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@buf = dso_local global [3072 x i8] zeroinitializer, align 16
4
5define dso_local void @test1(i16 signext %0, i16 signext %1) nounwind {
6; CHECK-LABEL: test1:
7; CHECK:       # %bb.0:
8; CHECK-NEXT:    vpxord %zmm0, %zmm0, %zmm0
9; CHECK-NEXT:    vmovdqu64 %zmm0, -{{[0-9]+}}(%rsp)
10; CHECK-NEXT:    movb $1, -{{[0-9]+}}(%rsp)
11; CHECK-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
12; CHECK-NEXT:    movw $8, -{{[0-9]+}}(%rsp)
13; CHECK-NEXT:    movb $8, -{{[0-9]+}}(%rsp)
14; CHECK-NEXT:    movw %si, -{{[0-9]+}}(%rsp)
15; CHECK-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
16; CHECK-NEXT:    movw %si, -{{[0-9]+}}(%rsp)
17; CHECK-NEXT:    ldtilecfg -{{[0-9]+}}(%rsp)
18; CHECK-NEXT:    movl $buf, %eax
19; CHECK-NEXT:    movl $32, %ecx
20; CHECK-NEXT:    movw $8, %dx
21; CHECK-NEXT:    tileloadd (%rax,%rcx), %tmm0
22; CHECK-NEXT:    movl $buf+1024, %eax
23; CHECK-NEXT:    tileloadd (%rax,%rcx), %tmm1
24; CHECK-NEXT:    movl $buf+2048, %eax
25; CHECK-NEXT:    tileloadd (%rax,%rcx), %tmm2
26; CHECK-NEXT:    tdpbssd %tmm1, %tmm0, %tmm2
27; CHECK-NEXT:    tilestored %tmm2, (%rax,%rcx)
28; CHECK-NEXT:    tilerelease
29; CHECK-NEXT:    vzeroupper
30; CHECK-NEXT:    jmp foo # TAILCALL
31  %3 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 8, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 0), i64 32)
32  %4 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 1024), i64 32)
33  %5 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 2048), i64 32)
34  %6 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %0, i16 %1, i16 8, x86_amx %5, x86_amx %3, x86_amx %4)
35  call void @llvm.dbg.value(metadata x86_amx %6, metadata !DILocalVariable(name: "1", scope: !2), metadata !DIExpression()), !dbg !3
36  tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 2048), i64 32, x86_amx %6)
37  tail call void @foo()
38  ret void
39}
40
41define dso_local void @test2(i16 signext %0, i16 signext %1) nounwind {
42; CHECK-LABEL: test2:
43; CHECK:       # %bb.0:
44; CHECK-NEXT:    pushq %rbp
45; CHECK-NEXT:    pushq %rbx
46; CHECK-NEXT:    subq $72, %rsp
47; CHECK-NEXT:    movl %esi, %ebx
48; CHECK-NEXT:    movl %edi, %ebp
49; CHECK-NEXT:    vpxord %zmm0, %zmm0, %zmm0
50; CHECK-NEXT:    vmovdqu64 %zmm0, {{[0-9]+}}(%rsp)
51; CHECK-NEXT:    vzeroupper
52; CHECK-NEXT:    callq foo
53; CHECK-NEXT:    movb $1, {{[0-9]+}}(%rsp)
54; CHECK-NEXT:    movb $8, {{[0-9]+}}(%rsp)
55; CHECK-NEXT:    movw %bx, {{[0-9]+}}(%rsp)
56; CHECK-NEXT:    movb %bpl, {{[0-9]+}}(%rsp)
57; CHECK-NEXT:    movw %bx, {{[0-9]+}}(%rsp)
58; CHECK-NEXT:    movb %bpl, {{[0-9]+}}(%rsp)
59; CHECK-NEXT:    movw $8, {{[0-9]+}}(%rsp)
60; CHECK-NEXT:    movb $8, {{[0-9]+}}(%rsp)
61; CHECK-NEXT:    movw %bx, {{[0-9]+}}(%rsp)
62; CHECK-NEXT:    movb %bpl, {{[0-9]+}}(%rsp)
63; CHECK-NEXT:    movw $8, {{[0-9]+}}(%rsp)
64; CHECK-NEXT:    ldtilecfg {{[0-9]+}}(%rsp)
65; CHECK-NEXT:    xorl %eax, %eax
66; CHECK-NEXT:    testb %al, %al
67; CHECK-NEXT:    jne .LBB1_3
68; CHECK-NEXT:  # %bb.1: # %if.true
69; CHECK-NEXT:    movw $8, %ax
70; CHECK-NEXT:    tilezero %tmm0
71; CHECK-NEXT:    movl $32, %ecx
72; CHECK-NEXT:    movl $buf+1024, %edx
73; CHECK-NEXT:    tileloadd (%rdx,%rcx), %tmm1
74; CHECK-NEXT:    movl $buf+2048, %edx
75; CHECK-NEXT:    tileloadd (%rdx,%rcx), %tmm2
76; CHECK-NEXT:    tdpbssd %tmm2, %tmm1, %tmm0
77; CHECK-NEXT:    tilestored %tmm0, (%rdx,%rcx)
78; CHECK-NEXT:    jmp .LBB1_2
79; CHECK-NEXT:  .LBB1_3: # %if.false
80; CHECK-NEXT:    movl $buf, %eax
81; CHECK-NEXT:    movl $32, %ecx
82; CHECK-NEXT:    movw $8, %dx
83; CHECK-NEXT:    tileloadd (%rax,%rcx), %tmm3
84; CHECK-NEXT:    movl $buf+1024, %eax
85; CHECK-NEXT:    tileloadd (%rax,%rcx), %tmm4
86; CHECK-NEXT:    movl $buf+2048, %eax
87; CHECK-NEXT:    tileloadd (%rax,%rcx), %tmm2
88; CHECK-NEXT:    tdpbssd %tmm2, %tmm4, %tmm3
89; CHECK-NEXT:    tilestored %tmm3, (%rax,%rcx)
90; CHECK-NEXT:  .LBB1_2: # %if.true
91; CHECK-NEXT:    addq $72, %rsp
92; CHECK-NEXT:    popq %rbx
93; CHECK-NEXT:    popq %rbp
94; CHECK-NEXT:    tilerelease
95; CHECK-NEXT:    retq
96  call void @foo()
97  br i1 undef, label %if.true, label %if.false
98
99if.true:
100  %t1 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %0, i16 8)
101  %t2 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 1024), i64 32)
102  %t3 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 2048), i64 32)
103  %t4 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %0, i16 %1, i16 8, x86_amx %t1, x86_amx %t2, x86_amx %t3)
104  tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 2048), i64 32, x86_amx %t4)
105  br label %exit
106
107if.false:
108  %t5 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 8, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 0), i64 32)
109  %t6 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 1024), i64 32)
110  %t7 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 2048), i64 32)
111  %t8 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %0, i16 %1, i16 8, x86_amx %t5, x86_amx %t6, x86_amx %t7)
112  tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 2048), i64 32, x86_amx %t8)
113  br label %exit
114
115exit:
116  ret void
117}
118
119define dso_local void @test3(i16 signext %0, i16 signext %1) nounwind {
120; CHECK-LABEL: test3:
121; CHECK:       # %bb.0:
122; CHECK-NEXT:    vpxord %zmm0, %zmm0, %zmm0
123; CHECK-NEXT:    vmovdqu64 %zmm0, -{{[0-9]+}}(%rsp)
124; CHECK-NEXT:    xorl %eax, %eax
125; CHECK-NEXT:    testb %al, %al
126; CHECK-NEXT:    jne .LBB2_2
127; CHECK-NEXT:  # %bb.1: # %if.true
128; CHECK-NEXT:    incl %edi
129; CHECK-NEXT:    jmp .LBB2_3
130; CHECK-NEXT:  .LBB2_2: # %if.false
131; CHECK-NEXT:    decl %edi
132; CHECK-NEXT:  .LBB2_3: # %exit
133; CHECK-NEXT:    movb $1, -{{[0-9]+}}(%rsp)
134; CHECK-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
135; CHECK-NEXT:    movw %si, -{{[0-9]+}}(%rsp)
136; CHECK-NEXT:    ldtilecfg -{{[0-9]+}}(%rsp)
137; CHECK-NEXT:    tilezero %tmm0
138; CHECK-NEXT:    movl $buf, %eax
139; CHECK-NEXT:    movl $32, %ecx
140; CHECK-NEXT:    tilestored %tmm0, (%rax,%rcx)
141; CHECK-NEXT:    tilerelease
142; CHECK-NEXT:    vzeroupper
143; CHECK-NEXT:    retq
144  br i1 undef, label %if.true, label %if.false
145
146if.true:
147  %3 = add i16 %0, 1
148  br label %exit
149
150if.false:
151  %4 = sub i16 %0, 1
152  br label %exit
153
154exit:
155  %5 = phi i16 [ %3, %if.true ], [ %4, %if.false ]
156  %6 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %5, i16 %1)
157  tail call void @llvm.x86.tilestored64.internal(i16 %5, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 0), i64 32, x86_amx %6)
158  ret void
159}
160
161; TODO: There's PRA Tile Register Configure bug needs to fix later.
162define dso_local void @test4(i16 signext %0, i16 signext %1) nounwind {
163; CHECK-LABEL: test4:
164; CHECK:       # %bb.0:
165; CHECK-NEXT:    vpxord %zmm0, %zmm0, %zmm0
166; CHECK-NEXT:    vmovdqu64 %zmm0, -{{[0-9]+}}(%rsp)
167; CHECK-NEXT:    xorl %eax, %eax
168; CHECK-NEXT:    testb %al, %al
169; CHECK-NEXT:    jne .LBB3_3
170; CHECK-NEXT:  # %bb.1: # %if.true
171; CHECK-NEXT:    incl %edi
172; CHECK-NEXT:    xorl %eax, %eax
173; CHECK-NEXT:    testb %al, %al
174; CHECK-NEXT:    jne .LBB3_4
175; CHECK-NEXT:  .LBB3_2: # %amx2
176; CHECK-NEXT:    movb $1, -{{[0-9]+}}(%rsp)
177; CHECK-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
178; CHECK-NEXT:    movw %si, -{{[0-9]+}}(%rsp)
179; CHECK-NEXT:    ldtilecfg -{{[0-9]+}}(%rsp)
180; CHECK-NEXT:    movl $32, %eax
181; CHECK-NEXT:    movl $buf+1024, %ecx
182; CHECK-NEXT:    tileloadd (%rcx,%rax), %tmm0
183; CHECK-NEXT:    movl $buf, %ecx
184; CHECK-NEXT:    tilestored %tmm0, (%rcx,%rax)
185; CHECK-NEXT:    tilerelease
186; CHECK-NEXT:    vzeroupper
187; CHECK-NEXT:    retq
188; CHECK-NEXT:  .LBB3_3: # %if.false
189; CHECK-NEXT:    decl %edi
190; CHECK-NEXT:    xorl %eax, %eax
191; CHECK-NEXT:    testb %al, %al
192; CHECK-NEXT:    jne .LBB3_2
193; CHECK-NEXT:  .LBB3_4: # %amx1
194; CHECK-NEXT:    ldtilecfg -{{[0-9]+}}(%rsp)
195; CHECK-NEXT:    tilezero %tmm0
196; CHECK-NEXT:    movl $buf, %eax
197; CHECK-NEXT:    movl $32, %ecx
198; CHECK-NEXT:    tilestored %tmm0, (%rax,%rcx)
199; CHECK-NEXT:    tilerelease
200; CHECK-NEXT:    vzeroupper
201; CHECK-NEXT:    retq
202  br i1 undef, label %if.true, label %if.false
203
204if.true:
205  %3 = add i16 %0, 1
206  br i1 undef, label %amx1, label %amx2
207
208if.false:
209  %4 = sub i16 %0, 1
210  br i1 undef, label %amx2, label %amx1
211
212amx1:
213  %5 = phi i16 [ %3, %if.true ], [ %4, %if.false ]
214  %6 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %5, i16 %1)
215  tail call void @llvm.x86.tilestored64.internal(i16 %5, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 0), i64 32, x86_amx %6)
216  br label %exit
217
218amx2:
219  %7 = phi i16 [ %3, %if.true ], [ %4, %if.false ]
220  %8 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %7, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 1024), i64 32)
221  tail call void @llvm.x86.tilestored64.internal(i16 %7, i16 %1, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 0), i64 32, x86_amx %8)
222  br label %exit
223
224exit:
225  ret void
226}
227
228define dso_local void @test5(i16 signext %0, i16 signext %1) nounwind {
229; CHECK-LABEL: test5:
230; CHECK:       # %bb.0: # %entry
231; CHECK-NEXT:    # kill: def $esi killed $esi def $rsi
232; CHECK-NEXT:    vpxord %zmm0, %zmm0, %zmm0
233; CHECK-NEXT:    vmovdqu64 %zmm0, -{{[0-9]+}}(%rsp)
234; CHECK-NEXT:    xorl %eax, %eax
235; CHECK-NEXT:    movl $buf, %r8d
236; CHECK-NEXT:    movl $32, %edx
237; CHECK-NEXT:    leal -1(%rsi), %ecx
238; CHECK-NEXT:    jmp .LBB4_1
239; CHECK-NEXT:    .p2align 4, 0x90
240; CHECK-NEXT:  .LBB4_3: # %if.false
241; CHECK-NEXT:    # in Loop: Header=BB4_1 Depth=1
242; CHECK-NEXT:    movl %ecx, %esi
243; CHECK-NEXT:    cmpw $7, %si
244; CHECK-NEXT:    jne .LBB4_5
245; CHECK-NEXT:  .LBB4_1: # %loop.bb1
246; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
247; CHECK-NEXT:    movb $1, -{{[0-9]+}}(%rsp)
248; CHECK-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
249; CHECK-NEXT:    movw %si, -{{[0-9]+}}(%rsp)
250; CHECK-NEXT:    ldtilecfg -{{[0-9]+}}(%rsp)
251; CHECK-NEXT:    testb %al, %al
252; CHECK-NEXT:    jne .LBB4_3
253; CHECK-NEXT:  # %bb.2: # %if.true
254; CHECK-NEXT:    # in Loop: Header=BB4_1 Depth=1
255; CHECK-NEXT:    tilezero %tmm0
256; CHECK-NEXT:    tilestored %tmm0, (%r8,%rdx)
257; CHECK-NEXT:    cmpw $7, %si
258; CHECK-NEXT:    je .LBB4_1
259; CHECK-NEXT:  .LBB4_5: # %exit
260; CHECK-NEXT:    tilerelease
261; CHECK-NEXT:    vzeroupper
262; CHECK-NEXT:    retq
263entry:
264  br label %loop.bb1
265
266loop.bb1:
267  %2 = phi i16 [ %1, %entry ], [ %5, %loop.bb2 ]
268  br i1 undef, label %if.true, label %if.false
269
270if.true:
271  %3 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %0, i16 %2)
272  tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %2, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 0), i64 32, x86_amx %3)
273  br label %loop.bb2
274
275if.false:
276  %4 = sub i16 %1, 1
277  br label %loop.bb2
278
279loop.bb2:
280  %5 = phi i16 [ %2, %if.true ], [ %4, %if.false ]
281  %6 = icmp eq i16 %5, 7
282  br i1 %6, label %loop.bb1, label %exit
283
284exit:
285  ret void
286}
287
288define dso_local void @test6(i16 signext %0) nounwind {
289; CHECK-LABEL: test6:
290; CHECK:       # %bb.0: # %entry
291; CHECK-NEXT:    # kill: def $edi killed $edi def $rdi
292; CHECK-NEXT:    vpxord %zmm0, %zmm0, %zmm0
293; CHECK-NEXT:    vmovdqu64 %zmm0, -{{[0-9]+}}(%rsp)
294; CHECK-NEXT:    xorl %r8d, %r8d
295; CHECK-NEXT:    movl $buf, %ecx
296; CHECK-NEXT:    movl $32, %edx
297; CHECK-NEXT:    xorl %esi, %esi
298; CHECK-NEXT:    jmp .LBB5_1
299; CHECK-NEXT:    .p2align 4, 0x90
300; CHECK-NEXT:  .LBB5_3: # %if.false
301; CHECK-NEXT:    # in Loop: Header=BB5_1 Depth=1
302; CHECK-NEXT:    decl %esi
303; CHECK-NEXT:  .LBB5_4: # %loop.bb2
304; CHECK-NEXT:    # in Loop: Header=BB5_1 Depth=1
305; CHECK-NEXT:    leal (%rdi,%rsi), %eax
306; CHECK-NEXT:    cmpw $7, %si
307; CHECK-NEXT:    movb $1, -{{[0-9]+}}(%rsp)
308; CHECK-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
309; CHECK-NEXT:    movw %ax, -{{[0-9]+}}(%rsp)
310; CHECK-NEXT:    ldtilecfg -{{[0-9]+}}(%rsp)
311; CHECK-NEXT:    tilezero %tmm0
312; CHECK-NEXT:    tilestored %tmm0, (%rcx,%rdx)
313; CHECK-NEXT:    jne .LBB5_5
314; CHECK-NEXT:  .LBB5_1: # %loop.bb1
315; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
316; CHECK-NEXT:    testb %r8b, %r8b
317; CHECK-NEXT:    jne .LBB5_3
318; CHECK-NEXT:  # %bb.2: # %if.true
319; CHECK-NEXT:    # in Loop: Header=BB5_1 Depth=1
320; CHECK-NEXT:    incl %esi
321; CHECK-NEXT:    jmp .LBB5_4
322; CHECK-NEXT:  .LBB5_5: # %exit
323; CHECK-NEXT:    tilerelease
324; CHECK-NEXT:    vzeroupper
325; CHECK-NEXT:    retq
326entry:
327  br label %loop.bb1
328
329loop.bb1:
330  %1 = phi i16 [ 0, %entry ], [ %4, %loop.bb2 ]
331  br i1 undef, label %if.true, label %if.false
332
333if.true:
334  %2 = add i16 %1, 1
335  br label %loop.bb2
336
337if.false:
338  %3 = sub i16 %1, 1
339  br label %loop.bb2
340
341loop.bb2:
342  %4 = phi i16 [ %2, %if.true ], [ %3, %if.false ]
343  %5 = icmp eq i16 %4, 7
344  %6 = add i16 %0, %4
345  %7 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %0, i16 %6)
346  tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %6, i8* getelementptr inbounds ([3072 x i8], [3072 x i8]* @buf, i64 0, i64 0), i64 32, x86_amx %7)
347  br i1 %5, label %loop.bb1, label %exit
348
349exit:
350  ret void
351}
352
353
354declare dso_local void @foo() nounwind
355declare void @llvm.dbg.value(metadata, metadata, metadata)
356declare x86_amx @llvm.x86.tilezero.internal(i16, i16)
357declare x86_amx @llvm.x86.tileloadd64.internal(i16, i16, i8*, i64)
358declare x86_amx @llvm.x86.tdpbssd.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)
359declare void @llvm.x86.tilestored64.internal(i16, i16, i8*, i64, x86_amx)
360
361!llvm.dbg.cu = !{!0}
362!llvm.module.flags = !{!1}
363
364!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !DIFile(filename: "1", directory: "1"))
365!1 = !{i32 2, !"Debug Info Version", i32 3}
366!2 = distinct !DISubprogram(unit: !0)
367!3 = !DILocation(line: 1, column: 1, scope: !2)
368