: popq %rbx
57 ; CHECK-NEXT: popq %r14
58 ; CHECK-NEXT: popq %r15
59 ; CHECK-NEXT: popq %rbp
60 ; CHECK-NEXT: tilerelease
61 ; CHECK-NEXT: vzeroupper
65 ; EGPR: # %bb.0: # %entry
66 ; EGPR-NEXT: pushq %rbp # encoding: [0x55]
67 ; EGPR-NEXT: pushq %r15 # encoding: [0x41,0x57]
68 ; EGPR-NEXT: pushq %r14 # encoding: [0x41,0x56]
69 ; EGPR-NEXT: pushq %rbx # encoding: [0x53]
70 ; EGPR-NEXT: subq $4056, %rsp # encoding: [0x48,0x81,0xec,0xd8,0x0f,0x00,0x00]
71 ; EGPR-NEXT: # imm = 0xFD8
72 ; EGPR-NEXT: vxorps %xmm0, %xmm0, %xmm0 # encoding: [0xc5,0xf8,0x57,0xc0]
73 ; EGPR-NEXT: vmovups %zmm0, {{[0-9]+}}(%rsp) # encoding: [0x62,0xf1,0x7c,0x48,0x11,0x44,0x24,0x0f]
74 ; EGPR-NEXT: movb $1, {{[0-9]+}}(%rsp) # encoding: [0xc6,0x84,0x24,0xc0,0x03,0x00,0x00,0x01]
75 ; EGPR-NEXT: movb $8, {{[0-9]+}}(%rsp) # encoding: [0xc6,0x84,0x24,0xf0,0x03,0x00,0x00,0x08]
76 ; EGPR-NEXT: movw $8, {{[0-9]+}}(%rsp) # encoding: [0x66,0xc7,0x84,0x24,0xd0,0x03,0x00,0x00,0x08,0x00]
77 ; EGPR-NEXT: movb $8, {{[0-9]+}}(%rsp) # encoding: [0xc6,0x84,0x24,0xf1,0x03,0x00,0x00,0x08]
78 ; EGPR-NEXT: movw $8, {{[0-9]+}}(%rsp) # encoding: [0x66,0xc7,0x84,0x24,0xd2,0x03,0x00,0x00,0x08,0x00]
79 ; EGPR-NEXT: movb $8, {{[0-9]+}}(%rsp) # encoding: [0xc6,0x84,0x24,0xf2,0x03,0x00,0x00,0x08]
80 ; EGPR-NEXT: movw $8, {{[0-9]+}}(%rsp) # encoding: [0x66,0xc7,0x84,0x24,0xd4,0x03,0x00,0x00,0x08,0x00]
81 ; EGPR-NEXT: movb $8, {{[0-9]+}}(%rsp) # encoding: [0xc6,0x84,0x24,0xf3,0x03,0x00,0x00,0x08]
82 ; EGPR-NEXT: movw $8, {{[0-9]+}}(%rsp) # encoding: [0x66,0xc7,0x84,0x24,0xd6,0x03,0x00,0x00,0x08,0x00]
83 ; EGPR-NEXT: ldtilecfg {{[0-9]+}}(%rsp) # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x78,0x49,0x84,0x24,0xc0,0x03,0x00,0x00]
84 ; EGPR-NEXT: movl $64, %eax # encoding: [0xb8,0x40,0x00,0x00,0x00]
85 ; EGPR-NEXT: movw $8, %bp # encoding: [0x66,0xbd,0x08,0x00]
86 ; EGPR-NEXT: tileloadd (%rdi,%rax), %tmm3 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7b,0x4b,0x1c,0x07]
87 ; EGPR-NEXT: xorl %eax, %eax # encoding: [0x31,0xc0]
88 ; EGPR-NEXT: testb %al, %al # encoding: [0x84,0xc0]
89 ; EGPR-NEXT: jne .LBB0_3 # encoding: [0x75,A]
90 ; EGPR-NEXT: # fixup A - offset: 1, value: .LBB0_3-1, kind: FK_PCRel_1
91 ; EGPR-NEXT: # %bb.1: # %loop.header.preheader
92 ; EGPR-NEXT: movq %rdi, %rbx # encoding: [0x48,0x89,0xfb]
93 ; EGPR-NEXT: xorl %r14d, %r14d # encoding: [0x45,0x31,0xf6]
94 ; EGPR-NEXT: movl $32, %r15d # encoding: [0x41,0xbf,0x20,0x00,0x00,0x00]
95 ; EGPR-NEXT: .p2align 4
96 ; EGPR-NEXT: .LBB0_2: # %loop.header
97 ; EGPR-NEXT: # =>This Inner Loop Header: Depth=1
98 ; EGPR-NEXT: movabsq $64, %rax # encoding: [0x48,0xb8,0x40,0x00,0x00,0x00,0x00,0x00,0x00,0x00]
99 ; EGPR-NEXT: tilestored %tmm3, 3024(%rsp,%rax) # 1024-byte Folded Spill
100 ; EGPR-NEXT: # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7a,0x4b,0x9c,0x04,0xd0,0x0b,0x00,0x00]
101 ; EGPR-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
102 ; EGPR-NEXT: callq foo # encoding: [0xe8,A,A,A,A]
103 ; EGPR-NEXT: # fixup A - offset: 1, value: foo-4, kind: reloc_branch_4byte_pcrel
104 ; EGPR-NEXT: ldtilecfg {{[0-9]+}}(%rsp) # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x78,0x49,0x84,0x24,0xc0,0x03,0x00,0x00]
105 ; EGPR-NEXT: movabsq $64, %rax # encoding: [0x48,0xb8,0x40,0x00,0x00,0x00,0x00,0x00,0x00,0x00]
106 ; EGPR-NEXT: tileloadd 3024(%rsp,%rax), %tmm3 # 1024-byte Folded Reload
107 ; EGPR-NEXT: # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7b,0x4b,0x9c,0x04,0xd0,0x0b,0x00,0x00]
108 ; EGPR-NEXT: tileloadd (%rbx,%r15), %tmm0 # EVEX TO VEX Compression encoding: [0xc4,0xa2,0x7b,0x4b,0x04,0x3b]
109 ; EGPR-NEXT: tileloadd (%rbx,%r15), %tmm1 # EVEX TO VEX Compression encoding: [0xc4,0xa2,0x7b,0x4b,0x0c,0x3b]
110 ; EGPR-NEXT: tilestored %tmm3, 1024(%rsp,%rax) # 1024-byte Folded Spill
111 ; EGPR-NEXT: # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7a,0x4b,0x9c,0x04,0x00,0x04,0x00,0x00]
112 ; EGPR-NEXT: tileloadd 1024(%rsp,%rax), %tmm2 # 1024-byte Folded Reload
113 ; EGPR-NEXT: # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7b,0x4b,0x94,0x04,0x00,0x04,0x00,0x00]
114 ; EGPR-NEXT: tdpbssd %tmm1, %tmm0, %tmm2 # encoding: [0xc4,0xe2,0x73,0x5e,0xd0]
115 ; EGPR-NEXT: tilestored %tmm2, (%rbx,%r15) # EVEX TO VEX Compression encoding: [0xc4,0xa2,0x7a,0x4b,0x14,0x3b]
116 ; EGPR-NEXT: incl %r14d # encoding: [0x41,0xff,0xc6]
117 ; EGPR-NEXT: cmpw $100, %r14w # encoding: [0x66,0x41,0x83,0xfe,0x64]
118 ; EGPR-NEXT: jl .LBB0_2 # encoding: [0x7c,A]
119 ; EGPR-NEXT: # fixup A - offset: 1, value: .LBB0_2-1, kind: FK_PCRel_1
120 ; EGPR-NEXT: .LBB0_3: # %exit
121 ; EGPR-NEXT: addq $4056, %rsp # encoding: [0x48,0x81,0xc4,0xd8,0x0f,0x00,0x00]
122 ; EGPR-NEXT: # imm = 0xFD8
123 ; EGPR-NEXT: popq %rbx # encoding: [0x5b]
124 ; EGPR-NEXT: popq %r14 # encoding: [0x41,0x5e]
125 ; EGPR-NEXT: popq %r15 # encoding: [0x41,0x5f]
126 ; EGPR-NEXT: popq %rbp # encoding: [0x5d]
127 ; EGPR-NEXT: tilerelease # encoding: [0xc4,0xe2,0x78,0x49,0xc0]
128 ; EGPR-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
129 ; EGPR-NEXT: retq # encoding: [0xc3]
131 %t1 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 8, ptr %buf, i64 64)
132 br i1 undef, label %loop.header, label %exit
135 %ivphi = phi i16 [0, %entry], [%iv, %loop.latch]
140 %t2 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 8, ptr %buf, i64 32)
141 %t3 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 8, ptr %buf, i64 32)
142 %t4 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 8, i16 8, i16 8, x86_amx %t1, x86_amx %t2, x86_amx %t3)
143 tail call void @llvm.x86.tilestored64.internal(i16 8, i16 8, ptr %buf, i64 32, x86_amx %t4)
147 %iv = add i16 %ivphi, 1
148 %c = icmp slt i16 %iv, 100
149 br i1 %c, label %loop.header, label %exit
155 define dso_local void @test2(ptr%buf) nounwind {
156 ; CHECK-LABEL: test2:
157 ; CHECK: # %bb.0: # %entry
158 ; CHECK-NEXT: pushq %rbp
159 ; CHECK-NEXT: pushq %r15
160 ; CHECK-NEXT: pushq %r14
161 ; CHECK-NEXT: pushq %rbx
162 ; CHECK-NEXT: subq $72, %rsp
163 ; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0
164 ; CHECK-NEXT: vmovups %zmm0, {{[0-9]+}}(%rsp)
165 ; CHECK-NEXT: movb $1, {{[0-9]+}}(%rsp)
166 ; CHECK-NEXT: movb $8, {{[0-9]+}}(%rsp)
167 ; CHECK-NEXT: movw $8, {{[0-9]+}}(%rsp)
168 ; CHECK-NEXT: movb $8, {{[0-9]+}}(%rsp)
169 ; CHECK-NEXT: movw $8, {{[0-9]+}}(%rsp)
170 ; CHECK-NEXT: movb $8, {{[0-9]+}}(%rsp)
171 ; CHECK-NEXT: movw $8, {{[0-9]+}}(%rsp)
172 ; CHECK-NEXT: ldtilecfg {{[0-9]+}}(%rsp)
173 ; CHECK-NEXT: movw $8, %bp
174 ; CHECK-NEXT: tilezero %tmm0
175 ; CHECK-NEXT: xorl %eax, %eax
176 ; CHECK-NEXT: testb %al, %al
177 ; CHECK-NEXT: jne .LBB1_3
178 ; CHECK-NEXT: # %bb.1: # %loop.header.preheader
179 ; CHECK-NEXT: movq %rdi, %rbx
180 ; CHECK-NEXT: xorl %r14d, %r14d
181 ; CHECK-NEXT: movl $32, %r15d
182 ; CHECK-NEXT: .p2align 4
183 ; CHECK-NEXT: .LBB1_2: # %loop.header
184 ; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
185 ; CHECK-NEXT: tilezero %tmm0
186 ; CHECK-NEXT: vzeroupper
187 ; CHECK-NEXT: callq foo
188 ; CHECK-NEXT: ldtilecfg {{[0-9]+}}(%rsp)
189 ; CHECK-NEXT: tilezero %tmm2
190 ; CHECK-NEXT: tileloadd (%rbx,%r15), %tmm0
191 ; CHECK-NEXT: tileloadd (%rbx,%r15), %tmm1
192 ; CHECK-NEXT: tdpbssd %tmm1, %tmm0, %tmm2
193 ; CHECK-NEXT: tilestored %tmm2, (%rbx,%r15)
194 ; CHECK-NEXT: incl %r14d
195 ; CHECK-NEXT: cmpw $100, %r14w
196 ; CHECK-NEXT: jl .LBB1_2
197 ; CHECK-NEXT: .LBB1_3: # %exit
198 ; CHECK-NEXT: addq $72, %rsp
199 ; CHECK-NEXT: popq %rbx
200 ; CHECK-NEXT: popq %r14
201 ; CHECK-NEXT: popq %r15
202 ; CHECK-NEXT: popq %rbp
203 ; CHECK-NEXT: tilerelease
204 ; CHECK-NEXT: vzeroupper
208 ; EGPR: # %bb.0: # %entry
209 ; EGPR-NEXT: pushq %rbp # encoding: [0x55]
210 ; EGPR-NEXT: pushq %r15 # encoding: [0x41,0x57]
211 ; EGPR-NEXT: pushq %r14 # encoding: [0x41,0x56]
212 ; EGPR-NEXT: pushq %rbx # encoding: [0x53]
213 ; EGPR-NEXT: subq $72, %rsp # encoding: [0x48,0x83,0xec,0x48]
214 ; EGPR-NEXT: vxorps %xmm0, %xmm0, %xmm0 # encoding: [0xc5,0xf8,0x57,0xc0]
215 ; EGPR-NEXT: vmovups %zmm0, {{[0-9]+}}(%rsp) # encoding: [0x62,0xf1,0x7c,0x48,0x11,0x84,0x24,0x08,0x00,0x00,0x00]
216 ; EGPR-NEXT: movb $1, {{[0-9]+}}(%rsp) # encoding: [0xc6,0x44,0x24,0x08,0x01]
217 ; EGPR-NEXT: movb $8, {{[0-9]+}}(%rsp) # encoding: [0xc6,0x44,0x24,0x38,0x08]
218 ; EGPR-NEXT: movw $8, {{[0-9]+}}(%rsp) # encoding: [0x66,0xc7,0x44,0x24,0x18,0x08,0x00]
219 ; EGPR-NEXT: movb $8, {{[0-9]+}}(%rsp) # encoding: [0xc6,0x44,0x24,0x39,0x08]
220 ; EGPR-NEXT: movw $8, {{[0-9]+}}(%rsp) # encoding: [0x66,0xc7,0x44,0x24,0x1a,0x08,0x00]
221 ; EGPR-NEXT: movb $8, {{[0-9]+}}(%rsp) # encoding: [0xc6,0x44,0x24,0x3a,0x08]
222 ; EGPR-NEXT: movw $8, {{[0-9]+}}(%rsp) # encoding: [0x66,0xc7,0x44,0x24,0x1c,0x08,0x00]
223 ; EGPR-NEXT: ldtilecfg {{[0-9]+}}(%rsp) # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x78,0x49,0x44,0x24,0x08]
224 ; EGPR-NEXT: movw $8, %bp # encoding: [0x66,0xbd,0x08,0x00]
225 ; EGPR-NEXT: tilezero %tmm0 # encoding: [0xc4,0xe2,0x7b,0x49,0xc0]
226 ; EGPR-NEXT: xorl %eax, %eax # encoding: [0x31,0xc0]
227 ; EGPR-NEXT: testb %al, %al # encoding: [0x84,0xc0]
228 ; EGPR-NEXT: jne .LBB1_3 # encoding: [0x75,A]
229 ; EGPR-NEXT: # fixup A - offset: 1, value: .LBB1_3-1, kind: FK_PCRel_1
230 ; EGPR-NEXT: # %bb.1: # %loop.header.preheader
231 ; EGPR-NEXT: movq %rdi, %rbx # encoding: [0x48,0x89,0xfb]
232 ; EGPR-NEXT: xorl %r14d, %r14d # encoding: [0x45,0x31,0xf6]
233 ; EGPR-NEXT: movl $32, %r15d # encoding: [0x41,0xbf,0x20,0x00,0x00,0x00]
234 ; EGPR-NEXT: .p2align 4
235 ; EGPR-NEXT: .LBB1_2: # %loop.header
236 ; EGPR-NEXT: # =>This Inner Loop Header: Depth=1
237 ; EGPR-NEXT: tilezero %tmm0 # encoding: [0xc4,0xe2,0x7b,0x49,0xc0]
238 ; EGPR-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
239 ; EGPR-NEXT: callq foo # encoding: [0xe8,A,A,A,A]
240 ; EGPR-NEXT: # fixup A - offset: 1, value: foo-4, kind: reloc_branch_4byte_pcrel
241 ; EGPR-NEXT: ldtilecfg {{[0-9]+}}(%rsp) # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x78,0x49,0x44,0x24,0x08]
242 ; EGPR-NEXT: tilezero %tmm2 # encoding: [0xc4,0xe2,0x7b,0x49,0xd0]
243 ; EGPR-NEXT: tileloadd (%rbx,%r15), %tmm0 # EVEX TO VEX Compression encoding: [0xc4,0xa2,0x7b,0x4b,0x04,0x3b]
244 ; EGPR-NEXT: tileloadd (%rbx,%r15), %tmm1 # EVEX TO VEX Compression encoding: [0xc4,0xa2,0x7b,0x4b,0x0c,0x3b]
245 ; EGPR-NEXT: tdpbssd %tmm1, %tmm0, %tmm2 # encoding: [0xc4,0xe2,0x73,0x5e,0xd0]
246 ; EGPR-NEXT: tilestored %tmm2, (%rbx,%r15) # EVEX TO VEX Compression encoding: [0xc4,0xa2,0x7a,0x4b,0x14,0x3b]
247 ; EGPR-NEXT: incl %r14d # encoding: [0x41,0xff,0xc6]
248 ; EGPR-NEXT: cmpw $100, %r14w # encoding: [0x66,0x41,0x83,0xfe,0x64]
249 ; EGPR-NEXT: jl .LBB1_2 # encoding: [0x7c,A]
250 ; EGPR-NEXT: # fixup A - offset: 1, value: .LBB1_2-1, kind: FK_PCRel_1
251 ; EGPR-NEXT: .LBB1_3: # %exit
252 ; EGPR-NEXT: addq $72, %rsp # encoding: [0x48,0x83,0xc4,0x48]
253 ; EGPR-NEXT: popq %rbx # encoding: [0x5b]
254 ; EGPR-NEXT: popq %r14 # encoding: [0x41,0x5e]
255 ; EGPR-NEXT: popq %r15 # encoding: [0x41,0x5f]
256 ; EGPR-NEXT: popq %rbp # encoding: [0x5d]
257 ; EGPR-NEXT: tilerelease # encoding: [0xc4,0xe2,0x78,0x49,0xc0]
258 ; EGPR-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
259 ; EGPR-NEXT: retq # encoding: [0xc3]
261 %t1 = tail call x86_amx @llvm.x86.tilezero.internal(i16 8, i16 8)
262 br i1 undef, label %loop.header, label %exit
265 %ivphi = phi i16 [0, %entry], [%iv, %loop.latch]
270 %t2 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 8, ptr %buf, i64 32)
271 %t3 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 8, ptr %buf, i64 32)
272 %t4 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 8, i16 8, i16 8, x86_amx %t1, x86_amx %t2, x86_amx %t3)
273 tail call void @llvm.x86.tilestored64.internal(i16 8, i16 8, ptr %buf, i64 32, x86_amx %t4)
277 %iv = add i16 %ivphi, 1
278 %c = icmp slt i16 %iv, 100
279 br i1 %c, label %loop.header, label %exit
285 declare dso_local void @foo()
286 declare x86_amx @llvm.x86.tilezero.internal(i16, i16)
287 declare x86_amx @llvm.x86.tileloadd64.internal(i16, i16, ptr, i64)
288 declare x86_amx @llvm.x86.tdpbssd.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)
289 declare void @llvm.x86.tilestored64.internal(i16, i16, ptr, i64, x86_amx)