[llvm] [AMDGPU] Add regbankselect rules for G_ADD/SUB and variants (PR #159860)
Anshil Gandhi via llvm-commits
llvm-commits at lists.llvm.org
Mon Oct 20 09:37:45 PDT 2025
https://github.com/gandhi56 updated https://github.com/llvm/llvm-project/pull/159860
>From c3b719ebf8190e12189abd87b0778d50e99b8c08 Mon Sep 17 00:00:00 2001
From: Anshil Gandhi <Anshil.Gandhi at amd.com>
Date: Fri, 19 Sep 2025 17:03:33 -0500
Subject: [PATCH 1/5] [AMDGPU] Add regbankselect rules for G_ADD/SUB and
variants
Introduce add/sub support for S64 and V2S16 types. Additionally,
add rules for G_UADDO, G_USUBO, G_UADDE and G_USUBE as they are
needed for S64 addition/subtraction.
---
.../Target/AMDGPU/AMDGPURegBankLegalize.cpp | 6 +-
.../AMDGPU/AMDGPURegBankLegalizeRules.cpp | 16 +-
llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll | 535 ++++++++++++++++++
.../regbankselect-add-overflow.s32.mir | 172 ++++++
.../GlobalISel/regbankselect-add.s16.mir | 4 +-
.../GlobalISel/regbankselect-add.s64.mir | 76 +++
.../GlobalISel/regbankselect-add.v2s16.mir | 16 +-
.../regbankselect-sub-overflow.s32.mir | 172 ++++++
.../GlobalISel/regbankselect-sub.s16.mir | 104 ++++
...lect-sub.mir => regbankselect-sub.s32.mir} | 4 +-
.../GlobalISel/regbankselect-sub.s64.mir | 77 +++
.../GlobalISel/regbankselect-sub.v2s16.mir | 82 +++
llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll | 533 +++++++++++++++++
13 files changed, 1777 insertions(+), 20 deletions(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add-overflow.s32.mir
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.s64.mir
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub-overflow.s32.mir
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s16.mir
rename llvm/test/CodeGen/AMDGPU/GlobalISel/{regbankselect-sub.mir => regbankselect-sub.s32.mir} (86%)
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s64.mir
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir
create mode 100644 llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalize.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalize.cpp
index e1879598f098a..d7932448846fa 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalize.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalize.cpp
@@ -482,9 +482,9 @@ bool AMDGPURegBankLegalize::runOnMachineFunction(MachineFunction &MF) {
}
}
- assert(!getAnySgprS1(MRI).isValid() &&
- "Registers with sgpr reg bank and S1 LLT are not legal after "
- "AMDGPURegBankLegalize. Should lower to sgpr S32");
+ // assert(!getAnySgprS1(MRI).isValid() &&
+ // "Registers with sgpr reg bank and S1 LLT are not legal after "
+ // "AMDGPURegBankLegalize. Should lower to sgpr S32");
return true;
}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
index a67b12a22589c..7fe8ab30c0648 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
@@ -470,7 +470,21 @@ RegBankLegalizeRules::RegBankLegalizeRules(const GCNSubtarget &_ST,
.Uni(S16, {{Sgpr32Trunc}, {Sgpr32AExt, Sgpr32AExt}})
.Div(S16, {{Vgpr16}, {Vgpr16, Vgpr16}})
.Uni(S32, {{Sgpr32}, {Sgpr32, Sgpr32}})
- .Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32}});
+ .Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32}})
+ /// TODO: SALU does not support packed math addition. Scalarize into two S16 additions.
+ .Uni(V2S16, {{SgprV2S16}, {Sgpr32AExt, Sgpr32AExt}})
+ .Div(V2S16, {{VgprV2S16}, {VgprV2S16, VgprV2S16}})
+ .Uni(S64, {{Sgpr64}, {Sgpr64, Sgpr64}})
+ .Div(S64, {{Vgpr64}, {Vgpr64, Vgpr64}});
+
+ /// TODO: Correct these rules, related to overflow detection.
+ addRulesForGOpcs({G_UADDO, G_USUBO}, Standard)
+ .Uni(S32, {{Sgpr32, Sgpr32Trunc}, {Sgpr32, Sgpr32}})
+ .Div(S32, {{Vgpr32, Vcc}, {Vgpr32, Vgpr32}});
+
+ addRulesForGOpcs({G_UADDE, G_USUBE}, Standard)
+ .Uni(S32, {{Sgpr32, Sgpr32Trunc}, {Sgpr32, Sgpr32, Sgpr32AExtBoolInReg}})
+ .Div(S32, {{Vgpr32, Vcc}, {Vgpr32, Vgpr32, Vcc}});
addRulesForGOpcs({G_MUL}, Standard).Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32}});
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll
new file mode 100644
index 0000000000000..ba401cac2621f
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll
@@ -0,0 +1,535 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=hawaii < %s | FileCheck -check-prefix=GFX7 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=fiji < %s | FileCheck -check-prefix=GFX8 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX12 %s
+
+define i16 @s_add_i16(i16 inreg %a, i16 inreg %b) {
+; GFX7-LABEL: s_add_i16:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_add_i32 s16, s16, s17
+; GFX7-NEXT: v_mov_b32_e32 v0, s16
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_add_i16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_add_i32 s16, s16, s17
+; GFX9-NEXT: v_mov_b32_e32 v0, s16
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: s_add_i16:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: s_add_i32 s16, s16, s17
+; GFX8-NEXT: v_mov_b32_e32 v0, s16
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: s_add_i16:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: s_add_i32 s16, s16, s17
+; GFX10-NEXT: v_mov_b32_e32 v0, s16
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: s_add_i16:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_add_i32 s0, s0, s1
+; GFX11-NEXT: v_mov_b32_e32 v0, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: s_add_i16:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_add_co_i32 s0, s0, s1
+; GFX12-NEXT: s_wait_alu 0xfffe
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = add i16 %a, %b
+ ret i16 %c
+}
+
+define i16 @v_add_i16(i16 %a, i16 %b) {
+; GFX7-LABEL: v_add_i16:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_add_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_add_i16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_u16_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_add_i16:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_u16_e32 v0, v0, v1
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_add_i16:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_nc_u16 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_add_i16:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_nc_u16 v0.l, v0.l, v1.l
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_add_i16:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_nc_u16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = add i16 %a, %b
+ ret i16 %c
+}
+
+define i32 @s_add_i32(i32 inreg %a, i32 inreg %b) {
+; GFX7-LABEL: s_add_i32:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_add_i32 s16, s16, s17
+; GFX7-NEXT: v_mov_b32_e32 v0, s16
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_add_i32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_add_i32 s16, s16, s17
+; GFX9-NEXT: v_mov_b32_e32 v0, s16
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: s_add_i32:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: s_add_i32 s16, s16, s17
+; GFX8-NEXT: v_mov_b32_e32 v0, s16
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: s_add_i32:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: s_add_i32 s16, s16, s17
+; GFX10-NEXT: v_mov_b32_e32 v0, s16
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: s_add_i32:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_add_i32 s0, s0, s1
+; GFX11-NEXT: v_mov_b32_e32 v0, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: s_add_i32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_add_co_i32 s0, s0, s1
+; GFX12-NEXT: s_wait_alu 0xfffe
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = add i32 %a, %b
+ ret i32 %c
+}
+
+define i32 @v_add_i32(i32 %a, i32 %b) {
+; GFX7-LABEL: v_add_i32:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_add_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_add_i32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_u32_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_add_i32:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_u32_e32 v0, vcc, v0, v1
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_add_i32:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_nc_u32_e32 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_add_i32:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_nc_u32_e32 v0, v0, v1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_add_i32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_nc_u32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = add i32 %a, %b
+ ret i32 %c
+}
+
+; TODO: Add test for s_add_v2i16
+define <2 x i16> @v_add_v2i16(<2 x i16> %a, <2 x i16> %b) {
+; GFX7-LABEL: v_add_v2i16:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_add_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT: v_add_i32_e32 v1, vcc, v1, v3
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_add_v2i16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_add_u16 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_add_v2i16:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_u16_e32 v2, v0, v1
+; GFX8-NEXT: v_add_u16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT: v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_add_v2i16:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_pk_add_u16 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_add_v2i16:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_add_u16 v0, v0, v1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_add_v2i16:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_pk_add_u16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = add <2 x i16> %a, %b
+ ret <2 x i16> %c
+}
+
+define i64 @s_add_i64(i64 inreg %a, i64 inreg %b) {
+; GFX7-LABEL: s_add_i64:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_add_u32 s4, s16, s18
+; GFX7-NEXT: s_addc_u32 s5, s17, s19
+; GFX7-NEXT: v_mov_b32_e32 v0, s4
+; GFX7-NEXT: v_mov_b32_e32 v1, s5
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_add_i64:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_add_u32 s4, s16, s18
+; GFX9-NEXT: s_addc_u32 s5, s17, s19
+; GFX9-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-NEXT: v_mov_b32_e32 v1, s5
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: s_add_i64:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: s_add_u32 s4, s16, s18
+; GFX8-NEXT: s_addc_u32 s5, s17, s19
+; GFX8-NEXT: v_mov_b32_e32 v0, s4
+; GFX8-NEXT: v_mov_b32_e32 v1, s5
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: s_add_i64:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: s_add_u32 s4, s16, s18
+; GFX10-NEXT: s_addc_u32 s5, s17, s19
+; GFX10-NEXT: v_mov_b32_e32 v0, s4
+; GFX10-NEXT: v_mov_b32_e32 v1, s5
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: s_add_i64:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_add_u32 s0, s0, s2
+; GFX11-NEXT: s_addc_u32 s1, s1, s3
+; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: s_add_i64:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_add_nc_u64 s[0:1], s[0:1], s[2:3]
+; GFX12-NEXT: s_wait_alu 0xfffe
+; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = add i64 %a, %b
+ ret i64 %c
+}
+
+define i64 @v_add_i64(i64 %a, i64 %b) {
+; GFX7-LABEL: v_add_i64:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_add_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_add_i64:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2
+; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_add_i64:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_u32_e32 v0, vcc, v0, v2
+; GFX8-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_add_i64:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
+; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_add_i64:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
+; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_add_i64:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
+; GFX12-NEXT: s_wait_alu 0xfffd
+; GFX12-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = add i64 %a, %b
+ ret i64 %c
+}
+
+define void @s_uaddo_uadde(i64 inreg %a, i64 inreg %b, ptr addrspace(1) %res, ptr addrspace(1) %carry) {
+; GFX7-LABEL: s_uaddo_uadde:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_add_u32 s4, s16, s18
+; GFX7-NEXT: s_addc_u32 s5, s17, s19
+; GFX7-NEXT: v_mov_b32_e32 v4, s4
+; GFX7-NEXT: s_mov_b32 s6, 0
+; GFX7-NEXT: s_cselect_b32 s8, 1, 0
+; GFX7-NEXT: v_mov_b32_e32 v5, s5
+; GFX7-NEXT: s_mov_b32 s7, 0xf000
+; GFX7-NEXT: s_mov_b64 s[4:5], 0
+; GFX7-NEXT: buffer_store_dwordx2 v[4:5], v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT: v_mov_b32_e32 v0, s8
+; GFX7-NEXT: buffer_store_dword v0, v[2:3], s[4:7], 0 addr64
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_uaddo_uadde:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_add_u32 s4, s16, s18
+; GFX9-NEXT: s_addc_u32 s5, s17, s19
+; GFX9-NEXT: v_mov_b32_e32 v4, s4
+; GFX9-NEXT: s_cselect_b32 s6, 1, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, s5
+; GFX9-NEXT: global_store_dwordx2 v[0:1], v[4:5], off
+; GFX9-NEXT: v_mov_b32_e32 v0, s6
+; GFX9-NEXT: global_store_dword v[2:3], v0, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: s_uaddo_uadde:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: s_add_u32 s4, s16, s18
+; GFX8-NEXT: s_addc_u32 s5, s17, s19
+; GFX8-NEXT: v_mov_b32_e32 v4, s4
+; GFX8-NEXT: s_cselect_b32 s6, 1, 0
+; GFX8-NEXT: v_mov_b32_e32 v5, s5
+; GFX8-NEXT: flat_store_dwordx2 v[0:1], v[4:5]
+; GFX8-NEXT: v_mov_b32_e32 v0, s6
+; GFX8-NEXT: flat_store_dword v[2:3], v0
+; GFX8-NEXT: s_waitcnt vmcnt(0)
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: s_uaddo_uadde:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: s_add_u32 s4, s16, s18
+; GFX10-NEXT: s_addc_u32 s5, s17, s19
+; GFX10-NEXT: s_cselect_b32 s6, 1, 0
+; GFX10-NEXT: v_mov_b32_e32 v4, s4
+; GFX10-NEXT: v_mov_b32_e32 v5, s5
+; GFX10-NEXT: v_mov_b32_e32 v6, s6
+; GFX10-NEXT: global_store_dwordx2 v[0:1], v[4:5], off
+; GFX10-NEXT: global_store_dword v[2:3], v6, off
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: s_uaddo_uadde:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_add_u32 s0, s0, s2
+; GFX11-NEXT: s_addc_u32 s1, s1, s3
+; GFX11-NEXT: s_cselect_b32 s2, 1, 0
+; GFX11-NEXT: v_dual_mov_b32 v5, s1 :: v_dual_mov_b32 v4, s0
+; GFX11-NEXT: v_mov_b32_e32 v6, s2
+; GFX11-NEXT: global_store_b64 v[0:1], v[4:5], off
+; GFX11-NEXT: global_store_b32 v[2:3], v6, off
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: s_uaddo_uadde:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_add_co_u32 s0, s0, s2
+; GFX12-NEXT: s_add_co_ci_u32 s1, s1, s3
+; GFX12-NEXT: s_cselect_b32 s2, 1, 0
+; GFX12-NEXT: s_wait_alu 0xfffe
+; GFX12-NEXT: v_dual_mov_b32 v5, s1 :: v_dual_mov_b32 v4, s0
+; GFX12-NEXT: v_mov_b32_e32 v6, s2
+; GFX12-NEXT: global_store_b64 v[0:1], v[4:5], off
+; GFX12-NEXT: global_store_b32 v[2:3], v6, off
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %uaddo = call {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
+ %add = extractvalue {i64, i1} %uaddo, 0
+ %of = extractvalue {i64, i1} %uaddo, 1
+ %of32 = select i1 %of, i32 1, i32 0
+ store i64 %add, ptr addrspace(1) %res
+ store i32 %of32, ptr addrspace(1) %carry
+ ret void
+}
+
+define void @v_uaddo_uadde(i64 %a, i64 %b, ptr addrspace(1) %res, ptr addrspace(1) %carry) {
+; GFX7-LABEL: v_uaddo_uadde:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_add_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX7-NEXT: s_mov_b32 s6, 0
+; GFX7-NEXT: s_mov_b32 s7, 0xf000
+; GFX7-NEXT: s_mov_b64 s[4:5], 0
+; GFX7-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
+; GFX7-NEXT: buffer_store_dwordx2 v[0:1], v[4:5], s[4:7], 0 addr64
+; GFX7-NEXT: buffer_store_dword v2, v[6:7], s[4:7], 0 addr64
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_uaddo_uadde:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2
+; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
+; GFX9-NEXT: global_store_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT: global_store_dword v[6:7], v2, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_uaddo_uadde:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_u32_e32 v0, vcc, v0, v2
+; GFX8-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
+; GFX8-NEXT: flat_store_dwordx2 v[4:5], v[0:1]
+; GFX8-NEXT: flat_store_dword v[6:7], v2
+; GFX8-NEXT: s_waitcnt vmcnt(0)
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_uaddo_uadde:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
+; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX10-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo
+; GFX10-NEXT: global_store_dwordx2 v[4:5], v[0:1], off
+; GFX10-NEXT: global_store_dword v[6:7], v2, off
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_uaddo_uadde:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
+; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX11-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo
+; GFX11-NEXT: global_store_b64 v[4:5], v[0:1], off
+; GFX11-NEXT: global_store_b32 v[6:7], v2, off
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_uaddo_uadde:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2
+; GFX12-NEXT: s_wait_alu 0xfffd
+; GFX12-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX12-NEXT: s_wait_alu 0xfffd
+; GFX12-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo
+; GFX12-NEXT: global_store_b64 v[4:5], v[0:1], off
+; GFX12-NEXT: global_store_b32 v[6:7], v2, off
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %uaddo = call {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
+ %add = extractvalue {i64, i1} %uaddo, 0
+ %of = extractvalue {i64, i1} %uaddo, 1
+ %of32 = select i1 %of, i32 1, i32 0
+ store i64 %add, ptr addrspace(1) %res
+ store i32 %of32, ptr addrspace(1) %carry
+ ret void
+}
+
+declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add-overflow.s32.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add-overflow.s32.mir
new file mode 100644
index 0000000000000..1bc7d37033555
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add-overflow.s32.mir
@@ -0,0 +1,172 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=amdgcn -mcpu=fiji -run-pass="amdgpu-regbankselect,amdgpu-regbanklegalize" %s -o - | FileCheck %s
+
+---
+name: uaddo_s32_ss
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $sgpr1
+ ; CHECK-LABEL: name: uaddo_s32_ss
+ ; CHECK: liveins: $sgpr0, $sgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+ ; CHECK-NEXT: [[UADDO:%[0-9]+]]:sgpr(s32), [[UADDO1:%[0-9]+]]:sgpr(s32) = G_UADDO [[COPY]], [[COPY1]]
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[UADDO1]](s32)
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $sgpr1
+ %2:_(s32), %3:_(s1) = G_UADDO %0, %1
+...
+
+---
+name: uaddo_s32_sv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $vgpr1
+ ; CHECK-LABEL: name: uaddo_s32_sv
+ ; CHECK: liveins: $sgpr0, $vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+ ; CHECK-NEXT: [[UADDO:%[0-9]+]]:vgpr(s32), [[UADDO1:%[0-9]+]]:vcc(s1) = G_UADDO [[COPY2]], [[COPY1]]
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32), %3:_(s1) = G_UADDO %0, %1
+...
+
+---
+name: uaddo_s32_vs
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $sgpr1
+ ; CHECK-LABEL: name: uaddo_s32_vs
+ ; CHECK: liveins: $vgpr0, $sgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+ ; CHECK-NEXT: [[UADDO:%[0-9]+]]:vgpr(s32), [[UADDO1:%[0-9]+]]:vcc(s1) = G_UADDO [[COPY]], [[COPY2]]
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $sgpr1
+ %2:_(s32), %3:_(s1) = G_UADDO %0, %1
+...
+
+---
+name: uaddo_s32_vv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; CHECK-LABEL: name: uaddo_s32_vv
+ ; CHECK: liveins: $vgpr0, $vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; CHECK-NEXT: [[UADDO:%[0-9]+]]:vgpr(s32), [[UADDO1:%[0-9]+]]:vcc(s1) = G_UADDO [[COPY]], [[COPY1]]
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32), %3:_(s1) = G_UADDO %0, %1
+...
+
+---
+name: uadde_s32_ss
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $sgpr1, $sgpr2
+ ; CHECK-LABEL: name: uadde_s32_ss
+ ; CHECK: liveins: $sgpr0, $sgpr1, $sgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+ ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[COPY2]], [[C]]
+ ; CHECK-NEXT: [[UADDE:%[0-9]+]]:sgpr(s32), [[UADDE1:%[0-9]+]]:sgpr(s32) = G_UADDE [[COPY]], [[COPY1]], [[AND]]
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[UADDE1]](s32)
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $sgpr1
+ %2:_(s32) = COPY $sgpr2
+ %3:_(s1) = G_TRUNC %2
+ %4:_(s32), %5:_(s1) = G_UADDE %0, %1, %3
+...
+
+---
+name: uadde_s32_sv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $vgpr1, $sgpr2
+ ; CHECK-LABEL: name: uadde_s32_sv
+ ; CHECK: liveins: $sgpr0, $vgpr1, $sgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+ ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+ ; CHECK-NEXT: [[AMDGPU_COPY_VCC_SCC:%[0-9]+]]:vcc(s1) = G_AMDGPU_COPY_VCC_SCC [[COPY2]](s32)
+ ; CHECK-NEXT: [[UADDE:%[0-9]+]]:vgpr(s32), [[UADDE1:%[0-9]+]]:vcc(s1) = G_UADDE [[COPY3]], [[COPY1]], [[AMDGPU_COPY_VCC_SCC]]
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = COPY $sgpr2
+ %3:_(s1) = G_TRUNC %2
+ %4:_(s32), %5:_(s1) = G_UADDE %0, %1, %3
+...
+
+---
+name: uadde_s32_vs
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $sgpr1, $sgpr2
+ ; CHECK-LABEL: name: uadde_s32_vs
+ ; CHECK: liveins: $vgpr0, $sgpr1, $sgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+ ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+ ; CHECK-NEXT: [[AMDGPU_COPY_VCC_SCC:%[0-9]+]]:vcc(s1) = G_AMDGPU_COPY_VCC_SCC [[COPY2]](s32)
+ ; CHECK-NEXT: [[UADDE:%[0-9]+]]:vgpr(s32), [[UADDE1:%[0-9]+]]:vcc(s1) = G_UADDE [[COPY]], [[COPY3]], [[AMDGPU_COPY_VCC_SCC]]
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $sgpr1
+ %2:_(s32) = COPY $sgpr2
+ %3:_(s1) = G_TRUNC %2
+ %4:_(s32), %5:_(s1) = G_UADDE %0, %1, %3
+...
+
+---
+name: uadde_s32_vv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-LABEL: name: uadde_s32_vv
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+ ; CHECK-NEXT: [[C:%[0-9]+]]:vgpr(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[COPY2]], [[C]]
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:vgpr(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(ne), [[AND]](s32), [[C1]]
+ ; CHECK-NEXT: [[UADDE:%[0-9]+]]:vgpr(s32), [[UADDE1:%[0-9]+]]:vcc(s1) = G_UADDE [[COPY]], [[COPY1]], [[ICMP]]
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = COPY $vgpr2
+ %3:_(s1) = G_TRUNC %2
+ %4:_(s32), %5:_(s1) = G_UADDE %0, %1, %3
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.s16.mir
index 54ee69fcb2204..c9785ded1b827 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.s16.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.s16.mir
@@ -1,6 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
-# RUN: llc -mtriple=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=fiji -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=fiji -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
---
name: add_s16_ss
legalized: true
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.s64.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.s64.mir
new file mode 100644
index 0000000000000..e3230a2420451
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.s64.mir
@@ -0,0 +1,76 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -o - | FileCheck %s
+
+---
+name: add_s64_ss
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0_sgpr1, $sgpr2_sgpr3
+ ; CHECK-LABEL: name: add_s64_ss
+ ; CHECK: liveins: $sgpr0_sgpr1, $sgpr2_sgpr3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s64) = COPY $sgpr2_sgpr3
+ ; CHECK-NEXT: [[ADD:%[0-9]+]]:sgpr(s64) = G_ADD [[COPY]], [[COPY1]]
+ %0:_(s64) = COPY $sgpr0_sgpr1
+ %1:_(s64) = COPY $sgpr2_sgpr3
+ %2:_(s64) = G_ADD %0, %1
+...
+
+---
+name: add_s64_sv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0_sgpr1, $vgpr0_vgpr1
+ ; CHECK-LABEL: name: add_s64_sv
+ ; CHECK: liveins: $sgpr0_sgpr1, $vgpr0_vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s64) = COPY [[COPY]](s64)
+ ; CHECK-NEXT: [[ADD:%[0-9]+]]:vgpr(s64) = G_ADD [[COPY2]], [[COPY1]]
+ %0:_(s64) = COPY $sgpr0_sgpr1
+ %1:_(s64) = COPY $vgpr0_vgpr1
+ %2:_(s64) = G_ADD %0, %1
+...
+
+---
+name: add_s64_vs
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0_sgpr1, $vgpr0_vgpr1
+ ; CHECK-LABEL: name: add_s64_vs
+ ; CHECK: liveins: $sgpr0_sgpr1, $vgpr0_vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s64) = COPY [[COPY1]](s64)
+ ; CHECK-NEXT: [[ADD:%[0-9]+]]:vgpr(s64) = G_ADD [[COPY]], [[COPY2]]
+ %0:_(s64) = COPY $vgpr0_vgpr1
+ %1:_(s64) = COPY $sgpr0_sgpr1
+ %2:_(s64) = G_ADD %0, %1
+...
+
+---
+name: add_s64_vv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+ ; CHECK-LABEL: name: add_s64_vv
+ ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
+ ; CHECK-NEXT: [[ADD:%[0-9]+]]:vgpr(s64) = G_ADD [[COPY]], [[COPY1]]
+ %0:_(s64) = COPY $vgpr0_vgpr1
+ %1:_(s64) = COPY $vgpr2_vgpr3
+ %2:_(s64) = G_ADD %0, %1
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
index 97018fac13a87..6fdad715f1b14 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
@@ -1,6 +1,6 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
---
name: add_v2s16_ss
@@ -14,16 +14,8 @@ body: |
; CHECK-NEXT: {{ $}}
; 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: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>)
- ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
- ; CHECK-NEXT: [[LSHR1:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST1]], [[C1]](s32)
- ; CHECK-NEXT: [[ADD:%[0-9]+]]:sgpr(s32) = G_ADD [[BITCAST]], [[BITCAST1]]
- ; CHECK-NEXT: [[ADD1:%[0-9]+]]:sgpr(s32) = G_ADD [[LSHR]], [[LSHR1]]
- ; CHECK-NEXT: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ADD]](s32), [[ADD1]](s32)
- ; CHECK-NEXT: S_ENDPGM 0, implicit [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+ ; CHECK-NEXT: [[ADD:%[0-9]+]]:sgpr(<2 x s16>) = G_ADD [[COPY]], [[COPY1]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[ADD]](<2 x s16>)
%0:_(<2 x s16>) = COPY $sgpr0
%1:_(<2 x s16>) = COPY $sgpr1
%2:_(<2 x s16>) = G_ADD %0, %1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub-overflow.s32.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub-overflow.s32.mir
new file mode 100644
index 0000000000000..f9058047d065b
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub-overflow.s32.mir
@@ -0,0 +1,172 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn -mcpu=fiji -run-pass="amdgpu-regbankselect,amdgpu-regbanklegalize" %s -o - | FileCheck %s
+
+---
+name: usubo_s32_ss
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $sgpr1
+ ; CHECK-LABEL: name: usubo_s32_ss
+ ; CHECK: liveins: $sgpr0, $sgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+ ; CHECK-NEXT: [[USUBO:%[0-9]+]]:sgpr(s32), [[USUBO1:%[0-9]+]]:sgpr(s32) = G_USUBO [[COPY]], [[COPY1]]
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[USUBO1]](s32)
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $sgpr1
+ %2:_(s32), %3:_(s1) = G_USUBO %0, %1
+...
+
+---
+name: usubo_s32_sv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $vgpr1
+ ; CHECK-LABEL: name: usubo_s32_sv
+ ; CHECK: liveins: $sgpr0, $vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+ ; CHECK-NEXT: [[USUBO:%[0-9]+]]:vgpr(s32), [[USUBO1:%[0-9]+]]:vcc(s1) = G_USUBO [[COPY2]], [[COPY1]]
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32), %3:_(s1) = G_USUBO %0, %1
+...
+
+---
+name: usubo_s32_vs
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $sgpr1
+ ; CHECK-LABEL: name: usubo_s32_vs
+ ; CHECK: liveins: $vgpr0, $sgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+ ; CHECK-NEXT: [[USUBO:%[0-9]+]]:vgpr(s32), [[USUBO1:%[0-9]+]]:vcc(s1) = G_USUBO [[COPY]], [[COPY2]]
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $sgpr1
+ %2:_(s32), %3:_(s1) = G_USUBO %0, %1
+...
+
+---
+name: usubo_s32_vv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; CHECK-LABEL: name: usubo_s32_vv
+ ; CHECK: liveins: $vgpr0, $vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; CHECK-NEXT: [[USUBO:%[0-9]+]]:vgpr(s32), [[USUBO1:%[0-9]+]]:vcc(s1) = G_USUBO [[COPY]], [[COPY1]]
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32), %3:_(s1) = G_USUBO %0, %1
+...
+
+---
+name: usube_s32_ss
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $sgpr1, $sgpr2
+ ; CHECK-LABEL: name: usube_s32_ss
+ ; CHECK: liveins: $sgpr0, $sgpr1, $sgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+ ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[COPY2]], [[C]]
+ ; CHECK-NEXT: [[USUBE:%[0-9]+]]:sgpr(s32), [[USUBE1:%[0-9]+]]:sgpr(s32) = G_USUBE [[COPY]], [[COPY1]], [[AND]]
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[USUBE1]](s32)
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $sgpr1
+ %2:_(s32) = COPY $sgpr2
+ %3:_(s1) = G_TRUNC %2
+ %4:_(s32), %5:_(s1) = G_USUBE %0, %1, %3
+...
+
+---
+name: usube_s32_sv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $vgpr1, $sgpr2
+ ; CHECK-LABEL: name: usube_s32_sv
+ ; CHECK: liveins: $sgpr0, $vgpr1, $sgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+ ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+ ; CHECK-NEXT: [[AMDGPU_COPY_VCC_SCC:%[0-9]+]]:vcc(s1) = G_AMDGPU_COPY_VCC_SCC [[COPY2]](s32)
+ ; CHECK-NEXT: [[USUBE:%[0-9]+]]:vgpr(s32), [[USUBE1:%[0-9]+]]:vcc(s1) = G_USUBE [[COPY3]], [[COPY1]], [[AMDGPU_COPY_VCC_SCC]]
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = COPY $sgpr2
+ %3:_(s1) = G_TRUNC %2
+ %4:_(s32), %5:_(s1) = G_USUBE %0, %1, %3
+...
+
+---
+name: usube_s32_vs
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $sgpr1, $sgpr2
+ ; CHECK-LABEL: name: usube_s32_vs
+ ; CHECK: liveins: $vgpr0, $sgpr1, $sgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+ ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+ ; CHECK-NEXT: [[AMDGPU_COPY_VCC_SCC:%[0-9]+]]:vcc(s1) = G_AMDGPU_COPY_VCC_SCC [[COPY2]](s32)
+ ; CHECK-NEXT: [[USUBE:%[0-9]+]]:vgpr(s32), [[USUBE1:%[0-9]+]]:vcc(s1) = G_USUBE [[COPY]], [[COPY3]], [[AMDGPU_COPY_VCC_SCC]]
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $sgpr1
+ %2:_(s32) = COPY $sgpr2
+ %3:_(s1) = G_TRUNC %2
+ %4:_(s32), %5:_(s1) = G_USUBE %0, %1, %3
+...
+
+---
+name: usube_s32_vv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-LABEL: name: usube_s32_vv
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+ ; CHECK-NEXT: [[C:%[0-9]+]]:vgpr(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[COPY2]], [[C]]
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:vgpr(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(ne), [[AND]](s32), [[C1]]
+ ; CHECK-NEXT: [[USUBE:%[0-9]+]]:vgpr(s32), [[USUBE1:%[0-9]+]]:vcc(s1) = G_USUBE [[COPY]], [[COPY1]], [[ICMP]]
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s32) = COPY $vgpr2
+ %3:_(s1) = G_TRUNC %2
+ %4:_(s32), %5:_(s1) = G_USUBE %0, %1, %3
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s16.mir
new file mode 100644
index 0000000000000..61be6b0bf3081
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s16.mir
@@ -0,0 +1,104 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
+
+---
+name: sub_s16_ss
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $sgpr1
+ ; CHECK-LABEL: name: sub_s16_ss
+ ; CHECK: liveins: $sgpr0, $sgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s16) = G_TRUNC [[COPY]](s32)
+ ; CHECK-NEXT: [[TRUNC1:%[0-9]+]]:sgpr(s16) = G_TRUNC [[COPY1]](s32)
+ ; CHECK-NEXT: [[ANYEXT:%[0-9]+]]:sgpr(s32) = G_ANYEXT [[TRUNC]](s16)
+ ; CHECK-NEXT: [[ANYEXT1:%[0-9]+]]:sgpr(s32) = G_ANYEXT [[TRUNC1]](s16)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:sgpr(s32) = G_SUB [[ANYEXT]], [[ANYEXT1]]
+ ; CHECK-NEXT: [[TRUNC2:%[0-9]+]]:sgpr(s16) = G_TRUNC [[SUB]](s32)
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[TRUNC2]](s16)
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $sgpr1
+ %2:_(s16) = G_TRUNC %0
+ %3:_(s16) = G_TRUNC %1
+ %4:_(s16) = G_SUB %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: sub_s16_sv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $vgpr0
+ ; CHECK-LABEL: name: sub_s16_sv
+ ; CHECK: liveins: $sgpr0, $vgpr0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s16) = G_TRUNC [[COPY]](s32)
+ ; CHECK-NEXT: [[TRUNC1:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY1]](s32)
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s16) = COPY [[TRUNC]](s16)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:vgpr(s16) = G_SUB [[COPY2]], [[TRUNC1]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[SUB]](s16)
+ %0:_(s32) = COPY $sgpr0
+ %1:_(s32) = COPY $vgpr0
+ %2:_(s16) = G_TRUNC %0
+ %3:_(s16) = G_TRUNC %1
+ %4:_(s16) = G_SUB %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: sub_s16_vs
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $vgpr0
+ ; CHECK-LABEL: name: sub_s16_vs
+ ; CHECK: liveins: $sgpr0, $vgpr0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY]](s32)
+ ; CHECK-NEXT: [[TRUNC1:%[0-9]+]]:sgpr(s16) = G_TRUNC [[COPY1]](s32)
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s16) = COPY [[TRUNC1]](s16)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:vgpr(s16) = G_SUB [[TRUNC]], [[COPY2]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[SUB]](s16)
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $sgpr0
+ %2:_(s16) = G_TRUNC %0
+ %3:_(s16) = G_TRUNC %1
+ %4:_(s16) = G_SUB %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: sub_s16_vv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; CHECK-LABEL: name: sub_s16_vv
+ ; CHECK: liveins: $vgpr0, $vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY]](s32)
+ ; CHECK-NEXT: [[TRUNC1:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY1]](s32)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:vgpr(s16) = G_SUB [[TRUNC]], [[TRUNC1]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[SUB]](s16)
+ %0:_(s32) = COPY $vgpr0
+ %1:_(s32) = COPY $vgpr1
+ %2:_(s16) = G_TRUNC %0
+ %3:_(s16) = G_TRUNC %1
+ %4:_(s16) = G_SUB %2, %3
+ S_ENDPGM 0, implicit %4
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s32.mir
similarity index 86%
rename from llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.mir
rename to llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s32.mir
index b0199d3ad5cd1..4024e992d2360 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s32.mir
@@ -1,5 +1,5 @@
-# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn -mcpu=fiji -run-pass="amdgpu-regbankselect,amdgpu-regbanklegalize" %s -o - | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
---
name: sub_s32_ss
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s64.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s64.mir
new file mode 100644
index 0000000000000..84c288c7e88dd
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.s64.mir
@@ -0,0 +1,77 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
+
+---
+name: sub_s64_ss
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0_sgpr1, $sgpr0_sgpr1
+ ; CHECK-LABEL: name: sub_s64_ss
+ ; CHECK: liveins: $sgpr0_sgpr1, $sgpr0_sgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:sgpr(s64) = G_SUB [[COPY]], [[COPY1]]
+ %0:_(s64) = COPY $sgpr0_sgpr1
+ %1:_(s64) = COPY $sgpr0_sgpr1
+ %2:_(s64) = G_SUB %0, %1
+...
+
+---
+name: sub_s64_sv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0_sgpr1, $vgpr0_vgpr1
+ ; CHECK-LABEL: name: sub_s64_sv
+ ; CHECK: liveins: $sgpr0_sgpr1, $vgpr0_vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s64) = COPY [[COPY]](s64)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:vgpr(s64) = G_SUB [[COPY2]], [[COPY1]]
+ %0:_(s64) = COPY $sgpr0_sgpr1
+ %1:_(s64) = COPY $vgpr0_vgpr1
+ %2:_(s64) = G_SUB %0, %1
+...
+
+---
+name: sub_s64_vs
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0_sgpr1, $vgpr0_vgpr1
+ ; CHECK-LABEL: name: sub_s64_vs
+ ; CHECK: liveins: $sgpr0_sgpr1, $vgpr0_vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s64) = COPY [[COPY1]](s64)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:vgpr(s64) = G_SUB [[COPY]], [[COPY2]]
+ %0:_(s64) = COPY $vgpr0_vgpr1
+ %1:_(s64) = COPY $sgpr0_sgpr1
+ %2:_(s64) = G_SUB %0, %1
+...
+
+---
+name: sub_s64_vv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+ ; CHECK-LABEL: name: sub_s64_vv
+ ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:vgpr(s64) = G_SUB [[COPY]], [[COPY1]]
+ %0:_(s64) = COPY $vgpr0_vgpr1
+ %1:_(s64) = COPY $vgpr2_vgpr3
+ %2:_(s64) = G_SUB %0, %1
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir
new file mode 100644
index 0000000000000..c881ff3115e87
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir
@@ -0,0 +1,82 @@
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
+
+---
+name: sub_v2s16_ss
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $sgpr1
+ ; CHECK-LABEL: name: sub_v2s16_ss
+ ; CHECK: liveins: $sgpr0, $sgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr1
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:sgpr(<2 x s16>) = G_SUB [[COPY]], [[COPY1]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[SUB]](<2 x s16>)
+ %0:_(<2 x s16>) = COPY $sgpr0
+ %1:_(<2 x s16>) = COPY $sgpr1
+ %2:_(<2 x s16>) = G_SUB %0, %1
+ S_ENDPGM 0, implicit %2
+...
+
+---
+name: sub_v2s16_sv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $vgpr0
+ ; CHECK-LABEL: name: sub_v2s16_sv
+ ; CHECK: liveins: $sgpr0, $vgpr0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY]](<2 x s16>)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:vgpr(<2 x s16>) = G_SUB [[COPY2]], [[COPY1]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[SUB]](<2 x s16>)
+ %0:_(<2 x s16>) = COPY $sgpr0
+ %1:_(<2 x s16>) = COPY $vgpr0
+ %2:_(<2 x s16>) = G_SUB %0, %1
+ S_ENDPGM 0, implicit %2
+...
+
+---
+name: sub_v2s16_vs
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $sgpr0, $vgpr0
+ ; CHECK-LABEL: name: sub_v2s16_vs
+ ; CHECK: liveins: $sgpr0, $vgpr0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY1]](<2 x s16>)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:vgpr(<2 x s16>) = G_SUB [[COPY]], [[COPY2]]
+ %0:_(<2 x s16>) = COPY $vgpr0
+ %1:_(<2 x s16>) = COPY $sgpr0
+ %2:_(<2 x s16>) = G_SUB %0, %1
+...
+
+---
+name: sub_v2s16_vv
+legalized: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; CHECK-LABEL: name: sub_v2s16_vv
+ ; CHECK: liveins: $vgpr0, $vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr1
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:vgpr(<2 x s16>) = G_SUB [[COPY]], [[COPY1]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[SUB]](<2 x s16>)
+ %0:_(<2 x s16>) = COPY $vgpr0
+ %1:_(<2 x s16>) = COPY $vgpr1
+ %2:_(<2 x s16>) = G_SUB %0, %1
+ S_ENDPGM 0, implicit %2
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll
new file mode 100644
index 0000000000000..414e86518123d
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll
@@ -0,0 +1,533 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=hawaii < %s | FileCheck -check-prefix=GFX7 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=fiji < %s | FileCheck -check-prefix=GFX8 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX12 %s
+
+define i16 @s_sub_i16(i16 inreg %a, i16 inreg %b) {
+; GFX7-LABEL: s_sub_i16:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_sub_i32 s4, s16, s17
+; GFX7-NEXT: v_mov_b32_e32 v0, s4
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_sub_i16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_sub_i32 s4, s16, s17
+; GFX9-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: s_sub_i16:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: s_sub_i32 s4, s16, s17
+; GFX8-NEXT: v_mov_b32_e32 v0, s4
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: s_sub_i16:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: s_sub_i32 s4, s16, s17
+; GFX10-NEXT: v_mov_b32_e32 v0, s4
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: s_sub_i16:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_sub_i32 s0, s0, s1
+; GFX11-NEXT: v_mov_b32_e32 v0, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: s_sub_i16:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_sub_co_i32 s0, s0, s1
+; GFX12-NEXT: s_wait_alu 0xfffe
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = sub i16 %a, %b
+ ret i16 %c
+}
+
+define i16 @v_sub_i16(i16 %a, i16 %b) {
+; GFX7-LABEL: v_sub_i16:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_sub_i16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_sub_u16_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_sub_i16:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_sub_u16_e32 v0, v0, v1
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_sub_i16:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_sub_nc_u16 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_sub_i16:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_sub_nc_u16 v0.l, v0.l, v1.l
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_sub_i16:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_nc_u16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = sub i16 %a, %b
+ ret i16 %c
+}
+
+define i32 @s_sub_i32(i32 inreg %a, i32 inreg %b) {
+; GFX7-LABEL: s_sub_i32:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_sub_i32 s4, s16, s17
+; GFX7-NEXT: v_mov_b32_e32 v0, s4
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_sub_i32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_sub_i32 s4, s16, s17
+; GFX9-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: s_sub_i32:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: s_sub_i32 s4, s16, s17
+; GFX8-NEXT: v_mov_b32_e32 v0, s4
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: s_sub_i32:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: s_sub_i32 s4, s16, s17
+; GFX10-NEXT: v_mov_b32_e32 v0, s4
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: s_sub_i32:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_sub_i32 s0, s0, s1
+; GFX11-NEXT: v_mov_b32_e32 v0, s0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: s_sub_i32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_sub_co_i32 s0, s0, s1
+; GFX12-NEXT: s_wait_alu 0xfffe
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = sub i32 %a, %b
+ ret i32 %c
+}
+
+define i32 @v_sub_i32(i32 %a, i32 %b) {
+; GFX7-LABEL: v_sub_i32:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_sub_i32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_sub_u32_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_sub_i32:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_sub_u32_e32 v0, vcc, v0, v1
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_sub_i32:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_sub_nc_u32_e32 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_sub_i32:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_sub_nc_u32_e32 v0, v0, v1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_sub_i32:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_nc_u32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = sub i32 %a, %b
+ ret i32 %c
+}
+
+; TODO: sub test for s_sub_v2i16
+define <2 x i16> @v_sub_v2i16(<2 x i16> %a, <2 x i16> %b) {
+; GFX7-LABEL: v_sub_v2i16:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_sub_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT: v_sub_i32_e32 v1, vcc, v1, v3
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_sub_v2i16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_sub_i16 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_sub_v2i16:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_sub_u16_e32 v2, v0, v1
+; GFX8-NEXT: v_sub_u16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT: v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_sub_v2i16:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_pk_sub_i16 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_sub_v2i16:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_sub_i16 v0, v0, v1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_sub_v2i16:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_pk_sub_i16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = sub <2 x i16> %a, %b
+ ret <2 x i16> %c
+}
+
+define i64 @s_sub_i64(i64 inreg %a, i64 inreg %b) {
+; GFX7-LABEL: s_sub_i64:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_sub_u32 s4, s16, s18
+; GFX7-NEXT: s_subb_u32 s5, s17, s19
+; GFX7-NEXT: v_mov_b32_e32 v0, s4
+; GFX7-NEXT: v_mov_b32_e32 v1, s5
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_sub_i64:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_sub_u32 s4, s16, s18
+; GFX9-NEXT: s_subb_u32 s5, s17, s19
+; GFX9-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-NEXT: v_mov_b32_e32 v1, s5
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: s_sub_i64:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: s_sub_u32 s4, s16, s18
+; GFX8-NEXT: s_subb_u32 s5, s17, s19
+; GFX8-NEXT: v_mov_b32_e32 v0, s4
+; GFX8-NEXT: v_mov_b32_e32 v1, s5
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: s_sub_i64:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: s_sub_u32 s4, s16, s18
+; GFX10-NEXT: s_subb_u32 s5, s17, s19
+; GFX10-NEXT: v_mov_b32_e32 v0, s4
+; GFX10-NEXT: v_mov_b32_e32 v1, s5
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: s_sub_i64:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_sub_u32 s0, s0, s2
+; GFX11-NEXT: s_subb_u32 s1, s1, s3
+; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: s_sub_i64:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_sub_nc_u64 s[0:1], s[0:1], s[2:3]
+; GFX12-NEXT: s_wait_alu 0xfffe
+; GFX12-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = sub i64 %a, %b
+ ret i64 %c
+}
+
+define i64 @v_sub_i64(i64 %a, i64 %b) {
+; GFX7-LABEL: v_sub_i64:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_sub_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_sub_i64:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, v0, v2
+; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_sub_i64:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_sub_u32_e32 v0, vcc, v0, v2
+; GFX8-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_sub_i64:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_sub_co_u32 v0, vcc_lo, v0, v2
+; GFX10-NEXT: v_sub_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_sub_i64:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_sub_co_u32 v0, vcc_lo, v0, v2
+; GFX11-NEXT: v_sub_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_sub_i64:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_co_u32 v0, vcc_lo, v0, v2
+; GFX12-NEXT: s_wait_alu 0xfffd
+; GFX12-NEXT: v_sub_co_ci_u32_e64 v1, null, v1, v3, vcc_lo
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %c = sub i64 %a, %b
+ ret i64 %c
+}
+
+define void @s_usubo_usube(i64 inreg %a, i64 inreg %b, ptr addrspace(1) %res, ptr addrspace(1) %carry) {
+; GFX7-LABEL: s_usubo_usube:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_sub_u32 s4, s16, s18
+; GFX7-NEXT: s_subb_u32 s5, s17, s19
+; GFX7-NEXT: v_mov_b32_e32 v4, s4
+; GFX7-NEXT: s_mov_b32 s6, 0
+; GFX7-NEXT: s_cselect_b32 s8, 1, 0
+; GFX7-NEXT: v_mov_b32_e32 v5, s5
+; GFX7-NEXT: s_mov_b32 s7, 0xf000
+; GFX7-NEXT: s_mov_b64 s[4:5], 0
+; GFX7-NEXT: buffer_store_dwordx2 v[4:5], v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT: v_mov_b32_e32 v0, s8
+; GFX7-NEXT: buffer_store_dword v0, v[2:3], s[4:7], 0 addr64
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_usubo_usube:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_sub_u32 s4, s16, s18
+; GFX9-NEXT: s_subb_u32 s5, s17, s19
+; GFX9-NEXT: v_mov_b32_e32 v4, s4
+; GFX9-NEXT: s_cselect_b32 s6, 1, 0
+; GFX9-NEXT: v_mov_b32_e32 v5, s5
+; GFX9-NEXT: global_store_dwordx2 v[0:1], v[4:5], off
+; GFX9-NEXT: v_mov_b32_e32 v0, s6
+; GFX9-NEXT: global_store_dword v[2:3], v0, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: s_usubo_usube:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: s_sub_u32 s4, s16, s18
+; GFX8-NEXT: s_subb_u32 s5, s17, s19
+; GFX8-NEXT: v_mov_b32_e32 v4, s4
+; GFX8-NEXT: s_cselect_b32 s6, 1, 0
+; GFX8-NEXT: v_mov_b32_e32 v5, s5
+; GFX8-NEXT: flat_store_dwordx2 v[0:1], v[4:5]
+; GFX8-NEXT: v_mov_b32_e32 v0, s6
+; GFX8-NEXT: flat_store_dword v[2:3], v0
+; GFX8-NEXT: s_waitcnt vmcnt(0)
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: s_usubo_usube:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: s_sub_u32 s4, s16, s18
+; GFX10-NEXT: s_subb_u32 s5, s17, s19
+; GFX10-NEXT: s_cselect_b32 s6, 1, 0
+; GFX10-NEXT: v_mov_b32_e32 v4, s4
+; GFX10-NEXT: v_mov_b32_e32 v5, s5
+; GFX10-NEXT: v_mov_b32_e32 v6, s6
+; GFX10-NEXT: global_store_dwordx2 v[0:1], v[4:5], off
+; GFX10-NEXT: global_store_dword v[2:3], v6, off
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: s_usubo_usube:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_sub_u32 s0, s0, s2
+; GFX11-NEXT: s_subb_u32 s1, s1, s3
+; GFX11-NEXT: s_cselect_b32 s2, 1, 0
+; GFX11-NEXT: v_dual_mov_b32 v5, s1 :: v_dual_mov_b32 v4, s0
+; GFX11-NEXT: v_mov_b32_e32 v6, s2
+; GFX11-NEXT: global_store_b64 v[0:1], v[4:5], off
+; GFX11-NEXT: global_store_b32 v[2:3], v6, off
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: s_usubo_usube:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_sub_co_u32 s0, s0, s2
+; GFX12-NEXT: s_sub_co_ci_u32 s1, s1, s3
+; GFX12-NEXT: s_cselect_b32 s2, 1, 0
+; GFX12-NEXT: s_wait_alu 0xfffe
+; GFX12-NEXT: v_dual_mov_b32 v5, s1 :: v_dual_mov_b32 v4, s0
+; GFX12-NEXT: v_mov_b32_e32 v6, s2
+; GFX12-NEXT: global_store_b64 v[0:1], v[4:5], off
+; GFX12-NEXT: global_store_b32 v[2:3], v6, off
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %usubo = call {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
+ %sub = extractvalue {i64, i1} %usubo, 0
+ %of = extractvalue {i64, i1} %usubo, 1
+ %of32 = select i1 %of, i32 1, i32 0
+ store i64 %sub, ptr addrspace(1) %res
+ store i32 %of32, ptr addrspace(1) %carry
+ ret void
+}
+
+define void @v_usubo_usube(i64 %a, i64 %b, ptr addrspace(1) %res, ptr addrspace(1) %carry) {
+; GFX7-LABEL: v_usubo_usube:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_sub_i32_e32 v0, vcc, v0, v2
+; GFX7-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc
+; GFX7-NEXT: s_mov_b32 s6, 0
+; GFX7-NEXT: s_mov_b32 s7, 0xf000
+; GFX7-NEXT: s_mov_b64 s[4:5], 0
+; GFX7-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
+; GFX7-NEXT: buffer_store_dwordx2 v[0:1], v[4:5], s[4:7], 0 addr64
+; GFX7-NEXT: buffer_store_dword v2, v[6:7], s[4:7], 0 addr64
+; GFX7-NEXT: s_waitcnt vmcnt(0)
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_usubo_usube:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, v0, v2
+; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
+; GFX9-NEXT: global_store_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT: global_store_dword v[6:7], v2, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-LABEL: v_usubo_usube:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_sub_u32_e32 v0, vcc, v0, v2
+; GFX8-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
+; GFX8-NEXT: flat_store_dwordx2 v[4:5], v[0:1]
+; GFX8-NEXT: flat_store_dword v[6:7], v2
+; GFX8-NEXT: s_waitcnt vmcnt(0)
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: v_usubo_usube:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_sub_co_u32 v0, vcc_lo, v0, v2
+; GFX10-NEXT: v_sub_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX10-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo
+; GFX10-NEXT: global_store_dwordx2 v[4:5], v[0:1], off
+; GFX10-NEXT: global_store_dword v[6:7], v2, off
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_usubo_usube:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_sub_co_u32 v0, vcc_lo, v0, v2
+; GFX11-NEXT: v_sub_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX11-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo
+; GFX11-NEXT: global_store_b64 v[4:5], v[0:1], off
+; GFX11-NEXT: global_store_b32 v[6:7], v2, off
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_usubo_usube:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_co_u32 v0, vcc_lo, v0, v2
+; GFX12-NEXT: s_wait_alu 0xfffd
+; GFX12-NEXT: v_sub_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX12-NEXT: s_wait_alu 0xfffd
+; GFX12-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo
+; GFX12-NEXT: global_store_b64 v[4:5], v[0:1], off
+; GFX12-NEXT: global_store_b32 v[6:7], v2, off
+; GFX12-NEXT: s_setpc_b64 s[30:31]
+ %usubo = call {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
+ %sub = extractvalue {i64, i1} %usubo, 0
+ %of = extractvalue {i64, i1} %usubo, 1
+ %of32 = select i1 %of, i32 1, i32 0
+ store i64 %sub, ptr addrspace(1) %res
+ store i32 %of32, ptr addrspace(1) %carry
+ ret void
+}
>From 278fa031e9078e683f22e8a29c18daebeec5032f Mon Sep 17 00:00:00 2001
From: Anshil Gandhi <Anshil.Gandhi at amd.com>
Date: Fri, 17 Oct 2025 14:30:25 -0500
Subject: [PATCH 2/5] Introduce scalarization of V2S16 for G_ADD/SUB
---
.../AMDGPU/AMDGPURegBankLegalizeHelper.cpp | 16 ++++++++++++++++
.../Target/AMDGPU/AMDGPURegBankLegalizeHelper.h | 2 ++
.../Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp | 4 +---
.../Target/AMDGPU/AMDGPURegBankLegalizeRules.h | 3 ++-
llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll | 4 +++-
.../GlobalISel/regbankselect-add.v2s16.mir | 8 ++++++--
.../GlobalISel/regbankselect-sub.v2s16.mir | 9 +++++++--
llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll | 4 +++-
8 files changed, 40 insertions(+), 10 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
index 540756653dd22..16bf63954f4d3 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
@@ -500,6 +500,20 @@ void RegBankLegalizeHelper::lowerUnpackMinMax(MachineInstr &MI) {
MI.eraseFromParent();
}
+void RegBankLegalizeHelper::lowerScalarizeV2S16(MachineInstr &MI) {
+ // Unpack the V2S16 operands into two S16 scalars each
+ auto Op1 = B.buildUnmerge({SgprRB, S16}, MI.getOperand(1).getReg());
+ auto Op2 = B.buildUnmerge({SgprRB, S16}, MI.getOperand(2).getReg());
+
+ // Perform scalar additions on S16 values
+ Register Lo = B.buildInstr(MI.getOpcode(), {SgprRB_S16}, {Op1.getReg(0), Op2.getReg(0)}).getReg(0);
+ Register Hi = B.buildInstr(MI.getOpcode(), {SgprRB_S16}, {Op1.getReg(1), Op2.getReg(1)}).getReg(0);
+
+ // Pack the results back into V2S16
+ 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));
@@ -804,6 +818,8 @@ void RegBankLegalizeHelper::lower(MachineInstr &MI,
}
break;
}
+ case ScalarizeV2S16:
+ return lowerScalarizeV2S16(MI);
case WidenMMOToS32:
return widenMMOToS32(cast<GAnyLoad>(MI));
}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h
index d937815bf4714..b19686ca6f0f7 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h
@@ -74,6 +74,7 @@ class RegBankLegalizeHelper {
MachineRegisterInfo::VRegAttrs SgprRB_S32 = {SgprRB, S32};
MachineRegisterInfo::VRegAttrs VgprRB_S32 = {VgprRB, S32};
MachineRegisterInfo::VRegAttrs VccRB_S1 = {VccRB, S1};
+ MachineRegisterInfo::VRegAttrs SgprRB_S16 = {SgprRB, S16};
public:
RegBankLegalizeHelper(MachineIRBuilder &B, const MachineUniformityInfo &MUI,
@@ -124,6 +125,7 @@ class RegBankLegalizeHelper {
void lowerSplitTo32Select(MachineInstr &MI);
void lowerSplitTo32SExtInReg(MachineInstr &MI);
void lowerUnpackMinMax(MachineInstr &MI);
+ void lowerScalarizeV2S16(MachineInstr &MI);
};
} // end namespace AMDGPU
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
index 7fe8ab30c0648..144c042180f77 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
@@ -471,13 +471,11 @@ RegBankLegalizeRules::RegBankLegalizeRules(const GCNSubtarget &_ST,
.Div(S16, {{Vgpr16}, {Vgpr16, Vgpr16}})
.Uni(S32, {{Sgpr32}, {Sgpr32, Sgpr32}})
.Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32}})
- /// TODO: SALU does not support packed math addition. Scalarize into two S16 additions.
- .Uni(V2S16, {{SgprV2S16}, {Sgpr32AExt, Sgpr32AExt}})
+ .Uni(V2S16, {{SgprV2S16}, {SgprV2S16, SgprV2S16}, ScalarizeV2S16})
.Div(V2S16, {{VgprV2S16}, {VgprV2S16, VgprV2S16}})
.Uni(S64, {{Sgpr64}, {Sgpr64, Sgpr64}})
.Div(S64, {{Vgpr64}, {Vgpr64, Vgpr64}});
- /// TODO: Correct these rules, related to overflow detection.
addRulesForGOpcs({G_UADDO, G_USUBO}, Standard)
.Uni(S32, {{Sgpr32, Sgpr32Trunc}, {Sgpr32, Sgpr32}})
.Div(S32, {{Vgpr32, Vcc}, {Vgpr32, Vgpr32}});
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h
index 93e0efda77fdd..cbef81e135197 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h
@@ -223,7 +223,8 @@ enum LoweringMethodID {
UniCstExt,
SplitLoad,
WidenLoad,
- WidenMMOToS32
+ WidenMMOToS32,
+ ScalarizeV2S16
};
enum FastRulesTypes {
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll
index ba401cac2621f..e211a5163c5e0 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll
@@ -196,7 +196,9 @@ define i32 @v_add_i32(i32 %a, i32 %b) {
ret i32 %c
}
-; TODO: Add test for s_add_v2i16
+; TODO: Add test for s_add_v2i16. Instruction selector currently fails
+; to handle G_UNMERGE_VALUES. Same in GlobalISel/sub.ll.
+
define <2 x i16> @v_add_v2i16(<2 x i16> %a, <2 x i16> %b) {
; GFX7-LABEL: v_add_v2i16:
; GFX7: ; %bb.0:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
index 6fdad715f1b14..95a8a143561e2 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
@@ -14,8 +14,12 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr1
- ; CHECK-NEXT: [[ADD:%[0-9]+]]:sgpr(<2 x s16>) = G_ADD [[COPY]], [[COPY1]]
- ; CHECK-NEXT: S_ENDPGM 0, implicit [[ADD]](<2 x s16>)
+ ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s16), [[UV1:%[0-9]+]]:sgpr(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+ ; CHECK-NEXT: [[UV2:%[0-9]+]]:sgpr(s16), [[UV3:%[0-9]+]]:sgpr(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+ ; CHECK-NEXT: [[ADD:%[0-9]+]]:sgpr(s16) = G_ADD [[UV]], [[UV2]]
+ ; CHECK-NEXT: [[ADD1:%[0-9]+]]:sgpr(s16) = G_ADD [[UV1]], [[UV3]]
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR [[ADD]](s16), [[ADD1]](s16)
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[BUILD_VECTOR]](<2 x s16>)
%0:_(<2 x s16>) = COPY $sgpr0
%1:_(<2 x s16>) = COPY $sgpr1
%2:_(<2 x s16>) = G_ADD %0, %1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir
index c881ff3115e87..22a2422d62930 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir
@@ -1,3 +1,4 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
@@ -13,8 +14,12 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr1
- ; CHECK-NEXT: [[SUB:%[0-9]+]]:sgpr(<2 x s16>) = G_SUB [[COPY]], [[COPY1]]
- ; CHECK-NEXT: S_ENDPGM 0, implicit [[SUB]](<2 x s16>)
+ ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s16), [[UV1:%[0-9]+]]:sgpr(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+ ; CHECK-NEXT: [[UV2:%[0-9]+]]:sgpr(s16), [[UV3:%[0-9]+]]:sgpr(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:sgpr(s16) = G_SUB [[UV]], [[UV2]]
+ ; CHECK-NEXT: [[SUB1:%[0-9]+]]:sgpr(s16) = G_SUB [[UV1]], [[UV3]]
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR [[SUB]](s16), [[SUB1]](s16)
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[BUILD_VECTOR]](<2 x s16>)
%0:_(<2 x s16>) = COPY $sgpr0
%1:_(<2 x s16>) = COPY $sgpr1
%2:_(<2 x s16>) = G_SUB %0, %1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll
index 414e86518123d..8b5958daac168 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/sub.ll
@@ -196,7 +196,9 @@ define i32 @v_sub_i32(i32 %a, i32 %b) {
ret i32 %c
}
-; TODO: sub test for s_sub_v2i16
+; TODO: Add test for s_sub_v2i16. Instruction selector currently fails
+; to handle G_UNMERGE_VALUES.
+
define <2 x i16> @v_sub_v2i16(<2 x i16> %a, <2 x i16> %b) {
; GFX7-LABEL: v_sub_v2i16:
; GFX7: ; %bb.0:
>From a000a1b61ef80404d88df5f8c6e04d8b22b8d6b8 Mon Sep 17 00:00:00 2001
From: Anshil Gandhi <Anshil.Gandhi at amd.com>
Date: Sun, 19 Oct 2025 12:21:05 -0500
Subject: [PATCH 3/5] Handle G_TRUNC on S1 SGPR
---
llvm/lib/Target/AMDGPU/AMDGPURegBankLegalize.cpp | 6 +++---
llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp | 3 ++-
.../AMDGPU/GlobalISel/regbankselect-add-overflow.s32.mir | 2 --
llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sext.mir | 2 --
.../AMDGPU/GlobalISel/regbankselect-sub-overflow.s32.mir | 2 --
llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-zext.mir | 2 --
6 files changed, 5 insertions(+), 12 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalize.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalize.cpp
index d7932448846fa..e1879598f098a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalize.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalize.cpp
@@ -482,9 +482,9 @@ bool AMDGPURegBankLegalize::runOnMachineFunction(MachineFunction &MF) {
}
}
- // assert(!getAnySgprS1(MRI).isValid() &&
- // "Registers with sgpr reg bank and S1 LLT are not legal after "
- // "AMDGPURegBankLegalize. Should lower to sgpr S32");
+ assert(!getAnySgprS1(MRI).isValid() &&
+ "Registers with sgpr reg bank and S1 LLT are not legal after "
+ "AMDGPURegBankLegalize. Should lower to sgpr S32");
return true;
}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
index 16bf63954f4d3..8821e290d4a4f 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
@@ -1136,7 +1136,8 @@ void RegBankLegalizeHelper::applyMappingDst(
assert(RB == SgprRB);
Register NewDst = MRI.createVirtualRegister(SgprRB_S32);
Op.setReg(NewDst);
- B.buildTrunc(Reg, NewDst);
+ if (!MRI.use_empty(Reg))
+ B.buildTrunc(Reg, NewDst);
break;
}
case InvalidMapping: {
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add-overflow.s32.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add-overflow.s32.mir
index 1bc7d37033555..2c12b52b2f068 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add-overflow.s32.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add-overflow.s32.mir
@@ -14,7 +14,6 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
; CHECK-NEXT: [[UADDO:%[0-9]+]]:sgpr(s32), [[UADDO1:%[0-9]+]]:sgpr(s32) = G_UADDO [[COPY]], [[COPY1]]
- ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[UADDO1]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s32) = COPY $sgpr1
%2:_(s32), %3:_(s1) = G_UADDO %0, %1
@@ -92,7 +91,6 @@ body: |
; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[COPY2]], [[C]]
; CHECK-NEXT: [[UADDE:%[0-9]+]]:sgpr(s32), [[UADDE1:%[0-9]+]]:sgpr(s32) = G_UADDE [[COPY]], [[COPY1]], [[AND]]
- ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[UADDE1]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s32) = COPY $sgpr1
%2:_(s32) = COPY $sgpr2
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sext.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sext.mir
index 7378c9366ec36..aaf8c7c6938c2 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sext.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sext.mir
@@ -76,7 +76,6 @@ body: |
; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -1
; CHECK-NEXT: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
; CHECK-NEXT: [[SELECT:%[0-9]+]]:sgpr(s32) = G_SELECT [[AND]](s32), [[C1]], [[C2]]
- ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s16) = G_TRUNC [[SELECT]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s32) = COPY $sgpr1
%2:_(s1) = G_ICMP intpred(eq), %0, %1
@@ -214,7 +213,6 @@ body: |
; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -1
; CHECK-NEXT: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
; CHECK-NEXT: [[SELECT:%[0-9]+]]:sgpr(s32) = G_SELECT [[AND]](s32), [[C1]], [[C2]]
- ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s16) = G_TRUNC [[SELECT]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s1) = G_TRUNC %0
%2:_(s16) = G_SEXT %1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub-overflow.s32.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub-overflow.s32.mir
index f9058047d065b..dbb07fac7a0a6 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub-overflow.s32.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub-overflow.s32.mir
@@ -14,7 +14,6 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
; CHECK-NEXT: [[USUBO:%[0-9]+]]:sgpr(s32), [[USUBO1:%[0-9]+]]:sgpr(s32) = G_USUBO [[COPY]], [[COPY1]]
- ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[USUBO1]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s32) = COPY $sgpr1
%2:_(s32), %3:_(s1) = G_USUBO %0, %1
@@ -92,7 +91,6 @@ body: |
; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[COPY2]], [[C]]
; CHECK-NEXT: [[USUBE:%[0-9]+]]:sgpr(s32), [[USUBE1:%[0-9]+]]:sgpr(s32) = G_USUBE [[COPY]], [[COPY1]], [[AND]]
- ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[USUBE1]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s32) = COPY $sgpr1
%2:_(s32) = COPY $sgpr2
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-zext.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-zext.mir
index 088c20a3137f7..773675cc5e28b 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-zext.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-zext.mir
@@ -72,7 +72,6 @@ body: |
; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[ICMP]], [[C]]
; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
; CHECK-NEXT: [[SELECT:%[0-9]+]]:sgpr(s32) = G_SELECT [[AND]](s32), [[C]], [[C1]]
- ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s16) = G_TRUNC [[SELECT]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s32) = COPY $sgpr1
%2:_(s1) = G_ICMP intpred(eq), %0, %1
@@ -208,7 +207,6 @@ body: |
; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[COPY]], [[C]]
; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
; CHECK-NEXT: [[SELECT:%[0-9]+]]:sgpr(s32) = G_SELECT [[AND]](s32), [[C]], [[C1]]
- ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s16) = G_TRUNC [[SELECT]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s1) = G_TRUNC %0
%2:_(s16) = G_ZEXT %1
>From c2e9fb7cf675327f7015918560127c35363f2975 Mon Sep 17 00:00:00 2001
From: Anshil Gandhi <Anshil.Gandhi at amd.com>
Date: Mon, 20 Oct 2025 09:58:44 -0500
Subject: [PATCH 4/5] Perform S32 G_ADD instead of unpacked addition for V2S16
---
.../Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp | 14 ++++----------
.../AMDGPU/GlobalISel/regbankselect-add.v2s16.mir | 11 +++++------
.../AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir | 11 +++++------
3 files changed, 14 insertions(+), 22 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
index 8821e290d4a4f..d69178b0358c8 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
@@ -501,16 +501,10 @@ void RegBankLegalizeHelper::lowerUnpackMinMax(MachineInstr &MI) {
}
void RegBankLegalizeHelper::lowerScalarizeV2S16(MachineInstr &MI) {
- // Unpack the V2S16 operands into two S16 scalars each
- auto Op1 = B.buildUnmerge({SgprRB, S16}, MI.getOperand(1).getReg());
- auto Op2 = B.buildUnmerge({SgprRB, S16}, MI.getOperand(2).getReg());
-
- // Perform scalar additions on S16 values
- Register Lo = B.buildInstr(MI.getOpcode(), {SgprRB_S16}, {Op1.getReg(0), Op2.getReg(0)}).getReg(0);
- Register Hi = B.buildInstr(MI.getOpcode(), {SgprRB_S16}, {Op1.getReg(1), Op2.getReg(1)}).getReg(0);
-
- // Pack the results back into V2S16
- B.buildBuildVectorTrunc(MI.getOperand(0).getReg(), {Lo, Hi});
+ auto CastOp1 = B.buildBitcast(SgprRB_S32, MI.getOperand(1).getReg());
+ auto CastOp2 = B.buildBitcast(SgprRB_S32, MI.getOperand(2).getReg());
+ auto NewInstr = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {CastOp1, CastOp2});
+ B.buildBitcast({SgprRB, V2S16}, NewInstr.getReg(0));
MI.eraseFromParent();
}
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
index 95a8a143561e2..b32f214edca1a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
@@ -14,12 +14,11 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr1
- ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s16), [[UV1:%[0-9]+]]:sgpr(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
- ; CHECK-NEXT: [[UV2:%[0-9]+]]:sgpr(s16), [[UV3:%[0-9]+]]:sgpr(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
- ; CHECK-NEXT: [[ADD:%[0-9]+]]:sgpr(s16) = G_ADD [[UV]], [[UV2]]
- ; CHECK-NEXT: [[ADD1:%[0-9]+]]:sgpr(s16) = G_ADD [[UV1]], [[UV3]]
- ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR [[ADD]](s16), [[ADD1]](s16)
- ; CHECK-NEXT: S_ENDPGM 0, implicit [[BUILD_VECTOR]](<2 x s16>)
+ ; CHECK-NEXT: [[BITCAST:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY]](<2 x s16>)
+ ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>)
+ ; CHECK-NEXT: [[ADD:%[0-9]+]]:sgpr(s32) = G_ADD [[BITCAST]], [[BITCAST1]]
+ ; CHECK-NEXT: [[BITCAST2:%[0-9]+]]:sgpr(<2 x s16>) = G_BITCAST [[ADD]](s32)
+ ; CHECK-NEXT: S_ENDPGM 0, implicit %2:sgpr(<2 x s16>)
%0:_(<2 x s16>) = COPY $sgpr0
%1:_(<2 x s16>) = COPY $sgpr1
%2:_(<2 x s16>) = G_ADD %0, %1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir
index 22a2422d62930..7aae537befeaf 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sub.v2s16.mir
@@ -14,12 +14,11 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr1
- ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s16), [[UV1:%[0-9]+]]:sgpr(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
- ; CHECK-NEXT: [[UV2:%[0-9]+]]:sgpr(s16), [[UV3:%[0-9]+]]:sgpr(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
- ; CHECK-NEXT: [[SUB:%[0-9]+]]:sgpr(s16) = G_SUB [[UV]], [[UV2]]
- ; CHECK-NEXT: [[SUB1:%[0-9]+]]:sgpr(s16) = G_SUB [[UV1]], [[UV3]]
- ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR [[SUB]](s16), [[SUB1]](s16)
- ; CHECK-NEXT: S_ENDPGM 0, implicit [[BUILD_VECTOR]](<2 x s16>)
+ ; CHECK-NEXT: [[BITCAST:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY]](<2 x s16>)
+ ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>)
+ ; CHECK-NEXT: [[SUB:%[0-9]+]]:sgpr(s32) = G_SUB [[BITCAST]], [[BITCAST1]]
+ ; CHECK-NEXT: [[BITCAST2:%[0-9]+]]:sgpr(<2 x s16>) = G_BITCAST [[SUB]](s32)
+ ; CHECK-NEXT: S_ENDPGM 0, implicit %2:sgpr(<2 x s16>)
%0:_(<2 x s16>) = COPY $sgpr0
%1:_(<2 x s16>) = COPY $sgpr1
%2:_(<2 x s16>) = G_SUB %0, %1
>From 08d51b9ceb8388faa32761b93cab052055996081 Mon Sep 17 00:00:00 2001
From: Anshil Gandhi <Anshil.Gandhi at amd.com>
Date: Mon, 20 Oct 2025 11:34:33 -0500
Subject: [PATCH 5/5] Removed redundant test commands and added comments to
add.ll for the missing test
---
llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp | 3 ++-
llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll | 5 +++--
.../CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir | 3 +--
3 files changed, 6 insertions(+), 5 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
index d69178b0358c8..f3d319e3d8308 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp
@@ -503,7 +503,8 @@ void RegBankLegalizeHelper::lowerUnpackMinMax(MachineInstr &MI) {
void RegBankLegalizeHelper::lowerScalarizeV2S16(MachineInstr &MI) {
auto CastOp1 = B.buildBitcast(SgprRB_S32, MI.getOperand(1).getReg());
auto CastOp2 = B.buildBitcast(SgprRB_S32, MI.getOperand(2).getReg());
- auto NewInstr = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {CastOp1, CastOp2});
+ auto NewInstr =
+ B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {CastOp1, CastOp2});
B.buildBitcast({SgprRB, V2S16}, NewInstr.getReg(0));
MI.eraseFromParent();
}
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll
index e211a5163c5e0..e48759ae9293a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/add.ll
@@ -196,8 +196,9 @@ define i32 @v_add_i32(i32 %a, i32 %b) {
ret i32 %c
}
-; TODO: Add test for s_add_v2i16. Instruction selector currently fails
-; to handle G_UNMERGE_VALUES. Same in GlobalISel/sub.ll.
+; TODO: Add test for s_add_v2i16. The current test does not work
+; due to a bug associated with
+; AMDGPURegBankLegalizeCombiner::tryEliminateReadAnyLane(..)
define <2 x i16> @v_add_v2i16(<2 x i16> %a, <2 x i16> %b) {
; GFX7-LABEL: v_add_v2i16:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
index b32f214edca1a..c0dc1986ec612 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-add.v2s16.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=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
-# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=amdgpu-regbankselect,amdgpu-regbanklegalize %s -verify-machineinstrs -o - | FileCheck %s
---
name: add_v2s16_ss
More information about the llvm-commits
mailing list