[llvm] [AMDGPU] Illegal VGPR to SGPR copy (PR #144008)
Akhilesh Moorthy via llvm-commits
llvm-commits at lists.llvm.org
Thu Jun 12 19:45:40 PDT 2025
https://github.com/isakhilesh created https://github.com/llvm/llvm-project/pull/144008
This patch resolves an instance of an illegal VGPR to SGPR copy by invoking `BuildMI()` when the source register is a VGPR and a destination register is a SGPR, and since we cannot copy data directly from a VGPR to a SGPR, we use `AMDGPU::V_READFIRSTLANE_B32`.
Fixes SWDEV-530052.
>From 5485f24fa95714d54b5b3d266b9ed2b7c24a4143 Mon Sep 17 00:00:00 2001
From: Akhilesh Moorthy <akhilesh.moorthy at amd.com>
Date: Mon, 9 Jun 2025 14:22:31 -0500
Subject: [PATCH] [AMDGPU] Illegal VGPR to SGPR copy
This patch resolves an instance of an illegal VGPR to SGPR copy by
invoking `BuildMI()` when the source register is a VGPR and a
destination register is a SGPR, and since we cannot copy data directly
from a VGPR to a SGPR, we use `AMDGPU::V_READFIRSTLANE_B32`.
Fixes SWDEV-530052.
---
llvm/lib/Target/AMDGPU/SIInstrInfo.cpp | 9 +-
...> call-args-inreg-no-sgpr-for-csrspill.ll} | 14 ++-
.../AMDGPU/illegal-sgpr-to-vgpr-copy.ll | 10 +-
...ev503538-move-to-valu-stack-srd-physreg.ll | 103 +++++++++++++++++-
...530052-illegal-vgpr-to-sgpr-copy-single.ll | 38 +++++++
...530052-illegal-vgpr-to-sgpr-copy-vector.ll | 42 +++++++
.../AMDGPU/tail-call-inreg-arguments.error.ll | 81 +++-----------
7 files changed, 218 insertions(+), 79 deletions(-)
rename llvm/test/CodeGen/AMDGPU/{call-args-inreg-no-sgpr-for-csrspill-xfail.ll => call-args-inreg-no-sgpr-for-csrspill.ll} (72%)
create mode 100644 llvm/test/CodeGen/AMDGPU/swdev530052-illegal-vgpr-to-sgpr-copy-single.ll
create mode 100644 llvm/test/CodeGen/AMDGPU/swdev530052-illegal-vgpr-to-sgpr-copy-vector.ll
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index 85276bd24bcf4..00f48af5d9e22 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -875,7 +875,14 @@ void SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
}
if (!AMDGPU::SReg_32RegClass.contains(SrcReg)) {
- reportIllegalCopy(this, MBB, MI, DL, DestReg, SrcReg, KillSrc);
+ // We invoke BuildMI() only when we have verified that the source register
+ // is a VGPR and the destination register is a SGPR, and since we cannot
+ // transfer data directly from VGPR to SGPR, we use
+ // AMDGPU::V_READFIRSTLANE_B32
+ assert(AMDGPU::SReg_32RegClass.contains(DestReg));
+ assert(AMDGPU::VGPR_32RegClass.contains(SrcReg));
+ BuildMI(MBB, MI, DL, this->get(AMDGPU::V_READFIRSTLANE_B32), DestReg)
+ .addReg(SrcReg);
return;
}
diff --git a/llvm/test/CodeGen/AMDGPU/call-args-inreg-no-sgpr-for-csrspill-xfail.ll b/llvm/test/CodeGen/AMDGPU/call-args-inreg-no-sgpr-for-csrspill.ll
similarity index 72%
rename from llvm/test/CodeGen/AMDGPU/call-args-inreg-no-sgpr-for-csrspill-xfail.ll
rename to llvm/test/CodeGen/AMDGPU/call-args-inreg-no-sgpr-for-csrspill.ll
index 34f4476f7fd6a..4b22b51cc7df2 100644
--- a/llvm/test/CodeGen/AMDGPU/call-args-inreg-no-sgpr-for-csrspill-xfail.ll
+++ b/llvm/test/CodeGen/AMDGPU/call-args-inreg-no-sgpr-for-csrspill.ll
@@ -1,6 +1,4 @@
-; RUN: not llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs=0 -filetype=null %s 2>&1 | FileCheck -enable-var-scope %s
-
-; CHECK: illegal VGPR to SGPR copy
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs=0 < %s | FileCheck -enable-var-scope %s
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
@@ -25,3 +23,13 @@ attributes #0 = { nounwind }
!llvm.module.flags = !{!0}
!0 = !{i32 1, !"amdhsa_code_object_version", i32 400}
+; CHECK: v_readlane_b32
+; CHECK: s_mov_b32
+; CHECK: v_writelane_b32
+; CHECK: s_swappc_b64
+; CHECK: s_or_saveexec_b64
+; CHECK: buffer_load_dword
+; CHECK: s_waitcnt
+; CHECK: s_addk_i32
+; CHECK: v_readfirstlane_b32
+; CHECK: s_mov_b64
\ No newline at end of file
diff --git a/llvm/test/CodeGen/AMDGPU/illegal-sgpr-to-vgpr-copy.ll b/llvm/test/CodeGen/AMDGPU/illegal-sgpr-to-vgpr-copy.ll
index 597f90c0f4e84..28be32fabab3c 100644
--- a/llvm/test/CodeGen/AMDGPU/illegal-sgpr-to-vgpr-copy.ll
+++ b/llvm/test/CodeGen/AMDGPU/illegal-sgpr-to-vgpr-copy.ll
@@ -1,8 +1,6 @@
; RUN: not llc -mtriple=amdgcn -verify-machineinstrs=0 < %s 2>&1 | FileCheck -check-prefix=ERR %s
; RUN: not llc -mtriple=amdgcn -verify-machineinstrs=0 < %s 2>&1 | FileCheck -check-prefix=GCN %s
-; ERR: error: <unknown>:0:0: in function illegal_vgpr_to_sgpr_copy_i32 void (): illegal VGPR to SGPR copy
-; GCN: ; illegal copy v1 to s9
define amdgpu_kernel void @illegal_vgpr_to_sgpr_copy_i32() #0 {
%vgpr = call i32 asm sideeffect "; def $0", "=${v1}"()
@@ -42,9 +40,7 @@ define amdgpu_kernel void @illegal_vgpr_to_sgpr_copy_v16i32() #0 {
ret void
}
-; ERR: error: <unknown>:0:0: in function illegal_agpr_to_sgpr_copy_i32 void (): illegal VGPR to SGPR copy
-; GCN: v_accvgpr_read_b32 [[COPY1:v[0-9]+]], a1
-; GCN: ; illegal copy [[COPY1]] to s9
+
define amdgpu_kernel void @illegal_agpr_to_sgpr_copy_i32() #1 {
%agpr = call i32 asm sideeffect "; def $0", "=${a1}"()
call void asm sideeffect "; use $0", "${s9}"(i32 %agpr)
@@ -54,7 +50,7 @@ define amdgpu_kernel void @illegal_agpr_to_sgpr_copy_i32() #1 {
; ERR: error: <unknown>:0:0: in function illegal_agpr_to_sgpr_copy_v2i32 void (): illegal VGPR to SGPR copy
; GCN-DAG: v_accvgpr_read_b32 v[[COPY1L:[0-9]+]], a0
; GCN-DAG: v_accvgpr_read_b32 v[[COPY1H:[0-9]+]], a1
-; GCN: ; illegal copy v[[[COPY1L]]:[[COPY1H]]] to s[10:11]
+; GCN: ; illegal copy v[0:1] to s[10:11]
define amdgpu_kernel void @illegal_agpr_to_sgpr_copy_v2i32() #1 {
%vgpr = call <2 x i32> asm sideeffect "; def $0", "=${a[0:1]}"()
call void asm sideeffect "; use $0", "${s[10:11]}"(<2 x i32> %vgpr)
@@ -62,4 +58,4 @@ define amdgpu_kernel void @illegal_agpr_to_sgpr_copy_v2i32() #1 {
}
attributes #0 = { nounwind }
-attributes #1 = { nounwind "target-cpu"="gfx908" }
+attributes #1 = { nounwind "target-cpu"="gfx908" }
\ No newline at end of file
diff --git a/llvm/test/CodeGen/AMDGPU/swdev503538-move-to-valu-stack-srd-physreg.ll b/llvm/test/CodeGen/AMDGPU/swdev503538-move-to-valu-stack-srd-physreg.ll
index f0b3d334af67d..820d35de9e1e0 100644
--- a/llvm/test/CodeGen/AMDGPU/swdev503538-move-to-valu-stack-srd-physreg.ll
+++ b/llvm/test/CodeGen/AMDGPU/swdev503538-move-to-valu-stack-srd-physreg.ll
@@ -1,14 +1,107 @@
-; RUN: not llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -verify-machineinstrs=0 -O0 2> %t.err < %s | FileCheck %s
-; RUN: FileCheck -check-prefix=ERR %s < %t.err
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -verify-machineinstrs=0 -O0 2> %t.err < %s | FileCheck %s
; FIXME: This error will be fixed by supporting arbitrary divergent
; dynamic allocas by performing a wave umax of the size.
-; ERR: error: <unknown>:0:0: in function move_to_valu_assert_srd_is_physreg_swdev503538 i32 (ptr addrspace(1)): illegal VGPR to SGPR copy
-
-; CHECK: ; illegal copy v0 to s32
define i32 @move_to_valu_assert_srd_is_physreg_swdev503538(ptr addrspace(1) %ptr) {
+; CHECK-LABEL: move_to_valu_assert_srd_is_physreg_swdev503538:
+; CHECK: ; %bb.0: ; %entry
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: s_mov_b32 s7, s33
+; CHECK-NEXT: s_mov_b32 s33, s32
+; CHECK-NEXT: s_xor_saveexec_b64 s[4:5], -1
+; CHECK-NEXT: buffer_store_dword v3, off, s[0:3], s33 ; 4-byte Folded Spill
+; CHECK-NEXT: s_mov_b64 exec, s[4:5]
+; CHECK-NEXT: s_add_i32 s32, s32, 0x400
+; CHECK-NEXT: v_mov_b32_e32 v2, v1
+; CHECK-NEXT: ; implicit-def: $sgpr4
+; CHECK-NEXT: ; implicit-def: $sgpr4
+; CHECK-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec
+; CHECK-NEXT: ; kill: def $vgpr1 killed $vgpr2 killed $exec
+; CHECK-NEXT: ; implicit-def: $sgpr4_sgpr5
+; CHECK-NEXT: v_mov_b32_e32 v0, s32
+; CHECK-NEXT: v_accvgpr_write_b32 a0, v0 ; Reload Reuse
+; CHECK-NEXT: v_readfirstlane_b32 s32, v0
+; CHECK-NEXT: v_accvgpr_write_b32 a1, v0 ; Reload Reuse
+; CHECK-NEXT: ; implicit-def: $sgpr4
+; CHECK-NEXT: s_mov_b64 s[4:5], exec
+; CHECK-NEXT: ; implicit-def: $vgpr3 : SGPR spill to VGPR lane
+; CHECK-NEXT: v_writelane_b32 v3, s4, 0
+; CHECK-NEXT: v_writelane_b32 v3, s5, 1
+; CHECK-NEXT: s_or_saveexec_b64 s[10:11], -1
+; CHECK-NEXT: v_accvgpr_write_b32 a2, v3 ; Reload Reuse
+; CHECK-NEXT: s_mov_b64 exec, s[10:11]
+; CHECK-NEXT: .LBB0_1: ; =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: s_or_saveexec_b64 s[10:11], -1
+; CHECK-NEXT: v_accvgpr_read_b32 v3, a2 ; Reload Reuse
+; CHECK-NEXT: s_mov_b64 exec, s[10:11]
+; CHECK-NEXT: v_accvgpr_read_b32 v0, a0 ; Reload Reuse
+; CHECK-NEXT: v_readfirstlane_b32 s4, v0
+; CHECK-NEXT: v_writelane_b32 v3, s4, 2
+; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], s4, v0
+; CHECK-NEXT: s_and_saveexec_b64 s[4:5], s[4:5]
+; CHECK-NEXT: v_writelane_b32 v3, s4, 3
+; CHECK-NEXT: v_writelane_b32 v3, s5, 4
+; CHECK-NEXT: s_or_saveexec_b64 s[10:11], -1
+; CHECK-NEXT: v_accvgpr_write_b32 a2, v3 ; Reload Reuse
+; CHECK-NEXT: s_mov_b64 exec, s[10:11]
+; CHECK-NEXT: ; %bb.2: ; in Loop: Header=BB0_1 Depth=1
+; CHECK-NEXT: s_or_saveexec_b64 s[10:11], -1
+; CHECK-NEXT: v_accvgpr_read_b32 v3, a2 ; Reload Reuse
+; CHECK-NEXT: s_mov_b64 exec, s[10:11]
+; CHECK-NEXT: v_readlane_b32 s4, v3, 3
+; CHECK-NEXT: v_readlane_b32 s5, v3, 4
+; CHECK-NEXT: v_readlane_b32 s6, v3, 2
+; CHECK-NEXT: s_nop 4
+; CHECK-NEXT: buffer_load_dword v0, off, s[0:3], s6
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: v_accvgpr_write_b32 a3, v0 ; Reload Reuse
+; CHECK-NEXT: s_xor_b64 exec, exec, s[4:5]
+; CHECK-NEXT: s_cbranch_execnz .LBB0_1
+; CHECK-NEXT: ; %bb.3:
+; CHECK-NEXT: s_or_saveexec_b64 s[10:11], -1
+; CHECK-NEXT: v_accvgpr_read_b32 v3, a2 ; Reload Reuse
+; CHECK-NEXT: s_mov_b64 exec, s[10:11]
+; CHECK-NEXT: v_readlane_b32 s4, v3, 0
+; CHECK-NEXT: v_readlane_b32 s5, v3, 1
+; CHECK-NEXT: s_mov_b64 exec, s[4:5]
+; CHECK-NEXT: s_mov_b32 s4, 0
+; CHECK-NEXT: v_writelane_b32 v3, s4, 5
+; CHECK-NEXT: s_or_saveexec_b64 s[10:11], -1
+; CHECK-NEXT: v_accvgpr_write_b32 a2, v3 ; Reload Reuse
+; CHECK-NEXT: s_mov_b64 exec, s[10:11]
+; CHECK-NEXT: .LBB0_4: ; %loadstoreloop
+; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: s_or_saveexec_b64 s[10:11], -1
+; CHECK-NEXT: v_accvgpr_read_b32 v3, a2 ; Reload Reuse
+; CHECK-NEXT: s_mov_b64 exec, s[10:11]
+; CHECK-NEXT: v_readlane_b32 s4, v3, 5
+; CHECK-NEXT: v_accvgpr_read_b32 v0, a1 ; Reload Reuse
+; CHECK-NEXT: v_add_u32_e64 v1, v0, s4
+; CHECK-NEXT: v_mov_b32_e32 v0, 0
+; CHECK-NEXT: buffer_store_byte v0, v1, s[0:3], 0 offen
+; CHECK-NEXT: s_mov_b32 s5, 1
+; CHECK-NEXT: s_add_i32 s4, s4, s5
+; CHECK-NEXT: s_mov_b32 s5, 0x800
+; CHECK-NEXT: s_cmp_lt_u32 s4, s5
+; CHECK-NEXT: v_writelane_b32 v3, s4, 5
+; CHECK-NEXT: s_mov_b64 s[10:11], exec
+; CHECK-NEXT: s_mov_b64 exec, -1
+; CHECK-NEXT: v_accvgpr_write_b32 a2, v3 ; Reload Reuse
+; CHECK-NEXT: s_mov_b64 exec, s[10:11]
+; CHECK-NEXT: s_cbranch_scc1 .LBB0_4
+; CHECK-NEXT: ; %bb.5: ; %Flow
+; CHECK-NEXT: ; %bb.6: ; %split
+; CHECK-NEXT: v_accvgpr_read_b32 v0, a3 ; Reload Reuse
+; CHECK-NEXT: s_mov_b32 s32, s33
+; CHECK-NEXT: s_xor_saveexec_b64 s[4:5], -1
+; CHECK-NEXT: buffer_load_dword v3, off, s[0:3], s33 ; 4-byte Folded Reload
+; CHECK-NEXT: s_mov_b64 exec, s[4:5]
+; CHECK-NEXT: s_mov_b32 s33, s7
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: s_setpc_b64 s[30:31]
entry:
%idx = load i32, ptr addrspace(1) %ptr, align 4
%zero = extractelement <4 x i32> zeroinitializer, i32 %idx
diff --git a/llvm/test/CodeGen/AMDGPU/swdev530052-illegal-vgpr-to-sgpr-copy-single.ll b/llvm/test/CodeGen/AMDGPU/swdev530052-illegal-vgpr-to-sgpr-copy-single.ll
new file mode 100644
index 0000000000000..0d898bd5cd8c3
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/swdev530052-illegal-vgpr-to-sgpr-copy-single.ll
@@ -0,0 +1,38 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 < %s | FileCheck %s
+
+declare void @test_buffer_load_sgpr_plus_imm_offset_noflags(i32 inreg)
+
+define void @test_load_zext(i32 %foo) {
+; CHECK-LABEL: test_load_zext:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: s_mov_b32 s0, s33
+; CHECK-NEXT: s_mov_b32 s33, s32
+; CHECK-NEXT: s_or_saveexec_b64 s[2:3], -1
+; CHECK-NEXT: scratch_store_dword off, v40, s33 ; 4-byte Folded Spill
+; CHECK-NEXT: s_mov_b64 exec, s[2:3]
+; CHECK-NEXT: s_add_i32 s32, s32, 16
+; CHECK-NEXT: v_writelane_b32 v40, s0, 2
+; CHECK-NEXT: s_getpc_b64 s[0:1]
+; CHECK-NEXT: s_add_u32 s0, s0, test_buffer_load_sgpr_plus_imm_offset_noflags at gotpcrel32@lo+4
+; CHECK-NEXT: s_addc_u32 s1, s1, test_buffer_load_sgpr_plus_imm_offset_noflags at gotpcrel32@hi+12
+; CHECK-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x0
+; CHECK-NEXT: v_writelane_b32 v40, s30, 0
+; CHECK-NEXT: v_readfirstlane_b32 s0, v0
+; CHECK-NEXT: v_writelane_b32 v40, s31, 1
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: s_swappc_b64 s[30:31], s[2:3]
+; CHECK-NEXT: v_readlane_b32 s31, v40, 1
+; CHECK-NEXT: v_readlane_b32 s30, v40, 0
+; CHECK-NEXT: s_mov_b32 s32, s33
+; CHECK-NEXT: v_readlane_b32 s0, v40, 2
+; CHECK-NEXT: s_or_saveexec_b64 s[2:3], -1
+; CHECK-NEXT: scratch_load_dword v40, off, s33 ; 4-byte Folded Reload
+; CHECK-NEXT: s_mov_b64 exec, s[2:3]
+; CHECK-NEXT: s_mov_b32 s33, s0
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+ call void @test_buffer_load_sgpr_plus_imm_offset_noflags(i32 %foo)
+ ret void
+ }
diff --git a/llvm/test/CodeGen/AMDGPU/swdev530052-illegal-vgpr-to-sgpr-copy-vector.ll b/llvm/test/CodeGen/AMDGPU/swdev530052-illegal-vgpr-to-sgpr-copy-vector.ll
new file mode 100644
index 0000000000000..63ecba2d56451
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/swdev530052-illegal-vgpr-to-sgpr-copy-vector.ll
@@ -0,0 +1,42 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 < %s | FileCheck %s
+
+define void @test_load_zext(<4 x i32> %LGV) {
+; CHECK-LABEL: test_load_zext:
+; CHECK: ; %bb.0: ; %.entry
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: s_mov_b32 s0, s33
+; CHECK-NEXT: s_mov_b32 s33, s32
+; CHECK-NEXT: s_or_saveexec_b64 s[2:3], -1
+; CHECK-NEXT: scratch_store_dword off, v40, s33 ; 4-byte Folded Spill
+; CHECK-NEXT: s_mov_b64 exec, s[2:3]
+; CHECK-NEXT: s_add_i32 s32, s32, 16
+; CHECK-NEXT: v_writelane_b32 v40, s0, 2
+; CHECK-NEXT: s_getpc_b64 s[0:1]
+; CHECK-NEXT: s_add_u32 s0, s0, test_buffer_load_sgpr_plus_imm_offset_noflags at gotpcrel32@lo+4
+; CHECK-NEXT: s_addc_u32 s1, s1, test_buffer_load_sgpr_plus_imm_offset_noflags at gotpcrel32@hi+12
+; CHECK-NEXT: s_load_dwordx2 s[16:17], s[0:1], 0x0
+; CHECK-NEXT: v_writelane_b32 v40, s30, 0
+; CHECK-NEXT: v_readfirstlane_b32 s0, v0
+; CHECK-NEXT: v_readfirstlane_b32 s1, v1
+; CHECK-NEXT: v_readfirstlane_b32 s2, v2
+; CHECK-NEXT: v_readfirstlane_b32 s3, v3
+; CHECK-NEXT: v_writelane_b32 v40, s31, 1
+; CHECK-NEXT: s_waitcnt lgkmcnt(0)
+; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
+; CHECK-NEXT: v_readlane_b32 s31, v40, 1
+; CHECK-NEXT: v_readlane_b32 s30, v40, 0
+; CHECK-NEXT: s_mov_b32 s32, s33
+; CHECK-NEXT: v_readlane_b32 s0, v40, 2
+; CHECK-NEXT: s_or_saveexec_b64 s[2:3], -1
+; CHECK-NEXT: scratch_load_dword v40, off, s33 ; 4-byte Folded Reload
+; CHECK-NEXT: s_mov_b64 exec, s[2:3]
+; CHECK-NEXT: s_mov_b32 s33, s0
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+.entry:
+ call void @test_buffer_load_sgpr_plus_imm_offset_noflags(<4 x i32> %LGV)
+ ret void
+}
+
+declare void @test_buffer_load_sgpr_plus_imm_offset_noflags(<4 x i32> inreg)
diff --git a/llvm/test/CodeGen/AMDGPU/tail-call-inreg-arguments.error.ll b/llvm/test/CodeGen/AMDGPU/tail-call-inreg-arguments.error.ll
index 242b5e9aeaf42..cf3e61e58bdcc 100644
--- a/llvm/test/CodeGen/AMDGPU/tail-call-inreg-arguments.error.ll
+++ b/llvm/test/CodeGen/AMDGPU/tail-call-inreg-arguments.error.ll
@@ -1,41 +1,8 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: not llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs=0 2> %t.err < %s | FileCheck %s
-; RUN: FileCheck -check-prefix=ERR %s < %t.err
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs=0 < %s | FileCheck %s
; FIXME: These tests cannot be tail called, and should be executed in a waterfall loop.
declare hidden void @void_func_i32_inreg(i32 inreg)
-
-; ERR: error: <unknown>:0:0: in function tail_call_i32_inreg_divergent void (i32): illegal VGPR to SGPR copy
-; ERR: error: <unknown>:0:0: in function indirect_tail_call_i32_inreg_divergent void (i32): illegal VGPR to SGPR copy
-
define void @tail_call_i32_inreg_divergent(i32 %vgpr) {
-; CHECK-LABEL: tail_call_i32_inreg_divergent:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_mov_b32 s16, s33
-; CHECK-NEXT: s_mov_b32 s33, s32
-; CHECK-NEXT: s_or_saveexec_b64 s[18:19], -1
-; CHECK-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
-; CHECK-NEXT: s_mov_b64 exec, s[18:19]
-; CHECK-NEXT: v_writelane_b32 v40, s16, 2
-; CHECK-NEXT: s_addk_i32 s32, 0x400
-; CHECK-NEXT: v_writelane_b32 v40, s30, 0
-; CHECK-NEXT: v_writelane_b32 v40, s31, 1
-; CHECK-NEXT: s_getpc_b64 s[16:17]
-; CHECK-NEXT: s_add_u32 s16, s16, void_func_i32_inreg at rel32@lo+4
-; CHECK-NEXT: s_addc_u32 s17, s17, void_func_i32_inreg at rel32@hi+12
-; CHECK-NEXT: ; illegal copy v0 to s0
-; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
-; CHECK-NEXT: v_readlane_b32 s31, v40, 1
-; CHECK-NEXT: v_readlane_b32 s30, v40, 0
-; CHECK-NEXT: s_mov_b32 s32, s33
-; CHECK-NEXT: v_readlane_b32 s4, v40, 2
-; CHECK-NEXT: s_or_saveexec_b64 s[6:7], -1
-; CHECK-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
-; CHECK-NEXT: s_mov_b64 exec, s[6:7]
-; CHECK-NEXT: s_mov_b32 s33, s4
-; CHECK-NEXT: s_waitcnt vmcnt(0)
-; CHECK-NEXT: s_setpc_b64 s[30:31]
tail call void @void_func_i32_inreg(i32 inreg %vgpr)
ret void
}
@@ -43,36 +10,24 @@ define void @tail_call_i32_inreg_divergent(i32 %vgpr) {
@constant = external hidden addrspace(4) constant ptr
define void @indirect_tail_call_i32_inreg_divergent(i32 %vgpr) {
-; CHECK-LABEL: indirect_tail_call_i32_inreg_divergent:
-; CHECK: ; %bb.0:
-; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_mov_b32 s16, s33
-; CHECK-NEXT: s_mov_b32 s33, s32
-; CHECK-NEXT: s_or_saveexec_b64 s[18:19], -1
-; CHECK-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
-; CHECK-NEXT: s_mov_b64 exec, s[18:19]
-; CHECK-NEXT: s_addk_i32 s32, 0x400
-; CHECK-NEXT: v_writelane_b32 v40, s16, 2
-; CHECK-NEXT: s_getpc_b64 s[16:17]
-; CHECK-NEXT: s_add_u32 s16, s16, constant at rel32@lo+4
-; CHECK-NEXT: s_addc_u32 s17, s17, constant at rel32@hi+12
-; CHECK-NEXT: s_load_dwordx2 s[16:17], s[16:17], 0x0
-; CHECK-NEXT: v_writelane_b32 v40, s30, 0
-; CHECK-NEXT: v_writelane_b32 v40, s31, 1
-; CHECK-NEXT: ; illegal copy v0 to s0
-; CHECK-NEXT: s_waitcnt lgkmcnt(0)
-; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
-; CHECK-NEXT: v_readlane_b32 s31, v40, 1
-; CHECK-NEXT: v_readlane_b32 s30, v40, 0
-; CHECK-NEXT: s_mov_b32 s32, s33
-; CHECK-NEXT: v_readlane_b32 s4, v40, 2
-; CHECK-NEXT: s_or_saveexec_b64 s[6:7], -1
-; CHECK-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
-; CHECK-NEXT: s_mov_b64 exec, s[6:7]
-; CHECK-NEXT: s_mov_b32 s33, s4
-; CHECK-NEXT: s_waitcnt vmcnt(0)
-; CHECK-NEXT: s_setpc_b64 s[30:31]
%fptr = load ptr, ptr addrspace(4) @constant, align 8
tail call void %fptr(i32 inreg %vgpr)
ret void
}
+;CHECK: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
+;CHECK: s_mov_b64 exec, s[18:19]
+;CHECK: v_writelane_b32 v40, s16, 2
+;CHECK: s_addk_i32 s32, 0x400
+;CHECK: v_writelane_b32 v40, s30, 0
+;CHECK: s_getpc_b64 s[16:17]
+;CHECK: s_add_u32 s16, s16, void_func_i32_inreg at rel32@lo+4
+;CHECK: s_addc_u32 s17, s17, void_func_i32_inreg at rel32@hi+12
+;CHECK: v_readfirstlane_b32 s0, v0
+;CHECK: v_writelane_b32 v40, s31, 1
+;CHECK: s_swappc_b64 s[30:31], s[16:17]
+;CHECK: v_readlane_b32 s31, v40, 1
+;CHECK: v_readlane_b32 s30, v40, 0
+;CHECK: s_mov_b32 s32, s33
+;CHECK: v_readlane_b32 s4, v40, 2
+;CHECK: s_or_saveexec_b64 s[6:7], -1
+;CHECK: buffer_load_dword v40, off, s[0:3], s33
\ No newline at end of file
More information about the llvm-commits
mailing list