AMDGPU: Try constant fold after folding immediate (llvm#141862)

arsenm · tomtor · commit c61896d7e765 · 2025-06-14T19:48:09.000+02:00
This helps avoid some regressions in a future patch. The or 0
pattern appears in the division tests because the reduce 64-bit
bit operation to a 32-bit one with half identity value is only
implemented for constants. We could fix that by using computeKnownBits.
Additionally the pattern disappears if I optimize the IR division
expansion, so that IR should probably be emitted more optimally in
the first place.
diff --git a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -1675,6 +1675,12 @@ bool SIFoldOperandsImpl::foldInstOperand(MachineInstr &MI,
       LLVM_DEBUG(dbgs() << "Folded source from " << MI << " into OpNo "
                         << static_cast<int>(Fold.UseOpNo) << " of "
                         << *Fold.UseMI);
+
+      if (Fold.isImm() && tryConstantFoldOp(Fold.UseMI)) {
+        LLVM_DEBUG(dbgs() << "Constant folded " << *Fold.UseMI);
+        Changed = true;
+      }
+
     } else if (Fold.Commuted) {
       // Restoring instruction's original operand order if fold has failed.
       TII->commuteInstruction(*Fold.UseMI, false);
diff --git a/llvm/test/CodeGen/AMDGPU/bit-op-reduce-width-known-bits.ll b/llvm/test/CodeGen/AMDGPU/bit-op-reduce-width-known-bits.ll
@@ -105,9 +105,8 @@ define i64 @v_xor_i64_known_i32_from_range_use_out_of_block(i64 %x) {
 ; CHECK-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; CHECK-NEXT:  ; %bb.1: ; %inc
 ; CHECK-NEXT:    v_not_b32_e32 v2, v4
-; CHECK-NEXT:    v_not_b32_e32 v3, 0
 ; CHECK-NEXT:    v_add_co_u32_e32 v2, vcc, v0, v2
-; CHECK-NEXT:    v_addc_co_u32_e32 v3, vcc, v1, v3, vcc
+; CHECK-NEXT:    v_addc_co_u32_e32 v3, vcc, -1, v1, vcc
 ; CHECK-NEXT:  ; %bb.2: ; %UnifiedReturnBlock
 ; CHECK-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; CHECK-NEXT:    v_mov_b32_e32 v0, v2
diff --git a/llvm/test/CodeGen/AMDGPU/constant-fold-imm-immreg.mir b/llvm/test/CodeGen/AMDGPU/constant-fold-imm-immreg.mir
@@ -961,3 +961,25 @@ body:             |
     S_ENDPGM 0, implicit %2, implicit %3
 
 ...
+
+---
+name:            constant_v_or_b32_uses_subreg_or_0_regression
+tracksRegLiveness: true
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1
+
+    ; GCN-LABEL: name: constant_v_or_b32_uses_subreg_or_0_regression
+    ; GCN: liveins: $vgpr0, $vgpr1
+    ; GCN-NEXT: {{  $}}
+    ; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; GCN-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[COPY]]
+    ; GCN-NEXT: S_ENDPGM 0, implicit [[COPY1]]
+  %0:vgpr_32 = COPY $vgpr0
+  %1:vgpr_32 = COPY $vgpr1
+  %2:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+  %3:vreg_64 = REG_SEQUENCE %2:vgpr_32, %subreg.sub0, %0:vgpr_32, %subreg.sub1
+  %4:vgpr_32 = V_OR_B32_e64 %3.sub0:vreg_64, %1, implicit $exec
+  S_ENDPGM 0, implicit %4
+
+...
diff --git a/llvm/test/CodeGen/AMDGPU/fold-imm-copy.mir b/llvm/test/CodeGen/AMDGPU/fold-imm-copy.mir
@@ -43,8 +43,7 @@ body:             |
     ; GCN-NEXT: [[DEF2:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
     ; GCN-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
     ; GCN-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE killed [[DEF]], %subreg.sub0, killed [[V_MOV_B32_e32_]], %subreg.sub1
-    ; GCN-NEXT: [[V_XOR_B32_e32_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e32 0, [[DEF1]], implicit $exec
-    ; GCN-NEXT: [[V_XOR_B32_e32_1:%[0-9]+]]:vgpr_32 = V_XOR_B32_e32 [[DEF2]], [[REG_SEQUENCE]].sub0, implicit $exec
+    ; GCN-NEXT: [[V_XOR_B32_e32_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e32 [[DEF2]], [[REG_SEQUENCE]].sub0, implicit $exec
     %0:vgpr_32 = IMPLICIT_DEF
     %1:vgpr_32 = IMPLICIT_DEF
     %2:vgpr_32 = IMPLICIT_DEF
diff --git a/llvm/test/CodeGen/AMDGPU/fold-zero-high-bits-skips-non-reg.mir b/llvm/test/CodeGen/AMDGPU/fold-zero-high-bits-skips-non-reg.mir
@@ -8,8 +8,8 @@ body: |
     ; CHECK-LABEL: name: test_tryFoldZeroHighBits_skips_nonreg
     ; CHECK: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
     ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_]], %subreg.sub1
-    ; CHECK-NEXT: [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 65535, 0, implicit $exec
-    ; CHECK-NEXT: S_NOP 0, implicit [[V_AND_B32_e64_]]
+    ; CHECK-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+    ; CHECK-NEXT: S_NOP 0, implicit [[V_MOV_B32_e32_1]]
   %0:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
   %1:vreg_64 = REG_SEQUENCE %0, %subreg.sub0, %0, %subreg.sub1
   %2:vgpr_32 = V_AND_B32_e64 65535, %1.sub0, implicit $exec
diff --git a/llvm/test/CodeGen/AMDGPU/sdiv64.ll b/llvm/test/CodeGen/AMDGPU/sdiv64.ll
@@ -404,12 +404,11 @@ define i64 @v_test_sdiv(i64 %x, i64 %y) {
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
 ; GCN-IR-NEXT:    v_add_i32_e32 v16, vcc, -1, v0
 ; GCN-IR-NEXT:    v_addc_u32_e32 v17, vcc, -1, v1, vcc
-; GCN-IR-NEXT:    v_not_b32_e32 v5, v10
+; GCN-IR-NEXT:    v_not_b32_e32 v4, v10
 ; GCN-IR-NEXT:    v_lshr_b64 v[8:9], v[6:7], v8
-; GCN-IR-NEXT:    v_not_b32_e32 v4, 0
-; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, v5, v11
+; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, v4, v11
 ; GCN-IR-NEXT:    v_mov_b32_e32 v10, 0
-; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, 0, v4, vcc
+; GCN-IR-NEXT:    v_addc_u32_e64 v7, s[4:5], -1, 0, vcc
 ; GCN-IR-NEXT:    s_mov_b64 s[10:11], 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v11, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v5, 0
diff --git a/llvm/test/CodeGen/AMDGPU/srem64.ll b/llvm/test/CodeGen/AMDGPU/srem64.ll
@@ -380,12 +380,11 @@ define i64 @v_test_srem(i64 %x, i64 %y) {
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
 ; GCN-IR-NEXT:    v_add_i32_e32 v16, vcc, -1, v2
 ; GCN-IR-NEXT:    v_addc_u32_e32 v17, vcc, -1, v3, vcc
-; GCN-IR-NEXT:    v_not_b32_e32 v7, v12
+; GCN-IR-NEXT:    v_not_b32_e32 v6, v12
 ; GCN-IR-NEXT:    v_lshr_b64 v[10:11], v[0:1], v8
-; GCN-IR-NEXT:    v_not_b32_e32 v6, 0
-; GCN-IR-NEXT:    v_add_i32_e32 v8, vcc, v7, v13
+; GCN-IR-NEXT:    v_add_i32_e32 v8, vcc, v6, v13
 ; GCN-IR-NEXT:    v_mov_b32_e32 v12, 0
-; GCN-IR-NEXT:    v_addc_u32_e32 v9, vcc, 0, v6, vcc
+; GCN-IR-NEXT:    v_addc_u32_e64 v9, s[4:5], -1, 0, vcc
 ; GCN-IR-NEXT:    s_mov_b64 s[10:11], 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v13, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
diff --git a/llvm/test/CodeGen/AMDGPU/udiv64.ll b/llvm/test/CodeGen/AMDGPU/udiv64.ll
@@ -348,10 +348,9 @@ define i64 @v_test_udiv_i64(i64 %x, i64 %y) {
 ; GCN-IR-NEXT:    v_lshr_b64 v[8:9], v[0:1], v10
 ; GCN-IR-NEXT:    v_addc_u32_e32 v13, vcc, -1, v3, vcc
 ; GCN-IR-NEXT:    v_not_b32_e32 v0, v14
-; GCN-IR-NEXT:    v_not_b32_e32 v1, 0
 ; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v15
 ; GCN-IR-NEXT:    v_mov_b32_e32 v10, 0
-; GCN-IR-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GCN-IR-NEXT:    v_addc_u32_e64 v1, s[4:5], -1, 0, vcc
 ; GCN-IR-NEXT:    s_mov_b64 s[10:11], 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v11, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
diff --git a/llvm/test/CodeGen/AMDGPU/urem64.ll b/llvm/test/CodeGen/AMDGPU/urem64.ll
@@ -355,12 +355,11 @@ define i64 @v_test_urem_i64(i64 %x, i64 %y) {
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
 ; GCN-IR-NEXT:    v_add_i32_e32 v14, vcc, -1, v2
 ; GCN-IR-NEXT:    v_addc_u32_e32 v15, vcc, -1, v3, vcc
-; GCN-IR-NEXT:    v_not_b32_e32 v7, v12
+; GCN-IR-NEXT:    v_not_b32_e32 v6, v12
 ; GCN-IR-NEXT:    v_lshr_b64 v[10:11], v[0:1], v8
-; GCN-IR-NEXT:    v_not_b32_e32 v6, 0
-; GCN-IR-NEXT:    v_add_i32_e32 v8, vcc, v7, v13
+; GCN-IR-NEXT:    v_add_i32_e32 v8, vcc, v6, v13
 ; GCN-IR-NEXT:    v_mov_b32_e32 v12, 0
-; GCN-IR-NEXT:    v_addc_u32_e32 v9, vcc, 0, v6, vcc
+; GCN-IR-NEXT:    v_addc_u32_e64 v9, s[4:5], -1, 0, vcc
 ; GCN-IR-NEXT:    s_mov_b64 s[10:11], 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v13, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
