[AMDGPU] Handle S_GETREG_B32 hazard on gfx1250 (#153848)

GFX1250 SPG says: S_GETREG_B32 does not wait for idle before executing.
The user must S_WAIT_ALU 0 before S_GETREG_B32 on:
STATUS, STATE_PRIV, EXCP_FLAG_PRIV, or EXCP_FLAG_USER.

Author: rampitec

llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp

@@ -1200,6 +1200,8 @@ void GCNHazardRecognizer::fixHazards(MachineInstr *MI) {
   fixShift64HighRegBug(MI);
   fixVALUMaskWriteHazard(MI);
   fixRequiredExportPriority(MI);
+  if (ST.requiresWaitIdleBeforeGetReg())
+    fixGetRegWaitIdle(MI);
 }
 
 static bool isVCmpXWritesExec(const SIInstrInfo &TII, const SIRegisterInfo &TRI,
@@ -3428,3 +3430,24 @@ bool GCNHazardRecognizer::fixRequiredExportPriority(MachineInstr *MI) {
 
   return true;
 }
+
+bool GCNHazardRecognizer::fixGetRegWaitIdle(MachineInstr *MI) {
+  if (!isSGetReg(MI->getOpcode()))
+    return false;
+
+  const SIInstrInfo *TII = ST.getInstrInfo();
+  switch (getHWReg(TII, *MI)) {
+  default:
+    return false;
+  case AMDGPU::Hwreg::ID_STATUS:
+  case AMDGPU::Hwreg::ID_STATE_PRIV:
+  case AMDGPU::Hwreg::ID_EXCP_FLAG_PRIV:
+  case AMDGPU::Hwreg::ID_EXCP_FLAG_USER:
+    break;
+  }
+
+  BuildMI(*MI->getParent(), MI, MI->getDebugLoc(),
+          TII->get(AMDGPU::S_WAITCNT_DEPCTR))
+      .addImm(0);
+  return true;
+}

llvm/lib/Target/AMDGPU/GCNHazardRecognizer.h

@@ -110,6 +110,7 @@ class GCNHazardRecognizer final : public ScheduleHazardRecognizer {
   bool fixShift64HighRegBug(MachineInstr *MI);
   bool fixVALUMaskWriteHazard(MachineInstr *MI);
   bool fixRequiredExportPriority(MachineInstr *MI);
+  bool fixGetRegWaitIdle(MachineInstr *MI);
 
   int checkMAIHazards(MachineInstr *MI);
   int checkMAIHazards908(MachineInstr *MI);

llvm/lib/Target/AMDGPU/GCNSubtarget.h

@@ -1801,6 +1801,10 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
   // instruction before "s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)".
   bool requiresNopBeforeDeallocVGPRs() const { return !GFX1250Insts; }
 
+  // \returns true if the subtarget needs S_WAIT_ALU 0 before S_GETREG_B32 on
+  // STATUS, STATE_PRIV, EXCP_FLAG_PRIV, or EXCP_FLAG_USER.
+  bool requiresWaitIdleBeforeGetReg() const { return GFX1250Insts; }
+
   bool isDynamicVGPREnabled() const { return DynamicVGPR; }
   unsigned getDynamicVGPRBlockSize() const {
     return DynamicVGPRBlockSize32 ? 32 : 16;

llvm/test/CodeGen/AMDGPU/hazard-getreg-waitalu.mir

@@ -0,0 +1,91 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1250 -run-pass post-RA-hazard-rec -o - %s | FileCheck --check-prefix=GCN %s
+
+---
+name: s_getreg_mode
+body: |
+  bb.0:
+    ; GCN-LABEL: name: s_getreg_mode
+    ; GCN: $sgpr0 = S_GETREG_B32 1, implicit $mode
+    $sgpr0 = S_GETREG_B32 1, implicit $mode
+...
+
+---
+name: s_getreg_status
+body: |
+  bb.0:
+    ; GCN-LABEL: name: s_getreg_status
+    ; GCN: S_WAITCNT_DEPCTR 0
+    ; GCN-NEXT: $sgpr0 = S_GETREG_B32 2, implicit $mode
+    $sgpr0 = S_GETREG_B32 2, implicit $mode
+...
+
+---
+name: s_getreg_status_masked
+body: |
+  bb.0:
+    ; GCN-LABEL: name: s_getreg_status_masked
+    ; GCN: S_WAITCNT_DEPCTR 0
+    ; GCN-NEXT: $sgpr0 = S_GETREG_B32 66, implicit $mode
+    $sgpr0 = S_GETREG_B32 66, implicit $mode
+...
+
+---
+name: s_getreg_state_priv
+body: |
+  bb.0:
+    ; GCN-LABEL: name: s_getreg_state_priv
+    ; GCN: S_WAITCNT_DEPCTR 0
+    ; GCN-NEXT: $sgpr0 = S_GETREG_B32 4, implicit $mode
+    $sgpr0 = S_GETREG_B32 4, implicit $mode
+...
+
+---
+name: s_getreg_excp_flag_priv
+body: |
+  bb.0:
+    ; GCN-LABEL: name: s_getreg_excp_flag_priv
+    ; GCN: S_WAITCNT_DEPCTR 0
+    ; GCN-NEXT: $sgpr0 = S_GETREG_B32 17, implicit $mode
+    $sgpr0 = S_GETREG_B32 17, implicit $mode
+...
+
+---
+name: s_getreg_excp_flag_user
+body: |
+  bb.0:
+    ; GCN-LABEL: name: s_getreg_excp_flag_user
+    ; GCN: S_WAITCNT_DEPCTR 0
+    ; GCN-NEXT: $sgpr0 = S_GETREG_B32 18, implicit $mode
+    $sgpr0 = S_GETREG_B32 18, implicit $mode
+...
+
+---
+name: s_getreg_status_in_bundle
+body: |
+  bb.0:
+    ; GCN-LABEL: name: s_getreg_status_in_bundle
+    ; GCN: BUNDLE {
+    ; GCN-NEXT:   S_NOP 0
+    ; GCN-NEXT:   S_WAITCNT_DEPCTR 0
+    ; GCN-NEXT:   $sgpr0 = S_GETREG_B32 2, implicit $mode
+    ; GCN-NEXT: }
+    BUNDLE {
+      S_NOP 0
+      $sgpr0 = S_GETREG_B32 2, implicit $mode
+    }
+...
+
+---
+name: s_getreg_status_top_of_bundle
+body: |
+  bb.0:
+    ; GCN-LABEL: name: s_getreg_status_top_of_bundle
+    ; GCN: BUNDLE {
+    ; GCN-NEXT:   S_WAITCNT_DEPCTR 0
+    ; GCN-NEXT:   $sgpr0 = S_GETREG_B32 2, implicit $mode
+    ; GCN-NEXT: }
+    BUNDLE {
+      $sgpr0 = S_GETREG_B32 2, implicit $mode
+    }
+...