[AMDGPU][GlobalISel] Add register bank legalization for G_SMIN/G_SMAX/G_UMIN/G_UMAX (llvm#159821)

mssefat · web-flow · commit 167c00e4a4b4 · 2025-10-10T14:34:15.000+02:00
[AMDGPU][GlobalISel] Add register bank legalization for
G_SMIN/G_SMAX/G_UMIN/G_UMAX

This patch adds register bank legalization support for min/max
operations in the AMDGPU GlobalISel pipeline.

- Add support S16, S32, and V2S16 types
- For V2S16 uniform operations implements UnpackMinMax lowering
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
@@ -468,6 +468,38 @@ void RegBankLegalizeHelper::lowerUnpackBitShift(MachineInstr &MI) {
   MI.eraseFromParent();
 }
 
+void RegBankLegalizeHelper::lowerUnpackMinMax(MachineInstr &MI) {
+  Register Lo, Hi;
+  switch (MI.getOpcode()) {
+  case AMDGPU::G_SMIN:
+  case AMDGPU::G_SMAX: {
+    // For signed operations, use sign extension
+    auto [Val0_Lo, Val0_Hi] = unpackSExt(MI.getOperand(1).getReg());
+    auto [Val1_Lo, Val1_Hi] = unpackSExt(MI.getOperand(2).getReg());
+    Lo = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {Val0_Lo, Val1_Lo})
+             .getReg(0);
+    Hi = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {Val0_Hi, Val1_Hi})
+             .getReg(0);
+    break;
+  }
+  case AMDGPU::G_UMIN:
+  case AMDGPU::G_UMAX: {
+    // For unsigned operations, use zero extension
+    auto [Val0_Lo, Val0_Hi] = unpackZExt(MI.getOperand(1).getReg());
+    auto [Val1_Lo, Val1_Hi] = unpackZExt(MI.getOperand(2).getReg());
+    Lo = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {Val0_Lo, Val1_Lo})
+             .getReg(0);
+    Hi = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {Val0_Hi, Val1_Hi})
+             .getReg(0);
+    break;
+  }
+  default:
+    llvm_unreachable("Unpack min/max lowering not implemented");
+  }
+  B.buildBuildVectorTrunc(MI.getOperand(0).getReg(), {Lo, Hi});
+  MI.eraseFromParent();
+}
+
 static bool isSignedBFE(MachineInstr &MI) {
   if (GIntrinsic *GI = dyn_cast<GIntrinsic>(&MI))
     return (GI->is(Intrinsic::amdgcn_sbfe));
@@ -654,6 +686,8 @@ void RegBankLegalizeHelper::lower(MachineInstr &MI,
   }
   case UnpackBitShift:
     return lowerUnpackBitShift(MI);
+  case UnpackMinMax:
+    return lowerUnpackMinMax(MI);
   case Ext32To64: {
     const RegisterBank *RB = MRI.getRegBank(MI.getOperand(0).getReg());
     MachineInstrBuilder Hi;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h
@@ -123,6 +123,7 @@ class RegBankLegalizeHelper {
   void lowerSplitTo32(MachineInstr &MI);
   void lowerSplitTo32Select(MachineInstr &MI);
   void lowerSplitTo32SExtInReg(MachineInstr &MI);
+  void lowerUnpackMinMax(MachineInstr &MI);
 };
 
 } // end namespace AMDGPU
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
@@ -522,6 +522,22 @@ RegBankLegalizeRules::RegBankLegalizeRules(const GCNSubtarget &_ST,
       .Uni(S64, {{Sgpr64}, {Sgpr64, Sgpr32, Sgpr32}, S_BFE})
       .Div(S64, {{Vgpr64}, {Vgpr64, Vgpr32, Vgpr32}, V_BFE});
 
+  addRulesForGOpcs({G_SMIN, G_SMAX}, Standard)
+      .Uni(S16, {{Sgpr32Trunc}, {Sgpr32SExt, Sgpr32SExt}})
+      .Div(S16, {{Vgpr16}, {Vgpr16, Vgpr16}})
+      .Uni(S32, {{Sgpr32}, {Sgpr32, Sgpr32}})
+      .Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32}})
+      .Uni(V2S16, {{SgprV2S16}, {SgprV2S16, SgprV2S16}, UnpackMinMax})
+      .Div(V2S16, {{VgprV2S16}, {VgprV2S16, VgprV2S16}});
+
+  addRulesForGOpcs({G_UMIN, G_UMAX}, Standard)
+      .Uni(S16, {{Sgpr32Trunc}, {Sgpr32ZExt, Sgpr32ZExt}})
+      .Div(S16, {{Vgpr16}, {Vgpr16, Vgpr16}})
+      .Uni(S32, {{Sgpr32}, {Sgpr32, Sgpr32}})
+      .Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32}})
+      .Uni(V2S16, {{SgprV2S16}, {SgprV2S16, SgprV2S16}, UnpackMinMax})
+      .Div(V2S16, {{VgprV2S16}, {VgprV2S16, VgprV2S16}});
+
   // Note: we only write S1 rules for G_IMPLICIT_DEF, G_CONSTANT, G_FCONSTANT
   // and G_FREEZE here, rest is trivially regbankselected earlier
   addRulesForGOpcs({G_IMPLICIT_DEF}).Any({{UniS1}, {{Sgpr32Trunc}, {}}});
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h
@@ -212,6 +212,7 @@ enum LoweringMethodID {
   VccExtToSel,
   UniExtToSel,
   UnpackBitShift,
+  UnpackMinMax,
   S_BFE,
   V_BFE,
   VgprToVccCopy,
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smax.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smax.mir
@@ -1,6 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-fast -o - %s  | FileCheck %s
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-greedy -o - %s  | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass='amdgpu-regbankselect,amdgpu-regbanklegalize' -o - %s | FileCheck %s
 
 ---
 name: smax_s32_ss
@@ -188,8 +187,7 @@ body: |
     ; CHECK-NEXT: [[ASHR:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST]], [[C]](s32)
     ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>)
     ; CHECK-NEXT: [[SEXT_INREG1:%[0-9]+]]:sgpr(s32) = G_SEXT_INREG [[BITCAST1]], 16
-    ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
-    ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST1]], [[C1]](s32)
+    ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST1]], [[C]](s32)
     ; CHECK-NEXT: [[SMAX:%[0-9]+]]:sgpr(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]]
     ; CHECK-NEXT: [[SMAX1:%[0-9]+]]:sgpr(s32) = G_SMAX [[ASHR]], [[ASHR1]]
     ; CHECK-NEXT: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[SMAX]](s32), [[SMAX1]](s32)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smin.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smin.mir
@@ -1,6 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-fast -o - %s  | FileCheck %s
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-greedy -o - %s  | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass='amdgpu-regbankselect,amdgpu-regbanklegalize' -o - %s | FileCheck %s
 
 ---
 name: smin_s32_ss
@@ -191,8 +190,7 @@ body: |
     ; CHECK-NEXT: [[ASHR:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST]], [[C]](s32)
     ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>)
     ; CHECK-NEXT: [[SEXT_INREG1:%[0-9]+]]:sgpr(s32) = G_SEXT_INREG [[BITCAST1]], 16
-    ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
-    ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST1]], [[C1]](s32)
+    ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST1]], [[C]](s32)
     ; CHECK-NEXT: [[SMIN:%[0-9]+]]:sgpr(s32) = G_SMIN [[SEXT_INREG]], [[SEXT_INREG1]]
     ; CHECK-NEXT: [[SMIN1:%[0-9]+]]:sgpr(s32) = G_SMIN [[ASHR]], [[ASHR1]]
     ; CHECK-NEXT: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[SMIN]](s32), [[SMIN1]](s32)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umax.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umax.mir
@@ -1,6 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-fast -o - %s  | FileCheck %s
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-greedy -o - %s  | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass='amdgpu-regbankselect,amdgpu-regbanklegalize' -o - %s | FileCheck %s
 
 ---
 name: umax_s32_ss
@@ -186,15 +185,13 @@ body: |
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr1
     ; CHECK-NEXT: [[BITCAST:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY]](<2 x s16>)
-    ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
-    ; CHECK-NEXT: [[LSHR:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST]], [[C]](s32)
-    ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535
-    ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST]], [[C1]]
+    ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535
+    ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST]], [[C]]
+    ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
+    ; CHECK-NEXT: [[LSHR:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST]], [[C1]](s32)
     ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>)
-    ; CHECK-NEXT: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
-    ; CHECK-NEXT: [[LSHR1:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST1]], [[C2]](s32)
-    ; CHECK-NEXT: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535
-    ; CHECK-NEXT: [[AND1:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST1]], [[C3]]
+    ; CHECK-NEXT: [[AND1:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST1]], [[C]]
+    ; CHECK-NEXT: [[LSHR1:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST1]], [[C1]](s32)
     ; CHECK-NEXT: [[UMAX:%[0-9]+]]:sgpr(s32) = G_UMAX [[AND]], [[AND1]]
     ; CHECK-NEXT: [[UMAX1:%[0-9]+]]:sgpr(s32) = G_UMAX [[LSHR]], [[LSHR1]]
     ; CHECK-NEXT: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[UMAX]](s32), [[UMAX1]](s32)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umin.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umin.mir
@@ -1,6 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-fast -o - %s  | FileCheck %s
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-greedy -o - %s  | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass='amdgpu-regbankselect,amdgpu-regbanklegalize' -o - %s | FileCheck %s
 
 ---
 name: umin_s32_ss
@@ -190,15 +189,13 @@ body: |
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr1
     ; CHECK-NEXT: [[BITCAST:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY]](<2 x s16>)
-    ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
-    ; CHECK-NEXT: [[LSHR:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST]], [[C]](s32)
-    ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535
-    ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST]], [[C1]]
+    ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535
+    ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST]], [[C]]
+    ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
+    ; CHECK-NEXT: [[LSHR:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST]], [[C1]](s32)
     ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>)
-    ; CHECK-NEXT: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
-    ; CHECK-NEXT: [[LSHR1:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST1]], [[C2]](s32)
-    ; CHECK-NEXT: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535
-    ; CHECK-NEXT: [[AND1:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST1]], [[C3]]
+    ; CHECK-NEXT: [[AND1:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST1]], [[C]]
+    ; CHECK-NEXT: [[LSHR1:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST1]], [[C1]](s32)
     ; CHECK-NEXT: [[UMIN:%[0-9]+]]:sgpr(s32) = G_UMIN [[AND]], [[AND1]]
     ; CHECK-NEXT: [[UMIN1:%[0-9]+]]:sgpr(s32) = G_UMIN [[LSHR]], [[LSHR1]]
     ; CHECK-NEXT: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[UMIN]](s32), [[UMIN1]](s32)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx802 < %s | FileCheck -check-prefixes=GFX89,GFX8 %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX89,GFX9 %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx802 < %s | FileCheck -check-prefixes=GFX89,GFX8 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX89,GFX9 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
 
 define i32 @test_min_max_ValK0_K1_i32(i32 %a) {
 ; GFX89-LABEL: test_min_max_ValK0_K1_i32:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx802 < %s | FileCheck -check-prefixes=GFX89,GFX8 %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX89,GFX9 %s
-; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx802 < %s | FileCheck -check-prefixes=GFX89,GFX8 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX89,GFX9 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
 
 define i32 @test_min_max_ValK0_K1_u32(i32 %a) {
 ; GFX89-LABEL: test_min_max_ValK0_K1_u32:
diff --git a/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll b/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll
@@ -6,12 +6,12 @@
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=SDAG-GFX12,SDAG-GFX12-TRUE16 %s
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=SDAG-GFX12,SDAG-GFX12-FAKE16 %s
 
-; RUN: llc -mtriple=amdgcn -mcpu=fiji -global-isel < %s | FileCheck -check-prefixes=GISEL-VI %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -global-isel < %s | FileCheck -check-prefixes=GISEL-GFX9 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1101 -mattr=+real-true16 -global-isel < %s | FileCheck -check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-TRUE16 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1101 -mattr=-real-true16 -global-isel < %s | FileCheck -check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-FAKE16 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 -global-isel < %s | FileCheck -check-prefixes=GISEL-GFX12,GISEL-GFX12-TRUE16 %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 -global-isel < %s | FileCheck -check-prefixes=GISEL-GFX12,GISEL-GFX12-FAKE16 %s
+; RUN: llc -mtriple=amdgcn -mcpu=fiji -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GISEL-VI %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GISEL-GFX9 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1101 -mattr=+real-true16 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-TRUE16 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1101 -mattr=-real-true16 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-FAKE16 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GISEL-GFX12,GISEL-GFX12-TRUE16 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GISEL-GFX12,GISEL-GFX12-FAKE16 %s
 
 ; <GFX9 has no V_SAT_PK, GFX9+ has V_SAT_PK, GFX11 has V_SAT_PK with t16
 
diff --git a/llvm/test/CodeGen/AMDGPU/vector-reduce-smax.ll b/llvm/test/CodeGen/AMDGPU/vector-reduce-smax.ll
diff --git a/llvm/test/CodeGen/AMDGPU/vector-reduce-smin.ll b/llvm/test/CodeGen/AMDGPU/vector-reduce-smin.ll
diff --git a/llvm/test/CodeGen/AMDGPU/vector-reduce-umax.ll b/llvm/test/CodeGen/AMDGPU/vector-reduce-umax.ll
diff --git a/llvm/test/CodeGen/AMDGPU/vector-reduce-umin.ll b/llvm/test/CodeGen/AMDGPU/vector-reduce-umin.ll
