AMDGPU: Fix temporal divergence introduced by machine-sink (llvm#67456)

Temporal divergence that was present in input or introduced in IR
transforms, like code-sinking or LICM, is handled in SIFixSGPRCopies
by changing sgpr source instr to vgpr instr.
After 5b657f5, that moved LICM after AMDGPUCodeGenPrepare,
machine-sinking can introduce temporal divergence by sinking
instructions outside of the cycle.
Add isSafeToSink callback in TargetInstrInfo.

Change-Id: I753744d3807cc05dfca687ed359fdceb9afa2fdc

Author: petar-avramovic

llvm/test/CodeGen/AMDGPU/machine-sink-temporal-divergence-swdev407790.ll

@@ -167,6 +167,7 @@ define protected amdgpu_kernel void @kernel_round1(ptr addrspace(1) nocapture no
 ; CHECK-NEXT:    s_or_b32 exec_lo, exec_lo, s59
 ; CHECK-NEXT:    s_add_i32 s58, s58, 4
 ; CHECK-NEXT:    s_add_i32 s4, s55, s58
+; CHECK-NEXT:    v_add_nc_u32_e32 v0, s58, v57
 ; CHECK-NEXT:    s_add_i32 s5, s4, 5
 ; CHECK-NEXT:    s_add_i32 s4, s4, 1
 ; CHECK-NEXT:    v_cmp_ge_u32_e32 vcc_lo, s5, v42
@@ -267,7 +268,7 @@ define protected amdgpu_kernel void @kernel_round1(ptr addrspace(1) nocapture no
 ; CHECK-NEXT:  .LBB0_16: ; %Flow43
 ; CHECK-NEXT:    ; in Loop: Header=BB0_5 Depth=1
 ; CHECK-NEXT:    s_or_b32 exec_lo, exec_lo, s57
-; CHECK-NEXT:    v_add_nc_u32_e32 v57, s58, v57
+; CHECK-NEXT:    v_mov_b32_e32 v57, v0
 ; CHECK-NEXT:  .LBB0_17: ; %Flow44
 ; CHECK-NEXT:    ; in Loop: Header=BB0_5 Depth=1
 ; CHECK-NEXT:    s_or_b32 exec_lo, exec_lo, s56
@@ -869,6 +870,7 @@ define protected amdgpu_kernel void @kernel_round1_short(ptr addrspace(1) nocapt
 ; CHECK-NEXT:    s_add_i32 s7, s7, 4
 ; CHECK-NEXT:    v_add_nc_u32_e32 v43, 1, v43
 ; CHECK-NEXT:    s_add_i32 s8, s4, s7
+; CHECK-NEXT:    v_add_nc_u32_e32 v0, s7, v47
 ; CHECK-NEXT:    s_add_i32 s9, s8, 5
 ; CHECK-NEXT:    s_add_i32 s8, s8, 1
 ; CHECK-NEXT:    v_cmp_ge_u32_e32 vcc_lo, s9, v41
@@ -879,7 +881,7 @@ define protected amdgpu_kernel void @kernel_round1_short(ptr addrspace(1) nocapt
 ; CHECK-NEXT:  ; %bb.4: ; %Flow3
 ; CHECK-NEXT:    ; in Loop: Header=BB1_1 Depth=1
 ; CHECK-NEXT:    s_or_b32 exec_lo, exec_lo, s6
-; CHECK-NEXT:    v_add_nc_u32_e32 v47, s7, v47
+; CHECK-NEXT:    v_mov_b32_e32 v47, v0
 ; CHECK-NEXT:  .LBB1_5: ; %Flow4
 ; CHECK-NEXT:    ; in Loop: Header=BB1_1 Depth=1
 ; CHECK-NEXT:    s_or_b32 exec_lo, exec_lo, s5

llvm/test/CodeGen/AMDGPU/machine-sink-temporal-divergence-swdev407790.mir

@@ -22,6 +22,7 @@ body: |
   ; CHECK-NEXT:   [[PHI:%[0-9]+]]:sreg_32 = PHI [[S_MOV_B32_1]], %bb.0, %6, %bb.1
   ; CHECK-NEXT:   [[PHI1:%[0-9]+]]:sreg_32 = PHI [[S_MOV_B32_]], %bb.0, %8, %bb.1
   ; CHECK-NEXT:   [[S_ADD_I32_:%[0-9]+]]:sreg_32 = S_ADD_I32 [[PHI1]], [[S_MOV_B32_2]], implicit-def dead $scc
+  ; CHECK-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[S_ADD_I32_]], [[S_ADD_I32_]], 0, implicit $exec
   ; CHECK-NEXT:   [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[S_ADD_I32_]], 0, 0, implicit $mode, implicit $exec
   ; CHECK-NEXT:   [[V_CMP_GT_F32_e64_:%[0-9]+]]:sreg_32 = nofpexcept V_CMP_GT_F32_e64 0, killed [[V_CVT_F32_U32_e64_]], 0, [[COPY]], 0, implicit $mode, implicit $exec
   ; CHECK-NEXT:   [[SI_IF_BREAK:%[0-9]+]]:sreg_32 = SI_IF_BREAK killed [[V_CMP_GT_F32_e64_]], [[PHI]], implicit-def dead $scc
@@ -30,7 +31,6 @@ body: |
   ; CHECK-NEXT: {{  $}}
   ; CHECK-NEXT: bb.2:
   ; CHECK-NEXT:   SI_END_CF [[SI_IF_BREAK]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec
-  ; CHECK-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[S_ADD_I32_]], [[S_ADD_I32_]], 0, implicit $exec
   ; CHECK-NEXT:   FLAT_STORE_DWORD [[COPY1]], [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32))
   ; CHECK-NEXT:   SI_RETURN
   bb.0:
@@ -83,6 +83,9 @@ body: |
   ; CHECK-NEXT:   [[PHI:%[0-9]+]]:sreg_32 = PHI [[S_MOV_B32_1]], %bb.0, %6, %bb.1
   ; CHECK-NEXT:   [[PHI1:%[0-9]+]]:sreg_32 = PHI [[S_MOV_B32_]], %bb.0, %8, %bb.1
   ; CHECK-NEXT:   [[S_ADD_I32_:%[0-9]+]]:sreg_32 = S_ADD_I32 [[PHI1]], [[S_MOV_B32_2]], implicit-def dead $scc
+  ; CHECK-NEXT:   [[S_ADD_I32_1:%[0-9]+]]:sreg_32 = S_ADD_I32 [[S_ADD_I32_]], [[S_MOV_B32_2]], implicit-def dead $scc
+  ; CHECK-NEXT:   [[S_ADD_I32_2:%[0-9]+]]:sreg_32 = S_ADD_I32 [[S_ADD_I32_1]], [[S_MOV_B32_2]], implicit-def dead $scc
+  ; CHECK-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[S_ADD_I32_2]], [[S_ADD_I32_2]], 0, implicit $exec
   ; CHECK-NEXT:   [[V_CVT_F32_U32_e64_:%[0-9]+]]:vgpr_32 = V_CVT_F32_U32_e64 [[S_ADD_I32_]], 0, 0, implicit $mode, implicit $exec
   ; CHECK-NEXT:   [[V_CMP_GT_F32_e64_:%[0-9]+]]:sreg_32 = nofpexcept V_CMP_GT_F32_e64 0, killed [[V_CVT_F32_U32_e64_]], 0, [[COPY]], 0, implicit $mode, implicit $exec
   ; CHECK-NEXT:   [[SI_IF_BREAK:%[0-9]+]]:sreg_32 = SI_IF_BREAK killed [[V_CMP_GT_F32_e64_]], [[PHI]], implicit-def dead $scc
@@ -91,9 +94,6 @@ body: |
   ; CHECK-NEXT: {{  $}}
   ; CHECK-NEXT: bb.2:
   ; CHECK-NEXT:   SI_END_CF [[SI_IF_BREAK]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec
-  ; CHECK-NEXT:   [[S_ADD_I32_1:%[0-9]+]]:sreg_32 = S_ADD_I32 [[S_ADD_I32_]], [[S_MOV_B32_2]], implicit-def dead $scc
-  ; CHECK-NEXT:   [[S_ADD_I32_2:%[0-9]+]]:sreg_32 = S_ADD_I32 [[S_ADD_I32_1]], [[S_MOV_B32_2]], implicit-def dead $scc
-  ; CHECK-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[S_ADD_I32_2]], [[S_ADD_I32_2]], 0, implicit $exec
   ; CHECK-NEXT:   FLAT_STORE_DWORD [[COPY1]], [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32))
   ; CHECK-NEXT:   SI_RETURN
   bb.0: