Skip to content

Commit 777b6de

Browse files
saiislamsaiislam
authored
[AMDGPU][NFC] Test autogenerated llc tests for COV5 (#74339)
Regenerate a few llc tests to test for COV5 instead of the default ABI version.
1 parent 4204b94 commit 777b6de

23 files changed

+3894
-4248
lines changed

‎llvm/lib/Target/AMDGPU/SIISelLowering.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1791,7 +1791,7 @@ SDValue SITargetLowering::lowerKernArgParameterPtr(SelectionDAG &DAG,
17911791
// We may not have the kernarg segment argument if we have no kernel
17921792
// arguments.
17931793
if (!InputPtrReg)
1794-
return DAG.getConstant(0, SL, PtrVT);
1794+
return DAG.getConstant(Offset, SL, PtrVT);
17951795

17961796
MachineRegisterInfo &MRI = DAG.getMachineFunction().getRegInfo();
17971797
SDValue BasePtr = DAG.getCopyFromReg(Chain, SL,

‎llvm/test/CodeGen/AMDGPU/abi-attribute-hints-undefined-behavior.ll

Lines changed: 21 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -276,10 +276,13 @@ define void @addrspacecast_requires_queue_ptr(ptr addrspace(5) %ptr.private, ptr
276276
; FIXEDABI-SDAG-LABEL: addrspacecast_requires_queue_ptr:
277277
; FIXEDABI-SDAG: ; %bb.0:
278278
; FIXEDABI-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
279-
; FIXEDABI-SDAG-NEXT: s_load_dwordx2 s[4:5], s[6:7], 0x40
279+
; FIXEDABI-SDAG-NEXT: s_mov_b64 s[4:5], 0xc0
280+
; FIXEDABI-SDAG-NEXT: s_load_dword s6, s[4:5], 0x0
281+
; FIXEDABI-SDAG-NEXT: s_mov_b64 s[4:5], 0xc4
282+
; FIXEDABI-SDAG-NEXT: s_load_dword s4, s[4:5], 0x0
280283
; FIXEDABI-SDAG-NEXT: v_cmp_ne_u32_e32 vcc, -1, v0
281284
; FIXEDABI-SDAG-NEXT: s_waitcnt lgkmcnt(0)
282-
; FIXEDABI-SDAG-NEXT: v_mov_b32_e32 v2, s5
285+
; FIXEDABI-SDAG-NEXT: v_mov_b32_e32 v2, s6
283286
; FIXEDABI-SDAG-NEXT: v_cndmask_b32_e32 v3, 0, v2, vcc
284287
; FIXEDABI-SDAG-NEXT: v_cndmask_b32_e32 v2, 0, v0, vcc
285288
; FIXEDABI-SDAG-NEXT: v_mov_b32_e32 v0, s4
@@ -297,12 +300,15 @@ define void @addrspacecast_requires_queue_ptr(ptr addrspace(5) %ptr.private, ptr
297300
; FIXEDABI-GISEL-LABEL: addrspacecast_requires_queue_ptr:
298301
; FIXEDABI-GISEL: ; %bb.0:
299302
; FIXEDABI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
300-
; FIXEDABI-GISEL-NEXT: s_load_dwordx2 s[4:5], s[6:7], 0x40
303+
; FIXEDABI-GISEL-NEXT: s_mov_b64 s[4:5], 0xc0
304+
; FIXEDABI-GISEL-NEXT: s_load_dword s6, s[4:5], 0x0
305+
; FIXEDABI-GISEL-NEXT: s_mov_b64 s[4:5], 0xc4
306+
; FIXEDABI-GISEL-NEXT: s_load_dword s4, s[4:5], 0x0
301307
; FIXEDABI-GISEL-NEXT: v_cmp_ne_u32_e32 vcc, -1, v0
302308
; FIXEDABI-GISEL-NEXT: v_cndmask_b32_e32 v2, 0, v0, vcc
303309
; FIXEDABI-GISEL-NEXT: s_waitcnt lgkmcnt(0)
304-
; FIXEDABI-GISEL-NEXT: v_mov_b32_e32 v0, s5
305-
; FIXEDABI-GISEL-NEXT: v_cndmask_b32_e32 v3, 0, v0, vcc
310+
; FIXEDABI-GISEL-NEXT: v_mov_b32_e32 v3, s6
311+
; FIXEDABI-GISEL-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc
306312
; FIXEDABI-GISEL-NEXT: v_mov_b32_e32 v4, s4
307313
; FIXEDABI-GISEL-NEXT: v_cmp_ne_u32_e32 vcc, -1, v1
308314
; FIXEDABI-GISEL-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc
@@ -325,7 +331,8 @@ define void @is_shared_requires_queue_ptr(ptr %ptr) #0 {
325331
; FIXEDABI-LABEL: is_shared_requires_queue_ptr:
326332
; FIXEDABI: ; %bb.0:
327333
; FIXEDABI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
328-
; FIXEDABI-NEXT: s_load_dword s4, s[6:7], 0x40
334+
; FIXEDABI-NEXT: s_mov_b64 s[4:5], 0xc4
335+
; FIXEDABI-NEXT: s_load_dword s4, s[4:5], 0x0
329336
; FIXEDABI-NEXT: s_waitcnt lgkmcnt(0)
330337
; FIXEDABI-NEXT: v_cmp_eq_u32_e32 vcc, s4, v1
331338
; FIXEDABI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
@@ -342,7 +349,8 @@ define void @is_private_requires_queue_ptr(ptr %ptr) #0 {
342349
; FIXEDABI-LABEL: is_private_requires_queue_ptr:
343350
; FIXEDABI: ; %bb.0:
344351
; FIXEDABI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
345-
; FIXEDABI-NEXT: s_load_dword s4, s[6:7], 0x44
352+
; FIXEDABI-NEXT: s_mov_b64 s[4:5], 0xc0
353+
; FIXEDABI-NEXT: s_load_dword s4, s[4:5], 0x0
346354
; FIXEDABI-NEXT: s_waitcnt lgkmcnt(0)
347355
; FIXEDABI-NEXT: v_cmp_eq_u32_e32 vcc, s4, v1
348356
; FIXEDABI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
@@ -359,7 +367,9 @@ define void @trap_requires_queue() #0 {
359367
; FIXEDABI-LABEL: trap_requires_queue:
360368
; FIXEDABI: ; %bb.0:
361369
; FIXEDABI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
362-
; FIXEDABI-NEXT: s_mov_b64 s[0:1], s[6:7]
370+
; FIXEDABI-NEXT: s_mov_b64 s[4:5], 0xc8
371+
; FIXEDABI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
372+
; FIXEDABI-NEXT: s_waitcnt lgkmcnt(0)
363373
; FIXEDABI-NEXT: s_trap 2
364374
call void @llvm.trap()
365375
unreachable
@@ -390,3 +400,6 @@ declare void @llvm.trap()
390400
declare void @llvm.debugtrap()
391401

392402
attributes #0 = { "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-queue-ptr" "amdgpu-no-work-group-id-x" "amdgpu-no-work-group-id-y" "amdgpu-no-work-group-id-z" "amdgpu-no-work-item-id-x" "amdgpu-no-work-item-id-y" "amdgpu-no-work-item-id-z" }
403+
404+
!llvm.module.flags = !{!0}
405+
!0 = !{i32 1, !"amdgpu_code_object_version", i32 500}

‎llvm/test/CodeGen/AMDGPU/addrspacecast.gfx6.ll

Lines changed: 7 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -24,7 +24,8 @@ define ptr @group_to_flat_addrspacecast(ptr addrspace(3) %ptr) {
2424
; CHECK-LABEL: group_to_flat_addrspacecast:
2525
; CHECK: ; %bb.0:
2626
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
27-
; CHECK-NEXT: s_load_dword s4, s[6:7], 0x10
27+
; CHECK-NEXT: s_mov_b64 s[4:5], 0xc4
28+
; CHECK-NEXT: s_load_dword s4, s[4:5], 0x0
2829
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, -1, v0
2930
; CHECK-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
3031
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
@@ -50,7 +51,8 @@ define ptr @private_to_flat_addrspacecast(ptr addrspace(5) %ptr) {
5051
; CHECK-LABEL: private_to_flat_addrspacecast:
5152
; CHECK: ; %bb.0:
5253
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
53-
; CHECK-NEXT: s_load_dword s4, s[6:7], 0x11
54+
; CHECK-NEXT: s_mov_b64 s[4:5], 0xc0
55+
; CHECK-NEXT: s_load_dword s4, s[4:5], 0x0
5456
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, -1, v0
5557
; CHECK-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
5658
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
@@ -204,3 +206,6 @@ define ptr addrspace(6) @addrspacecast_flat_null_to_constant32bit() {
204206
}
205207

206208
attributes #0 = { "amdgpu-32bit-address-high-bits"="0xffff8000" }
209+
210+
!llvm.module.flags = !{!0}
211+
!0 = !{i32 1, !"amdgpu_code_object_version", i32 500}

‎llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pow-codegen.ll

Lines changed: 33 additions & 30 deletions
Original file line numberDiff line numberDiff line change
@@ -146,13 +146,13 @@ define double @test_pow_fast_f64__integral_y(double %x, i32 %y.i) {
146146
; CHECK-NEXT: s_mov_b64 s[4:5], s[40:41]
147147
; CHECK-NEXT: v_writelane_b32 v40, s45, 13
148148
; CHECK-NEXT: v_mov_b32_e32 v41, v31
149+
; CHECK-NEXT: s_mov_b64 s[34:35], s[6:7]
149150
; CHECK-NEXT: s_mov_b32 s42, s15
150151
; CHECK-NEXT: s_mov_b32 s43, s14
151152
; CHECK-NEXT: s_mov_b32 s44, s13
152153
; CHECK-NEXT: s_mov_b32 s45, s12
153-
; CHECK-NEXT: s_mov_b64 s[34:35], s[10:11]
154-
; CHECK-NEXT: s_mov_b64 s[36:37], s[8:9]
155-
; CHECK-NEXT: s_mov_b64 s[38:39], s[6:7]
154+
; CHECK-NEXT: s_mov_b64 s[36:37], s[10:11]
155+
; CHECK-NEXT: s_mov_b64 s[38:39], s[8:9]
156156
; CHECK-NEXT: v_mov_b32_e32 v42, v2
157157
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
158158
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
@@ -163,9 +163,9 @@ define double @test_pow_fast_f64__integral_y(double %x, i32 %y.i) {
163163
; CHECK-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
164164
; CHECK-NEXT: s_mov_b64 s[4:5], s[40:41]
165165
; CHECK-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
166-
; CHECK-NEXT: s_mov_b64 s[6:7], s[38:39]
167-
; CHECK-NEXT: s_mov_b64 s[8:9], s[36:37]
168-
; CHECK-NEXT: s_mov_b64 s[10:11], s[34:35]
166+
; CHECK-NEXT: s_mov_b64 s[6:7], s[34:35]
167+
; CHECK-NEXT: s_mov_b64 s[8:9], s[38:39]
168+
; CHECK-NEXT: s_mov_b64 s[10:11], s[36:37]
169169
; CHECK-NEXT: s_mov_b32 s12, s45
170170
; CHECK-NEXT: s_mov_b32 s13, s44
171171
; CHECK-NEXT: s_mov_b32 s14, s43
@@ -285,13 +285,13 @@ define double @test_powr_fast_f64(double %x, double %y) {
285285
; CHECK-NEXT: buffer_store_dword v43, off, s[0:3], s33 ; 4-byte Folded Spill
286286
; CHECK-NEXT: v_writelane_b32 v40, s45, 13
287287
; CHECK-NEXT: v_mov_b32_e32 v43, v31
288+
; CHECK-NEXT: s_mov_b64 s[34:35], s[6:7]
288289
; CHECK-NEXT: s_mov_b32 s42, s15
289290
; CHECK-NEXT: s_mov_b32 s43, s14
290291
; CHECK-NEXT: s_mov_b32 s44, s13
291292
; CHECK-NEXT: s_mov_b32 s45, s12
292-
; CHECK-NEXT: s_mov_b64 s[34:35], s[10:11]
293-
; CHECK-NEXT: s_mov_b64 s[36:37], s[8:9]
294-
; CHECK-NEXT: s_mov_b64 s[38:39], s[6:7]
293+
; CHECK-NEXT: s_mov_b64 s[36:37], s[10:11]
294+
; CHECK-NEXT: s_mov_b64 s[38:39], s[8:9]
295295
; CHECK-NEXT: v_mov_b32_e32 v42, v3
296296
; CHECK-NEXT: v_mov_b32_e32 v41, v2
297297
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
@@ -302,9 +302,9 @@ define double @test_powr_fast_f64(double %x, double %y) {
302302
; CHECK-NEXT: s_addc_u32 s5, s5, _Z4exp2d@gotpcrel32@hi+12
303303
; CHECK-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
304304
; CHECK-NEXT: s_mov_b64 s[4:5], s[40:41]
305-
; CHECK-NEXT: s_mov_b64 s[6:7], s[38:39]
306-
; CHECK-NEXT: s_mov_b64 s[8:9], s[36:37]
307-
; CHECK-NEXT: s_mov_b64 s[10:11], s[34:35]
305+
; CHECK-NEXT: s_mov_b64 s[6:7], s[34:35]
306+
; CHECK-NEXT: s_mov_b64 s[8:9], s[38:39]
307+
; CHECK-NEXT: s_mov_b64 s[10:11], s[36:37]
308308
; CHECK-NEXT: s_mov_b32 s12, s45
309309
; CHECK-NEXT: s_mov_b32 s13, s44
310310
; CHECK-NEXT: s_mov_b32 s14, s43
@@ -430,13 +430,13 @@ define double @test_pown_fast_f64(double %x, i32 %y) {
430430
; CHECK-NEXT: s_mov_b64 s[4:5], s[40:41]
431431
; CHECK-NEXT: v_writelane_b32 v40, s45, 13
432432
; CHECK-NEXT: v_mov_b32_e32 v41, v31
433+
; CHECK-NEXT: s_mov_b64 s[34:35], s[6:7]
433434
; CHECK-NEXT: s_mov_b32 s42, s15
434435
; CHECK-NEXT: s_mov_b32 s43, s14
435436
; CHECK-NEXT: s_mov_b32 s44, s13
436437
; CHECK-NEXT: s_mov_b32 s45, s12
437-
; CHECK-NEXT: s_mov_b64 s[34:35], s[10:11]
438-
; CHECK-NEXT: s_mov_b64 s[36:37], s[8:9]
439-
; CHECK-NEXT: s_mov_b64 s[38:39], s[6:7]
438+
; CHECK-NEXT: s_mov_b64 s[36:37], s[10:11]
439+
; CHECK-NEXT: s_mov_b64 s[38:39], s[8:9]
440440
; CHECK-NEXT: v_mov_b32_e32 v42, v2
441441
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
442442
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
@@ -447,9 +447,9 @@ define double @test_pown_fast_f64(double %x, i32 %y) {
447447
; CHECK-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
448448
; CHECK-NEXT: s_mov_b64 s[4:5], s[40:41]
449449
; CHECK-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
450-
; CHECK-NEXT: s_mov_b64 s[6:7], s[38:39]
451-
; CHECK-NEXT: s_mov_b64 s[8:9], s[36:37]
452-
; CHECK-NEXT: s_mov_b64 s[10:11], s[34:35]
450+
; CHECK-NEXT: s_mov_b64 s[6:7], s[34:35]
451+
; CHECK-NEXT: s_mov_b64 s[8:9], s[38:39]
452+
; CHECK-NEXT: s_mov_b64 s[10:11], s[36:37]
453453
; CHECK-NEXT: s_mov_b32 s12, s45
454454
; CHECK-NEXT: s_mov_b32 s13, s44
455455
; CHECK-NEXT: s_mov_b32 s14, s43
@@ -571,13 +571,13 @@ define double @test_pown_fast_f64_known_even(double %x, i32 %y.arg) {
571571
; CHECK-NEXT: buffer_store_dword v42, off, s[0:3], s33 ; 4-byte Folded Spill
572572
; CHECK-NEXT: v_writelane_b32 v40, s45, 13
573573
; CHECK-NEXT: v_mov_b32_e32 v41, v31
574+
; CHECK-NEXT: s_mov_b64 s[34:35], s[6:7]
574575
; CHECK-NEXT: s_mov_b32 s42, s15
575576
; CHECK-NEXT: s_mov_b32 s43, s14
576577
; CHECK-NEXT: s_mov_b32 s44, s13
577578
; CHECK-NEXT: s_mov_b32 s45, s12
578-
; CHECK-NEXT: s_mov_b64 s[34:35], s[10:11]
579-
; CHECK-NEXT: s_mov_b64 s[36:37], s[8:9]
580-
; CHECK-NEXT: s_mov_b64 s[38:39], s[6:7]
579+
; CHECK-NEXT: s_mov_b64 s[36:37], s[10:11]
580+
; CHECK-NEXT: s_mov_b64 s[38:39], s[8:9]
581581
; CHECK-NEXT: v_lshlrev_b32_e32 v42, 1, v2
582582
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
583583
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
@@ -588,9 +588,9 @@ define double @test_pown_fast_f64_known_even(double %x, i32 %y.arg) {
588588
; CHECK-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
589589
; CHECK-NEXT: s_mov_b64 s[4:5], s[40:41]
590590
; CHECK-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
591-
; CHECK-NEXT: s_mov_b64 s[6:7], s[38:39]
592-
; CHECK-NEXT: s_mov_b64 s[8:9], s[36:37]
593-
; CHECK-NEXT: s_mov_b64 s[10:11], s[34:35]
591+
; CHECK-NEXT: s_mov_b64 s[6:7], s[34:35]
592+
; CHECK-NEXT: s_mov_b64 s[8:9], s[38:39]
593+
; CHECK-NEXT: s_mov_b64 s[10:11], s[36:37]
594594
; CHECK-NEXT: s_mov_b32 s12, s45
595595
; CHECK-NEXT: s_mov_b32 s13, s44
596596
; CHECK-NEXT: s_mov_b32 s14, s43
@@ -715,13 +715,13 @@ define double @test_pown_fast_f64_known_odd(double %x, i32 %y.arg) {
715715
; CHECK-NEXT: s_mov_b64 s[4:5], s[40:41]
716716
; CHECK-NEXT: v_writelane_b32 v40, s45, 13
717717
; CHECK-NEXT: v_mov_b32_e32 v41, v31
718+
; CHECK-NEXT: s_mov_b64 s[34:35], s[6:7]
718719
; CHECK-NEXT: s_mov_b32 s42, s15
719720
; CHECK-NEXT: s_mov_b32 s43, s14
720721
; CHECK-NEXT: s_mov_b32 s44, s13
721722
; CHECK-NEXT: s_mov_b32 s45, s12
722-
; CHECK-NEXT: s_mov_b64 s[34:35], s[10:11]
723-
; CHECK-NEXT: s_mov_b64 s[36:37], s[8:9]
724-
; CHECK-NEXT: s_mov_b64 s[38:39], s[6:7]
723+
; CHECK-NEXT: s_mov_b64 s[36:37], s[10:11]
724+
; CHECK-NEXT: s_mov_b64 s[38:39], s[8:9]
725725
; CHECK-NEXT: v_or_b32_e32 v43, 1, v2
726726
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
727727
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
@@ -732,9 +732,9 @@ define double @test_pown_fast_f64_known_odd(double %x, i32 %y.arg) {
732732
; CHECK-NEXT: s_load_dwordx2 s[16:17], s[4:5], 0x0
733733
; CHECK-NEXT: s_mov_b64 s[4:5], s[40:41]
734734
; CHECK-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
735-
; CHECK-NEXT: s_mov_b64 s[6:7], s[38:39]
736-
; CHECK-NEXT: s_mov_b64 s[8:9], s[36:37]
737-
; CHECK-NEXT: s_mov_b64 s[10:11], s[34:35]
735+
; CHECK-NEXT: s_mov_b64 s[6:7], s[34:35]
736+
; CHECK-NEXT: s_mov_b64 s[8:9], s[38:39]
737+
; CHECK-NEXT: s_mov_b64 s[10:11], s[36:37]
738738
; CHECK-NEXT: s_mov_b32 s12, s45
739739
; CHECK-NEXT: s_mov_b32 s13, s44
740740
; CHECK-NEXT: s_mov_b32 s14, s43
@@ -773,3 +773,6 @@ define double @test_pown_fast_f64_known_odd(double %x, i32 %y.arg) {
773773
%call = tail call fast double @_Z4powndi(double %x, i32 %y)
774774
ret double %call
775775
}
776+
777+
!llvm.module.flags = !{!0}
778+
!0 = !{i32 1, !"amdgpu_code_object_version", i32 500}

0 commit comments

Comments
 (0)