Fix tests

Author: lucas-rami

llvm/test/CodeGen/AMDGPU/buffer-fat-pointers-memcpy.ll

@@ -981,11 +981,11 @@ define amdgpu_kernel void @memcpy_known_medium(ptr addrspace(7) %src, ptr addrsp
 ; GISEL-GFX942-NEXT:    s_mov_b32 s2, s7
 ; GISEL-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-GFX942-NEXT:    s_or_b64 s[6:7], s[6:7], s[2:3]
-; GISEL-GFX942-NEXT:    v_mov_b32_e32 v0, 0x100
-; GISEL-GFX942-NEXT:    v_mov_b32_e32 v1, s16
+; GISEL-GFX942-NEXT:    v_mov_b32_e32 v0, s16
+; GISEL-GFX942-NEXT:    v_mov_b32_e32 v1, 0x100
 ; GISEL-GFX942-NEXT:  .LBB1_1: ; %load-store-loop
 ; GISEL-GFX942-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GISEL-GFX942-NEXT:    v_add_u32_e32 v62, s0, v1
+; GISEL-GFX942-NEXT:    v_add_u32_e32 v62, s0, v0
 ; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[2:5], v62, s[8:11], 0 offen
 ; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[6:9], v62, s[8:11], 0 offen offset:16
 ; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[10:13], v62, s[8:11], 0 offen offset:32
@@ -1002,9 +1002,9 @@ define amdgpu_kernel void @memcpy_known_medium(ptr addrspace(7) %src, ptr addrsp
 ; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[54:57], v62, s[8:11], 0 offen offset:208
 ; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[58:61], v62, s[8:11], 0 offen offset:224
 ; GISEL-GFX942-NEXT:    buffer_load_dwordx4 a[0:3], v62, s[8:11], 0 offen offset:240
-; GISEL-GFX942-NEXT:    v_add_u32_e32 v63, s12, v1
-; GISEL-GFX942-NEXT:    v_add_u32_e32 v1, 0x100, v1
-; GISEL-GFX942-NEXT:    v_cmp_lt_u32_e32 vcc, v1, v0
+; GISEL-GFX942-NEXT:    v_add_u32_e32 v63, s12, v0
+; GISEL-GFX942-NEXT:    v_add_u32_e32 v0, 0x100, v0
+; GISEL-GFX942-NEXT:    v_cmp_lt_u32_e32 vcc, v0, v1
 ; GISEL-GFX942-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-GFX942-NEXT:    scratch_store_dwordx4 off, a[0:3], off ; 16-byte Folded Spill
 ; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v63, s[4:7], 0 offen

llvm/test/CodeGen/AMDGPU/machine-scheduler-rematerialization-scoring.mir

@@ -28,29 +28,6 @@
 #           | 5 |
 #           +---+
 
---- |
-  define void @small_num_sgprs_as_spill() "amdgpu-num-sgpr"="85" {
-    ret void
-  }
-  define void @small_num_vgprs_as_spill() "amdgpu-num-vgpr"="14" {
-    ret void
-  }
-  define void @dont_remat_waves_per_eu() "amdgpu-flat-work-group-size"="1024,1024" "amdgpu-waves-per-eu"="7,7" {
-    ret void
-  }
-  define void @dont_remat_at_max_occ() "amdgpu-waves-per-eu"="8,8" {
-    ret void
-  }
-  define void @reduce_arch_and_acc_vgrp_spill() "amdgpu-waves-per-eu"="8,8" {
-    ret void
-  }
-  define void @reduce_spill_archvgpr_above_addressable_limit() "amdgpu-flat-work-group-size"="1,64" "amdgpu-waves-per-eu"="1,2" {
-    ret void
-  }
-  define void @reduce_spill_agpr_above_addressable_limit() "amdgpu-flat-work-group-size"="1,64" "amdgpu-waves-per-eu"="1,2" {
-    ret void
-  }
----
 # %32's defining and using region frequencies are identical therefore it is the
 # best register to rematerialize.
 name:            favor_same_frequency

llvm/test/tools/UpdateTestChecks/update_llc_test_checks/Inputs/amdgpu_generated_funcs.ll.generated.expected

@@ -83,11 +83,9 @@ attributes #0 = { noredzone nounwind ssp uwtable "frame-pointer"="all" }
 ; CHECK-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:8
 ; CHECK-NEXT:    buffer_store_dword v2, off, s[0:3], s33 offset:12
 ; CHECK-NEXT:    buffer_store_dword v3, off, s[0:3], s33 offset:16
-; CHECK-NEXT:    s_mov_b64 s[4:5], exec
-; CHECK-NEXT:    s_mov_b64 s[6:7], 0
-; CHECK-NEXT:    s_and_b64 s[6:7], s[4:5], s[6:7]
-; CHECK-NEXT:    s_xor_b64 s[4:5], s[6:7], s[4:5]
-; CHECK-NEXT:    s_mov_b64 exec, s[6:7]
+; CHECK-NEXT:    s_mov_b64 s[4:5], 0
+; CHECK-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
+; CHECK-NEXT:    s_xor_b64 s[4:5], exec, s[6:7]
 ; CHECK-NEXT:    s_cbranch_execz .LBB0_2
 ; CHECK-NEXT:  ; %bb.1:
 ; CHECK-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:4

llvm/test/tools/UpdateTestChecks/update_llc_test_checks/Inputs/amdgpu_generated_funcs.ll.nogenerated.expected

@@ -24,11 +24,9 @@ define dso_local i32 @check_boundaries() #0 {
 ; CHECK-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:8
 ; CHECK-NEXT:    buffer_store_dword v2, off, s[0:3], s33 offset:12
 ; CHECK-NEXT:    buffer_store_dword v3, off, s[0:3], s33 offset:16
-; CHECK-NEXT:    s_mov_b64 s[4:5], exec
-; CHECK-NEXT:    s_mov_b64 s[6:7], 0
-; CHECK-NEXT:    s_and_b64 s[6:7], s[4:5], s[6:7]
-; CHECK-NEXT:    s_xor_b64 s[4:5], s[6:7], s[4:5]
-; CHECK-NEXT:    s_mov_b64 exec, s[6:7]
+; CHECK-NEXT:    s_mov_b64 s[4:5], 0
+; CHECK-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
+; CHECK-NEXT:    s_xor_b64 s[4:5], exec, s[6:7]
 ; CHECK-NEXT:    s_cbranch_execz .LBB0_2
 ; CHECK-NEXT:  ; %bb.1:
 ; CHECK-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:4