[llvm] AMDGPU: Allocate special SGPRs before user SGPR arguments (PR #78234)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 16 00:02:56 PST 2024
https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/78234
Partially fixes #78121, but there are a few remaining issues:
1. We don't have a path to start packing excess SGPR arguments into VGPRs, and then to the stack.
2. I'm not sure if we should be allowing use of removed implicit arguments as user arguments
>From 367ff119aabd51c5e50adc718e30c25d0850d61a Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Tue, 16 Jan 2024 12:08:27 +0700
Subject: [PATCH] AMDGPU: Allocate special SGPRs before user SGPR arguments
Partially fixes #78121, but there are a few remaining issues:
1. We don't have a path to start packing excess SGPR arguments into
VGPRs, and then to the stack.
2. I'm not sure if we should be allowing use of removed implicit
arguments as user arguments
---
llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp | 11 +-
llvm/lib/Target/AMDGPU/SIISelLowering.cpp | 19 +-
.../amdgpu-simplify-libcall-pow-codegen.ll | 8 +-
llvm/test/CodeGen/AMDGPU/call-args-inreg.ll | 1583 +++++++++++++++++
.../CodeGen/AMDGPU/function-args-inreg.ll | 200 +++
.../CodeGen/AMDGPU/gfx-call-non-gfx-func.ll | 2 +-
.../AMDGPU/preserve-wwm-copy-dst-reg.ll | 48 +-
7 files changed, 1826 insertions(+), 45 deletions(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/call-args-inreg.ll
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
index cf2896f80f19bb..b9a3536036749a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
@@ -719,6 +719,10 @@ bool AMDGPUCallLowering::lowerFormalArguments(
if (!IsEntryFunc && !IsGraphics) {
// For the fixed ABI, pass workitem IDs in the last argument register.
TLI.allocateSpecialInputVGPRsFixed(CCInfo, MF, *TRI, *Info);
+
+ if (!Subtarget.enableFlatScratch())
+ CCInfo.AllocateReg(Info->getScratchRSrcReg());
+ TLI.allocateSpecialInputSGPRs(CCInfo, MF, *TRI, *Info);
}
IncomingValueAssigner Assigner(AssignFn);
@@ -732,13 +736,8 @@ bool AMDGPUCallLowering::lowerFormalArguments(
uint64_t StackSize = Assigner.StackSize;
// Start adding system SGPRs.
- if (IsEntryFunc) {
+ if (IsEntryFunc)
TLI.allocateSystemSGPRs(CCInfo, MF, *Info, CC, IsGraphics);
- } else {
- if (!Subtarget.enableFlatScratch())
- CCInfo.AllocateReg(Info->getScratchRSrcReg());
- TLI.allocateSpecialInputSGPRs(CCInfo, MF, *TRI, *Info);
- }
// When we tail call, we need to check if the callee's arguments will fit on
// the caller's stack. So, whenever we lower formal arguments, we should keep
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index a7f924114ed86c..4017b3780fab82 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -2782,15 +2782,19 @@ SDValue SITargetLowering::LowerFormalArguments(
} else if (!IsGraphics) {
// For the fixed ABI, pass workitem IDs in the last argument register.
allocateSpecialInputVGPRsFixed(CCInfo, MF, *TRI, *Info);
- }
- if (!IsKernel) {
- CCAssignFn *AssignFn = CCAssignFnForCall(CallConv, isVarArg);
- if (!IsGraphics && !Subtarget->enableFlatScratch()) {
+ if (!Subtarget->enableFlatScratch()) {
+ // FIXME: Sink this into allocateSpecialInputSGPRs
CCInfo.AllocateRegBlock(ArrayRef<MCPhysReg>{AMDGPU::SGPR0, AMDGPU::SGPR1,
AMDGPU::SGPR2, AMDGPU::SGPR3},
4);
}
+
+ allocateSpecialInputSGPRs(CCInfo, MF, *TRI, *Info);
+ }
+
+ if (!IsKernel) {
+ CCAssignFn *AssignFn = CCAssignFnForCall(CallConv, isVarArg);
CCInfo.AnalyzeFormalArguments(Splits, AssignFn);
}
@@ -2990,13 +2994,8 @@ SDValue SITargetLowering::LowerFormalArguments(
}
// Start adding system SGPRs.
- if (IsEntryFunc) {
+ if (IsEntryFunc)
allocateSystemSGPRs(CCInfo, MF, *Info, CallConv, IsGraphics);
- } else {
- CCInfo.AllocateReg(Info->getScratchRSrcReg());
- if (!IsGraphics)
- allocateSpecialInputSGPRs(CCInfo, MF, *TRI, *Info);
- }
auto &ArgUsageInfo =
DAG.getPass()->getAnalysis<AMDGPUArgumentUsageInfo>();
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pow-codegen.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pow-codegen.ll
index e53fa9650f9ec2..e65eca78106105 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pow-codegen.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pow-codegen.ll
@@ -147,13 +147,13 @@ define double @test_pow_fast_f64__integral_y(double %x, i32 %y.i) {
; CHECK-NEXT: v_writelane_b32 v40, s45, 13
; CHECK-NEXT: v_mov_b32_e32 v41, v31
; CHECK-NEXT: s_mov_b64 s[34:35], s[6:7]
+; CHECK-NEXT: v_mov_b32_e32 v42, v2
; CHECK-NEXT: s_mov_b32 s42, s15
; CHECK-NEXT: s_mov_b32 s43, s14
; CHECK-NEXT: s_mov_b32 s44, s13
; CHECK-NEXT: s_mov_b32 s45, s12
; CHECK-NEXT: s_mov_b64 s[36:37], s[10:11]
; CHECK-NEXT: s_mov_b64 s[38:39], s[8:9]
-; CHECK-NEXT: v_mov_b32_e32 v42, v2
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: v_cvt_f64_i32_e32 v[2:3], v42
@@ -286,14 +286,14 @@ define double @test_powr_fast_f64(double %x, double %y) {
; CHECK-NEXT: v_writelane_b32 v40, s45, 13
; CHECK-NEXT: v_mov_b32_e32 v43, v31
; CHECK-NEXT: s_mov_b64 s[34:35], s[6:7]
+; CHECK-NEXT: v_mov_b32_e32 v42, v3
+; CHECK-NEXT: v_mov_b32_e32 v41, v2
; CHECK-NEXT: s_mov_b32 s42, s15
; CHECK-NEXT: s_mov_b32 s43, s14
; CHECK-NEXT: s_mov_b32 s44, s13
; CHECK-NEXT: s_mov_b32 s45, s12
; CHECK-NEXT: s_mov_b64 s[36:37], s[10:11]
; CHECK-NEXT: s_mov_b64 s[38:39], s[8:9]
-; CHECK-NEXT: v_mov_b32_e32 v42, v3
-; CHECK-NEXT: v_mov_b32_e32 v41, v2
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: v_mul_f64 v[0:1], v[0:1], v[41:42]
@@ -431,13 +431,13 @@ define double @test_pown_fast_f64(double %x, i32 %y) {
; CHECK-NEXT: v_writelane_b32 v40, s45, 13
; CHECK-NEXT: v_mov_b32_e32 v41, v31
; CHECK-NEXT: s_mov_b64 s[34:35], s[6:7]
+; CHECK-NEXT: v_mov_b32_e32 v42, v2
; CHECK-NEXT: s_mov_b32 s42, s15
; CHECK-NEXT: s_mov_b32 s43, s14
; CHECK-NEXT: s_mov_b32 s44, s13
; CHECK-NEXT: s_mov_b32 s45, s12
; CHECK-NEXT: s_mov_b64 s[36:37], s[10:11]
; CHECK-NEXT: s_mov_b64 s[38:39], s[8:9]
-; CHECK-NEXT: v_mov_b32_e32 v42, v2
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: v_cvt_f64_i32_e32 v[2:3], v42
diff --git a/llvm/test/CodeGen/AMDGPU/call-args-inreg.ll b/llvm/test/CodeGen/AMDGPU/call-args-inreg.ll
new file mode 100644
index 00000000000000..42e9dce3747768
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/call-args-inreg.ll
@@ -0,0 +1,1583 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefix=GFX9 %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -enable-var-scope -check-prefix=GFX11 %s
+
+declare hidden void @external_void_func_i8_inreg(i8 inreg) #0
+declare hidden void @external_void_func_i16_inreg(i32 inreg) #0
+declare hidden void @external_void_func_i32_inreg(i32 inreg) #0
+declare hidden void @external_void_func_i64_inreg(i64 inreg) #0
+declare hidden void @external_void_func_v2i32_inreg(<2 x i32> inreg) #0
+declare hidden void @external_void_func_v3i32_inreg(<3 x i32> inreg) #0
+declare hidden void @external_void_func_v4i32_inreg(<4 x i32> inreg) #0
+declare hidden void @external_void_func_v8i32_inreg(<8 x i32> inreg) #0
+declare hidden void @external_void_func_v16i32_inreg(<16 x i32> inreg) #0
+declare hidden void @external_void_func_f16_inreg(half inreg) #0
+declare hidden void @external_void_func_bf16_inreg(bfloat inreg) #0
+declare hidden void @external_void_func_f32_inreg(float inreg) #0
+declare hidden void @external_void_func_f64_inreg(double inreg) #0
+declare hidden void @external_void_func_v2f16_inreg(<2 x half> inreg) #0
+declare hidden void @external_void_func_v2bf16_inreg(<2 x bfloat> inreg) #0
+declare hidden void @external_void_func_v3f16_inreg(<3 x half> inreg) #0
+declare hidden void @external_void_func_v4f16_inreg(<4 x half> inreg) #0
+
+declare hidden void @external_void_func_p0_inreg(ptr inreg) #0
+declare hidden void @external_void_func_p1_inreg(ptr addrspace(1) inreg) #0
+declare hidden void @external_void_func_p3_inreg(ptr addrspace(3) inreg) #0
+declare hidden void @external_void_func_v2p1_inreg(<2 x ptr addrspace(1)> inreg) #0
+declare hidden void @external_void_func_v2p5_inreg(<2 x ptr addrspace(5)> inreg) #0
+
+declare hidden void @external_void_func_i64_inreg_i32_inreg_i64_inreg(i64 inreg, i32 inreg, i64 inreg) #0
+
+declare hidden void @external_void_func_a15i32_inreg([15 x i32] inreg) #0
+declare hidden void @external_void_func_a16i32_inreg([16 x i32] inreg) #0
+declare hidden void @external_void_func_a15i32_inreg_i32_inreg([15 x i32] inreg, i32 inreg) #0
+declare hidden void @external_void_func_a15i32_inreg_i32_inreg__noimplicit([15 x i32] inreg, i32 inreg) #1
+
+define void @test_call_external_void_func_i8_inreg(i8 inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_i8_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_i8_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_i8_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_i8_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s1, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s2, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s2
+; GFX11-NEXT: v_writelane_b32 v40, s1, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_i8_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_i8_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_i8_inreg(i8 inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_i16_inreg(i16 inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_i16_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_i16_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_i16_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_i16_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s1, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s2, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s2
+; GFX11-NEXT: v_writelane_b32 v40, s1, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_i16_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_i16_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_i16_inreg(i16 inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_i32_inreg(i32 inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_i32_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_i32_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s1, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s2, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s2
+; GFX11-NEXT: v_writelane_b32 v40, s1, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_i32_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_i32_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_i32_inreg(i32 inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_i64_inreg(i64 inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_i64_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_i64_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_i64_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_i64_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s2, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s3, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s3
+; GFX11-NEXT: v_writelane_b32 v40, s2, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_i64_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_i64_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_i64_inreg(i64 inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_v2i32_inreg(<2 x i32> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v2i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_v2i32_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_v2i32_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v2i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s2, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s3, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s3
+; GFX11-NEXT: v_writelane_b32 v40, s2, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_v2i32_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_v2i32_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v2i32_inreg(<2 x i32> inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_v3i32_inreg(<3 x i32> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v3i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s17, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s17, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s2, s16
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[18:19]
+; GFX9-NEXT: s_add_u32 s18, s18, external_void_func_v3i32_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s19, s19, external_void_func_v3i32_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[18:19]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v3i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s3, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s16, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s16
+; GFX11-NEXT: v_writelane_b32 v40, s3, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[16:17]
+; GFX11-NEXT: s_add_u32 s16, s16, external_void_func_v3i32_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s17, s17, external_void_func_v3i32_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v3i32_inreg(<3 x i32> inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_v4i32_inreg(<4 x i32> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v4i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s18, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-NEXT: v_writelane_b32 v40, s18, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s3, s17
+; GFX9-NEXT: s_mov_b32 s2, s16
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[18:19]
+; GFX9-NEXT: s_add_u32 s18, s18, external_void_func_v4i32_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s19, s19, external_void_func_v4i32_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[18:19]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v4i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s16, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s17, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s17
+; GFX11-NEXT: v_writelane_b32 v40, s16, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[16:17]
+; GFX11-NEXT: s_add_u32 s16, s16, external_void_func_v4i32_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s17, s17, external_void_func_v4i32_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v4i32_inreg(<4 x i32> inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_v8i32_inreg(<8 x i32> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v8i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s22, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[24:25], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[24:25]
+; GFX9-NEXT: v_writelane_b32 v40, s22, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s3, s17
+; GFX9-NEXT: s_mov_b32 s2, s16
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: s_mov_b32 s16, s18
+; GFX9-NEXT: s_mov_b32 s17, s19
+; GFX9-NEXT: s_mov_b32 s18, s20
+; GFX9-NEXT: s_mov_b32 s19, s21
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[22:23]
+; GFX9-NEXT: s_add_u32 s22, s22, external_void_func_v8i32_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s23, s23, external_void_func_v8i32_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[22:23]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v8i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s18, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s19, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s19
+; GFX11-NEXT: v_writelane_b32 v40, s18, 2
+; GFX11-NEXT: s_mov_b32 s19, s17
+; GFX11-NEXT: s_mov_b32 s18, s16
+; GFX11-NEXT: s_mov_b32 s17, s7
+; GFX11-NEXT: s_mov_b32 s16, s6
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[20:21]
+; GFX11-NEXT: s_add_u32 s20, s20, external_void_func_v8i32_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s21, s21, external_void_func_v8i32_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[20:21]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v8i32_inreg(<8 x i32> inreg %arg)
+ ret void
+}
+
+; FIXME:
+; define void @test_call_external_void_func_v16i32_inreg(<16 x i32> inreg %arg) #0 {
+; call void @external_void_func_v16i32_inreg(<16 x i32> inreg %arg)
+; ret void
+; }
+
+define void @test_call_external_void_func_f16_inreg(half inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_f16_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_f16_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_f16_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_f16_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s1, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s2, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s2
+; GFX11-NEXT: v_writelane_b32 v40, s1, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_f16_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_f16_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_f16_inreg(half inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_bf16_inreg(bfloat inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_bf16_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_bf16_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_bf16_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_bf16_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s1, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s2, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s2
+; GFX11-NEXT: v_writelane_b32 v40, s1, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_bf16_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_bf16_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_bf16_inreg(bfloat inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_f32_inreg(float inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_f32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_f32_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_f32_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_f32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s1, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s2, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s2
+; GFX11-NEXT: v_writelane_b32 v40, s1, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_f32_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_f32_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_f32_inreg(float inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_f64_inreg(double inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_f64_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_f64_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_f64_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_f64_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s2, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s3, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s3
+; GFX11-NEXT: v_writelane_b32 v40, s2, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_f64_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_f64_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_f64_inreg(double inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_v2f16_inreg(<2 x half> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v2f16_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_v2f16_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_v2f16_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v2f16_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s1, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s2, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s2
+; GFX11-NEXT: v_writelane_b32 v40, s1, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_v2f16_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_v2f16_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v2f16_inreg(<2 x half> inreg %arg)
+ ret void
+}
+
+
+define void @test_call_external_void_func_v2bf16_inreg(<2 x bfloat> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v2bf16_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_v2bf16_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_v2bf16_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v2bf16_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s1, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s2, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s2
+; GFX11-NEXT: v_writelane_b32 v40, s1, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_v2bf16_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_v2bf16_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v2bf16_inreg(<2 x bfloat> inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_v3f16_inreg(<3 x half> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v3f16_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_v3f16_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_v3f16_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v3f16_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s2, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s3, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s3
+; GFX11-NEXT: v_writelane_b32 v40, s2, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_v3f16_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_v3f16_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v3f16_inreg(<3 x half> inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_v4f16_inreg(<4 x half> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v4f16_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_v4f16_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_v4f16_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v4f16_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s2, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s3, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s3
+; GFX11-NEXT: v_writelane_b32 v40, s2, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_v4f16_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_v4f16_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v4f16_inreg(<4 x half> inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_p0_inreg(ptr inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_p0_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_p0_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_p0_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_p0_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s2, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s3, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s3
+; GFX11-NEXT: v_writelane_b32 v40, s2, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_p0_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_p0_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_p0_inreg(ptr inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_p1_inreg(ptr addrspace(1) inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_p1_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_p1_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_p1_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_p1_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s2, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s3, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s3
+; GFX11-NEXT: v_writelane_b32 v40, s2, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_p1_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_p1_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_p1_inreg(ptr addrspace(1) inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_p3_inreg(ptr addrspace(3) inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_p3_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_p3_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_p3_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_p3_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s1, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s2, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s2
+; GFX11-NEXT: v_writelane_b32 v40, s1, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_p3_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_p3_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_p3_inreg(ptr addrspace(3) inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_v2p1_inreg(<2 x ptr addrspace(1)> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v2p1_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s18, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-NEXT: v_writelane_b32 v40, s18, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s3, s17
+; GFX9-NEXT: s_mov_b32 s2, s16
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[18:19]
+; GFX9-NEXT: s_add_u32 s18, s18, external_void_func_v2p1_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s19, s19, external_void_func_v2p1_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[18:19]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v2p1_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s16, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s17, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s17
+; GFX11-NEXT: v_writelane_b32 v40, s16, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[16:17]
+; GFX11-NEXT: s_add_u32 s16, s16, external_void_func_v2p1_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s17, s17, external_void_func_v2p1_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v2p1_inreg(<2 x ptr addrspace(1)> inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_v2p5_inreg(<2 x ptr addrspace(5)> inreg %arg) #0 {
+; GFX9-LABEL: test_call_external_void_func_v2p5_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s16, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-NEXT: v_writelane_b32 v40, s16, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[16:17]
+; GFX9-NEXT: s_add_u32 s16, s16, external_void_func_v2p5_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s17, s17, external_void_func_v2p5_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_v2p5_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s2, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s3, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s3
+; GFX11-NEXT: v_writelane_b32 v40, s2, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[2:3]
+; GFX11-NEXT: s_add_u32 s2, s2, external_void_func_v2p5_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s3, s3, external_void_func_v2p5_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_v2p5_inreg(<2 x ptr addrspace(5)> inreg %arg)
+ ret void
+}
+
+define void @test_call_external_void_func_i64_inreg_i32_inreg_i64_inreg(i64 inreg %arg0, i32 inreg %arg1, i64 inreg %arg2) #0 {
+; GFX9-LABEL: test_call_external_void_func_i64_inreg_i32_inreg_i64_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s19, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-NEXT: v_writelane_b32 v40, s19, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s3, s17
+; GFX9-NEXT: s_mov_b32 s2, s16
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: s_mov_b32 s16, s18
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[20:21]
+; GFX9-NEXT: s_add_u32 s20, s20, external_void_func_i64_inreg_i32_inreg_i64_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s21, s21, external_void_func_i64_inreg_i32_inreg_i64_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[20:21]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_i64_inreg_i32_inreg_i64_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s16, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s17, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s17
+; GFX11-NEXT: v_writelane_b32 v40, s16, 2
+; GFX11-NEXT: ; kill: def $sgpr6_sgpr7 killed $sgpr6_sgpr7 killed $sgpr7
+; GFX11-NEXT: s_mov_b32 s16, s6
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[18:19]
+; GFX11-NEXT: s_add_u32 s18, s18, external_void_func_i64_inreg_i32_inreg_i64_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s19, s19, external_void_func_i64_inreg_i32_inreg_i64_inreg at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[18:19]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_i64_inreg_i32_inreg_i64_inreg(i64 inreg %arg0, i32 inreg %arg1, i64 inreg %arg2)
+ ret void
+}
+
+define void @test_call_external_void_func_a15i32_inreg([15 x i32] inreg %arg0) #0 {
+; GFX9-LABEL: test_call_external_void_func_a15i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s29, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 vcc, -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, vcc
+; GFX9-NEXT: v_writelane_b32 v40, s29, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s3, s17
+; GFX9-NEXT: s_mov_b32 s2, s16
+; GFX9-NEXT: s_mov_b32 s1, s7
+; GFX9-NEXT: s_mov_b32 s0, s6
+; GFX9-NEXT: s_mov_b32 s16, s18
+; GFX9-NEXT: s_mov_b32 s17, s19
+; GFX9-NEXT: s_mov_b32 s18, s20
+; GFX9-NEXT: s_mov_b32 s19, s21
+; GFX9-NEXT: s_mov_b32 s20, s22
+; GFX9-NEXT: s_mov_b32 s21, s23
+; GFX9-NEXT: s_mov_b32 s22, s24
+; GFX9-NEXT: s_mov_b32 s23, s25
+; GFX9-NEXT: s_mov_b32 s24, s26
+; GFX9-NEXT: s_mov_b32 s25, s27
+; GFX9-NEXT: s_mov_b32 s26, s28
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 vcc
+; GFX9-NEXT: s_add_u32 vcc_lo, vcc_lo, external_void_func_a15i32_inreg at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 vcc_hi, vcc_hi, external_void_func_a15i32_inreg at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], vcc
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_a15i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s25, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s26, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s26
+; GFX11-NEXT: v_writelane_b32 v40, s25, 2
+; GFX11-NEXT: s_mov_b32 s26, s24
+; GFX11-NEXT: s_mov_b32 s25, s23
+; GFX11-NEXT: s_mov_b32 s24, s22
+; GFX11-NEXT: s_mov_b32 s23, s21
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: s_mov_b32 s22, s20
+; GFX11-NEXT: s_mov_b32 s21, s19
+; GFX11-NEXT: s_mov_b32 s20, s18
+; GFX11-NEXT: s_mov_b32 s19, s17
+; GFX11-NEXT: s_mov_b32 s18, s16
+; GFX11-NEXT: s_mov_b32 s17, s7
+; GFX11-NEXT: s_mov_b32 s16, s6
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_getpc_b64 s[28:29]
+; GFX11-NEXT: s_add_u32 s28, s28, external_void_func_a15i32_inreg at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s29, s29, external_void_func_a15i32_inreg at rel32@hi+12
+; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[28:29]
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_a15i32_inreg([15 x i32] inreg %arg0)
+ ret void
+}
+
+; FIXME:
+; define void @test_call_external_void_func_a16i32_inreg([16 x i32] inreg %arg0) #0 {
+; call void @external_void_func_a16i32_inreg([16 x i32] inreg %arg0)
+; ret void
+; }
+
+; FIXME:
+; define void @test_call_external_void_func_a15i32_inreg_i32_inreg([15 x i32] inreg %arg0, i32 inreg %arg1) #0 {
+; call void @external_void_func_a15i32_inreg_i32_inreg([15 x i32] inreg %arg0, i32 inreg %arg1)
+; ret void
+; }
+
+
+; FIXME: This should also fail
+define void @test_call_external_void_func_a15i32_inreg_i32_inreg([15 x i32] inreg %arg0, i32 inreg %arg1) #1 {
+; GFX9-LABEL: test_call_external_void_func_a15i32_inreg_i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b32 s23, s33
+; GFX9-NEXT: s_mov_b32 s33, s32
+; GFX9-NEXT: s_or_saveexec_b64 s[24:25], -1
+; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+; GFX9-NEXT: s_mov_b64 exec, s[24:25]
+; GFX9-NEXT: v_writelane_b32 v40, s23, 2
+; GFX9-NEXT: s_addk_i32 s32, 0x400
+; GFX9-NEXT: v_writelane_b32 v40, s30, 0
+; GFX9-NEXT: s_mov_b32 s3, s7
+; GFX9-NEXT: s_mov_b32 s2, s6
+; GFX9-NEXT: s_mov_b32 s1, s5
+; GFX9-NEXT: s_mov_b32 s0, s4
+; GFX9-NEXT: s_mov_b32 s4, s8
+; GFX9-NEXT: s_mov_b32 s5, s9
+; GFX9-NEXT: s_mov_b32 s6, s10
+; GFX9-NEXT: s_mov_b32 s7, s11
+; GFX9-NEXT: s_mov_b32 s8, s15
+; GFX9-NEXT: s_mov_b32 s9, s16
+; GFX9-NEXT: s_mov_b32 s10, s17
+; GFX9-NEXT: s_mov_b32 s11, s18
+; GFX9-NEXT: s_mov_b32 s15, s19
+; GFX9-NEXT: s_mov_b32 s16, s20
+; GFX9-NEXT: s_mov_b32 s17, s21
+; GFX9-NEXT: s_mov_b32 s18, s22
+; GFX9-NEXT: v_writelane_b32 v40, s31, 1
+; GFX9-NEXT: s_getpc_b64 s[24:25]
+; GFX9-NEXT: s_add_u32 s24, s24, external_void_func_a15i32_inreg_i32_inreg__noimplicit at rel32@lo+4
+; GFX9-NEXT: s_addc_u32 s25, s25, external_void_func_a15i32_inreg_i32_inreg__noimplicit at rel32@hi+12
+; GFX9-NEXT: s_swappc_b64 s[30:31], s[24:25]
+; GFX9-NEXT: v_readlane_b32 s31, v40, 1
+; GFX9-NEXT: v_readlane_b32 s30, v40, 0
+; GFX9-NEXT: v_readlane_b32 s4, v40, 2
+; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1
+; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
+; GFX9-NEXT: s_mov_b64 exec, s[6:7]
+; GFX9-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-NEXT: s_mov_b32 s33, s4
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: test_call_external_void_func_a15i32_inreg_i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: s_mov_b32 s19, s33
+; GFX11-NEXT: s_mov_b32 s33, s32
+; GFX11-NEXT: s_or_saveexec_b32 s20, -1
+; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill
+; GFX11-NEXT: s_mov_b32 exec_lo, s20
+; GFX11-NEXT: v_writelane_b32 v40, s19, 2
+; GFX11-NEXT: s_add_i32 s32, s32, 16
+; GFX11-NEXT: s_getpc_b64 s[20:21]
+; GFX11-NEXT: s_add_u32 s20, s20, external_void_func_a15i32_inreg_i32_inreg__noimplicit at rel32@lo+4
+; GFX11-NEXT: s_addc_u32 s21, s21, external_void_func_a15i32_inreg_i32_inreg__noimplicit at rel32@hi+12
+; GFX11-NEXT: v_writelane_b32 v40, s30, 0
+; GFX11-NEXT: v_writelane_b32 v40, s31, 1
+; GFX11-NEXT: s_swappc_b64 s[30:31], s[20:21]
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT: v_readlane_b32 s31, v40, 1
+; GFX11-NEXT: v_readlane_b32 s30, v40, 0
+; GFX11-NEXT: v_readlane_b32 s0, v40, 2
+; GFX11-NEXT: s_or_saveexec_b32 s1, -1
+; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload
+; GFX11-NEXT: s_mov_b32 exec_lo, s1
+; GFX11-NEXT: s_add_i32 s32, s32, -16
+; GFX11-NEXT: s_mov_b32 s33, s0
+; GFX11-NEXT: s_waitcnt vmcnt(0)
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ call void @external_void_func_a15i32_inreg_i32_inreg__noimplicit([15 x i32] inreg %arg0, i32 inreg %arg1)
+ ret void
+}
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-queue-ptr" "amdgpu-no-work-group-id-x" "amdgpu-no-work-group-id-y" "amdgpu-no-work-group-id-z" "amdgpu-no-work-item-id-x" "amdgpu-no-work-item-id-y" "amdgpu-no-work-item-id-z" }
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 1, !"amdgpu_code_object_version", i32 500}
diff --git a/llvm/test/CodeGen/AMDGPU/function-args-inreg.ll b/llvm/test/CodeGen/AMDGPU/function-args-inreg.ll
index 5f2a8c4ba98247..44a9127b4bd09c 100644
--- a/llvm/test/CodeGen/AMDGPU/function-args-inreg.ll
+++ b/llvm/test/CodeGen/AMDGPU/function-args-inreg.ll
@@ -2007,6 +2007,206 @@ define void @void_func_v16bf16_inreg(<16 x bfloat> inreg %arg0) #0 {
ret void
}
+define void @void_func_2_i32_inreg(i32 inreg %arg0, i32 inreg %arg1, ptr addrspace(1) %ptr) {
+; GFX9-LABEL: void_func_2_i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-NEXT: global_store_dword v[0:1], v2, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s5
+; GFX9-NEXT: global_store_dword v[0:1], v2, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: void_func_2_i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v2, s0 :: v_dual_mov_b32 v3, s1
+; GFX11-NEXT: global_store_b32 v[0:1], v2, off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: global_store_b32 v[0:1], v3, off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ store volatile i32 %arg0, ptr addrspace(1) %ptr
+ store volatile i32 %arg1, ptr addrspace(1) %ptr
+ ret void
+}
+
+define void @void_func_2_i64_inreg(i64 inreg %arg0, i64 inreg %arg1, ptr addrspace(1) %ptr) {
+; GFX9-LABEL: void_func_2_i64_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v4, s4
+; GFX9-NEXT: v_mov_b32_e32 v5, s5
+; GFX9-NEXT: v_mov_b32_e32 v2, s6
+; GFX9-NEXT: v_mov_b32_e32 v3, s7
+; GFX9-NEXT: global_store_dwordx2 v[0:1], v[4:5], off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: global_store_dwordx2 v[0:1], v[2:3], off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: void_func_2_i64_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v5, s1
+; GFX11-NEXT: v_dual_mov_b32 v4, s0 :: v_dual_mov_b32 v3, s3
+; GFX11-NEXT: global_store_b64 v[0:1], v[4:5], off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: global_store_b64 v[0:1], v[2:3], off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ store volatile i64 %arg0, ptr addrspace(1) %ptr
+ store volatile i64 %arg1, ptr addrspace(1) %ptr
+ ret void
+}
+
+define void @void_func_i64_inreg_i32_inreg_i64_inreg(i64 inreg %arg0, i32 inreg %arg1, i64 inreg %arg2, ptr addrspace(1) %ptr) {
+; GFX9-LABEL: void_func_i64_inreg_i32_inreg_i64_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v4, s4
+; GFX9-NEXT: v_mov_b32_e32 v5, s5
+; GFX9-NEXT: global_store_dwordx2 v[0:1], v[4:5], off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v4, s6
+; GFX9-NEXT: v_mov_b32_e32 v2, s7
+; GFX9-NEXT: v_mov_b32_e32 v3, s8
+; GFX9-NEXT: global_store_dword v[0:1], v4, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: global_store_dwordx2 v[0:1], v[2:3], off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: void_func_i64_inreg_i32_inreg_i64_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v2, s3 :: v_dual_mov_b32 v5, s1
+; GFX11-NEXT: v_dual_mov_b32 v4, s0 :: v_dual_mov_b32 v3, s4
+; GFX11-NEXT: v_mov_b32_e32 v6, s2
+; GFX11-NEXT: global_store_b64 v[0:1], v[4:5], off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: global_store_b32 v[0:1], v6, off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: global_store_b64 v[0:1], v[2:3], off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ store volatile i64 %arg0, ptr addrspace(1) %ptr
+ store volatile i32 %arg1, ptr addrspace(1) %ptr
+ store volatile i64 %arg2, ptr addrspace(1) %ptr
+ ret void
+}
+
+define void @void_func_5_i32_inreg(i32 inreg %arg0, i32 inreg %arg1, i32 inreg %arg2, i32 inreg %arg3, i32 inreg %arg4, ptr addrspace(1) %ptr) {
+; GFX9-LABEL: void_func_5_i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-NEXT: global_store_dword v[0:1], v2, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s5
+; GFX9-NEXT: global_store_dword v[0:1], v2, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s6
+; GFX9-NEXT: global_store_dword v[0:1], v2, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s7
+; GFX9-NEXT: global_store_dword v[0:1], v2, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s8
+; GFX9-NEXT: global_store_dword v[0:1], v2, off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: void_func_5_i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v2, s0 :: v_dual_mov_b32 v3, s1
+; GFX11-NEXT: v_dual_mov_b32 v4, s2 :: v_dual_mov_b32 v5, s3
+; GFX11-NEXT: v_mov_b32_e32 v6, s4
+; GFX11-NEXT: global_store_b32 v[0:1], v2, off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: global_store_b32 v[0:1], v3, off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: global_store_b32 v[0:1], v4, off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: global_store_b32 v[0:1], v5, off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: global_store_b32 v[0:1], v6, off dlc
+; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ store volatile i32 %arg0, ptr addrspace(1) %ptr
+ store volatile i32 %arg1, ptr addrspace(1) %ptr
+ store volatile i32 %arg2, ptr addrspace(1) %ptr
+ store volatile i32 %arg3, ptr addrspace(1) %ptr
+ store volatile i32 %arg4, ptr addrspace(1) %ptr
+ ret void
+}
+
+define void @void_func_a5i32_inreg([5 x i32] inreg %arg0, ptr addrspace(1) %ptr) {
+; GFX9-LABEL: void_func_a5i32_inreg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_mov_b32_e32 v2, s8
+; GFX9-NEXT: global_store_dword v[0:1], v2, off offset:16
+; GFX9-NEXT: v_mov_b32_e32 v5, s7
+; GFX9-NEXT: v_mov_b32_e32 v4, s6
+; GFX9-NEXT: v_mov_b32_e32 v3, s5
+; GFX9-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-NEXT: global_store_dwordx4 v[0:1], v[2:5], off
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: void_func_a5i32_inreg:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_dual_mov_b32 v6, s4 :: v_dual_mov_b32 v5, s3
+; GFX11-NEXT: v_dual_mov_b32 v4, s2 :: v_dual_mov_b32 v3, s1
+; GFX11-NEXT: v_mov_b32_e32 v2, s0
+; GFX11-NEXT: s_clause 0x1
+; GFX11-NEXT: global_store_b32 v[0:1], v6, off offset:16
+; GFX11-NEXT: global_store_b128 v[0:1], v[2:5], off
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+ store [5 x i32] %arg0, ptr addrspace(1) %ptr
+ ret void
+}
+
+; Force all implicit inputs to be required
+declare void @extern()
+
+define void @void_func_a13i32_inreg([13 x i32] inreg %arg0, ptr addrspace(1) %ptr) {
+ store [13 x i32] %arg0, ptr addrspace(1) %ptr
+ call void @extern()
+ ret void
+}
+
+; define void @void_func_a14i32_inreg([14 x i32] inreg %arg0, ptr addrspace(1) %ptr) {
+; store [14 x i32] %arg0, ptr addrspace(1) %ptr
+; call void @extern()
+; ret void
+; }
+
+; FIXME:
+; define void @void_func_a15i32_inreg([15 x i32] inreg %arg0, ptr addrspace(1) %ptr) {
+; store [15 x i32] %arg0, ptr addrspace(1) %ptr
+; call void @extern()
+; ret void
+; }
+
+; FIXME:
+; define void @void_func_a16i32_inreg([16 x i32] inreg %arg0, ptr addrspace(1) %ptr) {
+; store [16 x i32] %arg0, ptr addrspace(1) %ptr
+; call void @extern()
+; ret void
+; }
+
+; FIXME: Should still fail
+define void @void_func_a16i32_inreg__noimplicit([16 x i32] inreg %arg0, ptr addrspace(1) %ptr) {
+ store [16 x i32] %arg0, ptr addrspace(1) %ptr
+ ret void
+}
+
attributes #0 = { nounwind }
attributes #1 = { nounwind noinline }
diff --git a/llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll b/llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll
index 9ab8be0485eddb..bb0e116cb4d324 100644
--- a/llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll
+++ b/llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll
@@ -119,7 +119,7 @@ define amdgpu_gfx void @gfx_func() {
; GISEL-NEXT: v_writelane_b32 v40, s30, 26
; GISEL-NEXT: s_mov_b32 s34, extern_c_func at abs32@lo
; GISEL-NEXT: s_mov_b32 s35, extern_c_func at abs32@hi
-; GISEL-NEXT: s_mov_b64 s[8:9], s[4:5]
+; GISEL-NEXT: s_mov_b64 s[8:9], 0
; GISEL-NEXT: s_addk_i32 s32, 0x400
; GISEL-NEXT: v_writelane_b32 v40, s31, 27
; GISEL-NEXT: s_swappc_b64 s[30:31], s[34:35]
diff --git a/llvm/test/CodeGen/AMDGPU/preserve-wwm-copy-dst-reg.ll b/llvm/test/CodeGen/AMDGPU/preserve-wwm-copy-dst-reg.ll
index 132c955827980e..b63c8038f9c81b 100644
--- a/llvm/test/CodeGen/AMDGPU/preserve-wwm-copy-dst-reg.ll
+++ b/llvm/test/CodeGen/AMDGPU/preserve-wwm-copy-dst-reg.ll
@@ -20,14 +20,14 @@ define void @preserve_wwm_copy_dstreg(ptr %parg0, ptr %parg1, ptr %parg2) #0 {
; GFX906-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:148 ; 4-byte Folded Spill
; GFX906-NEXT: s_mov_b64 exec, s[18:19]
; GFX906-NEXT: ; implicit-def: $vgpr2
-; GFX906-NEXT: s_mov_b32 s22, s15
+; GFX906-NEXT: s_mov_b32 s21, s15
; GFX906-NEXT: v_writelane_b32 v2, s6, 0
; GFX906-NEXT: v_writelane_b32 v2, s7, 1
-; GFX906-NEXT: v_writelane_b32 v2, s22, 2
-; GFX906-NEXT: s_mov_b32 s23, s14
-; GFX906-NEXT: v_writelane_b32 v2, s23, 3
-; GFX906-NEXT: s_mov_b32 s21, s13
-; GFX906-NEXT: v_writelane_b32 v2, s21, 4
+; GFX906-NEXT: v_writelane_b32 v2, s21, 2
+; GFX906-NEXT: s_mov_b32 s22, s14
+; GFX906-NEXT: v_writelane_b32 v2, s22, 3
+; GFX906-NEXT: s_mov_b32 s23, s13
+; GFX906-NEXT: v_writelane_b32 v2, s23, 4
; GFX906-NEXT: s_mov_b32 s24, s12
; GFX906-NEXT: v_writelane_b32 v2, s24, 5
; GFX906-NEXT: s_mov_b64 s[26:27], s[10:11]
@@ -41,13 +41,13 @@ define void @preserve_wwm_copy_dstreg(ptr %parg0, ptr %parg1, ptr %parg2) #0 {
; GFX906-NEXT: s_addk_i32 s32, 0x2800
; GFX906-NEXT: v_writelane_b32 v41, s31, 1
; GFX906-NEXT: v_mov_b32_e32 v32, v31
+; GFX906-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; GFX906-NEXT: s_waitcnt vmcnt(0)
+; GFX906-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
; GFX906-NEXT: v_writelane_b32 v2, s5, 11
; GFX906-NEXT: s_or_saveexec_b64 s[34:35], -1
; GFX906-NEXT: v_mov_b32_e32 v33, v2
; GFX906-NEXT: s_mov_b64 exec, s[34:35]
-; GFX906-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
-; GFX906-NEXT: s_waitcnt vmcnt(0)
-; GFX906-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
; GFX906-NEXT: ;;#ASMSTART
; GFX906-NEXT: ; def v[0:31]
; GFX906-NEXT: ;;#ASMEND
@@ -141,10 +141,10 @@ define void @preserve_wwm_copy_dstreg(ptr %parg0, ptr %parg1, ptr %parg2) #0 {
; GFX906-NEXT: s_mov_b64 exec, s[34:35]
; GFX906-NEXT: v_readlane_b32 s16, v40, 22
; GFX906-NEXT: s_mov_b32 s12, s24
-; GFX906-NEXT: s_mov_b32 s13, s21
-; GFX906-NEXT: s_mov_b32 s14, s23
+; GFX906-NEXT: s_mov_b32 s13, s23
+; GFX906-NEXT: s_mov_b32 s14, s22
; GFX906-NEXT: v_mov_b32_e32 v31, v32
-; GFX906-NEXT: s_mov_b32 s15, s22
+; GFX906-NEXT: s_mov_b32 s15, s21
; GFX906-NEXT: s_mov_b64 s[10:11], s[26:27]
; GFX906-NEXT: v_readlane_b32 s17, v40, 23
; GFX906-NEXT: buffer_store_dword v32, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
@@ -399,14 +399,14 @@ define void @preserve_wwm_copy_dstreg(ptr %parg0, ptr %parg1, ptr %parg2) #0 {
; GFX908-NEXT: s_waitcnt vmcnt(0)
; GFX908-NEXT: s_mov_b64 exec, s[16:17]
; GFX908-NEXT: ; implicit-def: $vgpr2
-; GFX908-NEXT: s_mov_b32 s22, s15
+; GFX908-NEXT: s_mov_b32 s21, s15
; GFX908-NEXT: v_writelane_b32 v2, s6, 0
; GFX908-NEXT: v_writelane_b32 v2, s7, 1
-; GFX908-NEXT: v_writelane_b32 v2, s22, 2
-; GFX908-NEXT: s_mov_b32 s23, s14
-; GFX908-NEXT: v_writelane_b32 v2, s23, 3
-; GFX908-NEXT: s_mov_b32 s21, s13
-; GFX908-NEXT: v_writelane_b32 v2, s21, 4
+; GFX908-NEXT: v_writelane_b32 v2, s21, 2
+; GFX908-NEXT: s_mov_b32 s22, s14
+; GFX908-NEXT: v_writelane_b32 v2, s22, 3
+; GFX908-NEXT: s_mov_b32 s23, s13
+; GFX908-NEXT: v_writelane_b32 v2, s23, 4
; GFX908-NEXT: s_mov_b32 s24, s12
; GFX908-NEXT: v_writelane_b32 v2, s24, 5
; GFX908-NEXT: s_mov_b64 s[26:27], s[10:11]
@@ -416,13 +416,13 @@ define void @preserve_wwm_copy_dstreg(ptr %parg0, ptr %parg1, ptr %parg2) #0 {
; GFX908-NEXT: v_writelane_b32 v2, s9, 9
; GFX908-NEXT: v_writelane_b32 v2, s4, 10
; GFX908-NEXT: v_mov_b32_e32 v32, v31
+; GFX908-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
+; GFX908-NEXT: s_waitcnt vmcnt(0)
+; GFX908-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
; GFX908-NEXT: v_writelane_b32 v2, s5, 11
; GFX908-NEXT: s_or_saveexec_b64 s[34:35], -1
; GFX908-NEXT: v_mov_b32_e32 v33, v2
; GFX908-NEXT: s_mov_b64 exec, s[34:35]
-; GFX908-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
-; GFX908-NEXT: s_waitcnt vmcnt(0)
-; GFX908-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
; GFX908-NEXT: ;;#ASMSTART
; GFX908-NEXT: ; def v[0:31]
; GFX908-NEXT: ;;#ASMEND
@@ -516,10 +516,10 @@ define void @preserve_wwm_copy_dstreg(ptr %parg0, ptr %parg1, ptr %parg2) #0 {
; GFX908-NEXT: s_mov_b64 exec, s[34:35]
; GFX908-NEXT: v_readlane_b32 s16, v40, 22
; GFX908-NEXT: s_mov_b32 s12, s24
-; GFX908-NEXT: s_mov_b32 s13, s21
-; GFX908-NEXT: s_mov_b32 s14, s23
+; GFX908-NEXT: s_mov_b32 s13, s23
+; GFX908-NEXT: s_mov_b32 s14, s22
; GFX908-NEXT: v_mov_b32_e32 v31, v32
-; GFX908-NEXT: s_mov_b32 s15, s22
+; GFX908-NEXT: s_mov_b32 s15, s21
; GFX908-NEXT: s_mov_b64 s[10:11], s[26:27]
; GFX908-NEXT: v_readlane_b32 s17, v40, 23
; GFX908-NEXT: buffer_store_dword v32, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
More information about the llvm-commits
mailing list