[llvm] AMDGPU: Allocate special SGPRs before user SGPR arguments (PR #78234)
via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 16 00:03:12 PST 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Matt Arsenault (arsenm)
<details>
<summary>Changes</summary>
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
---
Patch is 90.73 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/78234.diff
7 Files Affected:
- (modified) llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp (+5-6)
- (modified) llvm/lib/Target/AMDGPU/SIISelLowering.cpp (+9-10)
- (modified) llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pow-codegen.ll (+4-4)
- (added) llvm/test/CodeGen/AMDGPU/call-args-inreg.ll (+1583)
- (modified) llvm/test/CodeGen/AMDGPU/function-args-inreg.ll (+200)
- (modified) llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll (+1-1)
- (modified) llvm/test/CodeGen/AMDGPU/preserve-wwm-copy-dst-reg.ll (+24-24)
``````````diff
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...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/78234
More information about the llvm-commits
mailing list