Skip to content

Commit 71acfe8

Browse files
JonChesterfieldJonChesterfield
committed
[AMDGPU] Implement variadic functions by IR lowering
1 parent ae4f300 commit 71acfe8

26 files changed

+4572
-22
lines changed

‎clang/lib/CodeGen/Targets/AMDGPU.cpp‎

Lines changed: 5 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -120,7 +120,11 @@ void AMDGPUABIInfo::computeInfo(CGFunctionInfo &FI) const {
120120

121121
Address AMDGPUABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
122122
QualType Ty) const {
123-
llvm_unreachable("AMDGPU does not support varargs");
123+
const bool IsIndirect = false;
124+
const bool AllowHigherAlign = false;
125+
return emitVoidPtrVAArg(CGF, VAListAddr, Ty, IsIndirect,
126+
getContext().getTypeInfoInChars(Ty),
127+
CharUnits::fromQuantity(4), AllowHigherAlign);
124128
}
125129

126130
ABIArgInfo AMDGPUABIInfo::classifyReturnType(QualType RetTy) const {

‎clang/test/CodeGen/voidptr-vaarg.c‎

Lines changed: 478 additions & 0 deletions
Large diffs are not rendered by default.

‎clang/test/CodeGenCXX/inline-then-fold-variadics.cpp‎

Lines changed: 181 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,181 @@
1+
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature
2+
// REQUIRES: webassembly-registered-target
3+
4+
// Simple calls to known variadic functions that are completely elided when
5+
// optimisations are on This is a functional check that the expand-variadic pass
6+
// is consistent with clang's va_arg handling
7+
8+
// When expand-variadics is added to the default pipeline, clang -O1 will
9+
// suffice here -Wno-varargs avoids warning second argument to 'va_start' is not
10+
// the last named parameter
11+
12+
// RUN: %clang_cc1 %s -triple wasm32-unknown-unknown -Wno-varargs -O1 -emit-llvm -o - | opt - -S --passes='module(expand-variadics,default<O1>)' --expand-variadics-override=optimize -o - | FileCheck %s
13+
14+
#include <stdarg.h>
15+
#include <stdint.h>
16+
17+
template <typename X, typename Y> static X first(...) {
18+
va_list va;
19+
__builtin_va_start(va, 0);
20+
X r = va_arg(va, X);
21+
va_end(va);
22+
return r;
23+
}
24+
25+
template <typename X, typename Y> static Y second(...) {
26+
va_list va;
27+
__builtin_va_start(va, 0);
28+
va_arg(va, X);
29+
Y r = va_arg(va, Y);
30+
va_end(va);
31+
return r;
32+
}
33+
34+
extern "C" {
35+
36+
// CHECK-LABEL: define {{[^@]+}}@first_pair_i32
37+
// CHECK-SAME: (i32 noundef returned [[X:%.*]], i32 noundef [[Y:%.*]])
38+
// CHECK-NEXT: entry:
39+
// CHECK-NEXT: ret i32 [[X]]
40+
//
41+
int first_pair_i32(int x, int y) { return first<int, int>(x, y); }
42+
43+
// CHECK-LABEL: define {{[^@]+}}@second_pair_i32
44+
// CHECK-SAME: (i32 noundef [[X:%.*]], i32 noundef returned [[Y:%.*]])
45+
// CHECK-NEXT: entry:
46+
// CHECK-NEXT: ret i32 [[Y]]
47+
//
48+
int second_pair_i32(int x, int y) { return second<int, int>(x, y); }
49+
50+
// CHECK-LABEL: define {{[^@]+}}@first_pair_f64
51+
// CHECK-SAME: (double noundef returned [[X:%.*]], double noundef [[Y:%.*]])
52+
// CHECK-NEXT: entry:
53+
// CHECK-NEXT: ret double [[X]]
54+
//
55+
double first_pair_f64(double x, double y) {
56+
return first<double, double>(x, y);
57+
}
58+
59+
// CHECK-LABEL: define {{[^@]+}}@second_pair_f64
60+
// CHECK-SAME: (double noundef [[X:%.*]], double noundef returned [[Y:%.*]])
61+
// CHECK-NEXT: entry:
62+
// CHECK-NEXT: ret double [[Y]]
63+
//
64+
double second_pair_f64(double x, double y) {
65+
return second<double, double>(x, y);
66+
}
67+
}
68+
69+
extern "C" {
70+
71+
// CHECK-LABEL: define {{[^@]+}}@first_i32_f64
72+
// CHECK-SAME: (i32 noundef returned [[X:%.*]], double noundef [[Y:%.*]])
73+
// CHECK-NEXT: entry:
74+
// CHECK-NEXT: ret i32 [[X]]
75+
//
76+
int first_i32_f64(int x, double y) { return first<int, double>(x, y); }
77+
78+
// CHECK-LABEL: define {{[^@]+}}@second_i32_f64
79+
// CHECK-SAME: (i32 noundef [[X:%.*]], double noundef returned [[Y:%.*]])
80+
// CHECK-NEXT: entry:
81+
// CHECK-NEXT: ret double [[Y]]
82+
//
83+
double second_i32_f64(int x, double y) { return second<int, double>(x, y); }
84+
85+
// CHECK-LABEL: define {{[^@]+}}@first_f64_i32
86+
// CHECK-SAME: (double noundef returned [[X:%.*]], i32 noundef [[Y:%.*]])
87+
// CHECK-NEXT: entry:
88+
// CHECK-NEXT: ret double [[X]]
89+
//
90+
double first_f64_i32(double x, int y) { return first<double, int>(x, y); }
91+
92+
// CHECK-LABEL: define {{[^@]+}}@second_f64_i32
93+
// CHECK-SAME: (double noundef [[X:%.*]], i32 noundef returned [[Y:%.*]])
94+
// CHECK-NEXT: entry:
95+
// CHECK-NEXT: ret i32 [[Y]]
96+
//
97+
int second_f64_i32(double x, int y) { return second<double, int>(x, y); }
98+
}
99+
100+
extern "C" {
101+
typedef uint64_t ulong2 __attribute__((__vector_size__(16), __aligned__(16)));
102+
103+
// CHECK-LABEL: define {{[^@]+}}@first_i32_ulong2
104+
// CHECK-SAME: (i32 noundef returned [[X:%.*]], ptr nocapture noundef readonly [[Y:%.*]])
105+
// CHECK-NEXT: entry:
106+
// CHECK-NEXT: ret i32 [[X]]
107+
//
108+
int first_i32_ulong2(int x, ulong2 *y) { return first<int, ulong2>(x, *y); }
109+
110+
// CHECK-LABEL: define {{[^@]+}}@second_i32_ulong2
111+
// CHECK-SAME: (i32 noundef [[X:%.*]], ptr nocapture noundef readonly [[Y:%.*]], ptr nocapture noundef writeonly [[R:%.*]]) local_unnamed_addr #[[ATTR1:[0-9]+]] {
112+
// CHECK-NEXT: entry:
113+
// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[Y]], align 16, !tbaa [[TBAA2:![0-9]+]]
114+
// CHECK-NEXT: store <2 x i64> [[TMP0]], ptr [[R]], align 16, !tbaa [[TBAA2]]
115+
// CHECK-NEXT: ret void
116+
//
117+
void second_i32_ulong2(int x, ulong2 *y, ulong2 *r) {
118+
*r = second<int, ulong2>(x, *y);
119+
}
120+
121+
// CHECK-LABEL: define {{[^@]+}}@first_ulong2_i32
122+
// CHECK-SAME: (ptr nocapture noundef readonly [[X:%.*]], i32 noundef [[Y:%.*]], ptr nocapture noundef writeonly [[R:%.*]]) local_unnamed_addr #[[ATTR1]] {
123+
// CHECK-NEXT: entry:
124+
// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[X]], align 16, !tbaa [[TBAA2]]
125+
// CHECK-NEXT: store <2 x i64> [[TMP0]], ptr [[R]], align 16, !tbaa [[TBAA2]]
126+
// CHECK-NEXT: ret void
127+
//
128+
void first_ulong2_i32(ulong2 *x, int y, ulong2 *r) {
129+
*r = first<ulong2, int>(*x, y);
130+
}
131+
132+
// CHECK-LABEL: define {{[^@]+}}@second_ulong2_i32
133+
// CHECK-SAME: (ptr nocapture noundef readonly [[X:%.*]], i32 noundef returned [[Y:%.*]])
134+
// CHECK-NEXT: entry:
135+
// CHECK-NEXT: ret i32 [[Y]]
136+
//
137+
int second_ulong2_i32(ulong2 *x, int y) { return second<ulong2, int>(*x, y); }
138+
}
139+
140+
// ascending alignment
141+
typedef struct {
142+
char c;
143+
short s;
144+
int i;
145+
long l;
146+
float f;
147+
double d;
148+
} asc;
149+
150+
extern "C" {
151+
152+
// CHECK-LABEL: define {{[^@]+}}@first_i32_asc
153+
// CHECK-SAME: (i32 noundef returned [[X:%.*]], ptr nocapture noundef readonly [[Y:%.*]])
154+
// CHECK-NEXT: entry:
155+
// CHECK-NEXT: ret i32 [[X]]
156+
//
157+
int first_i32_asc(int x, asc *y) { return first<int, asc>(x, *y); }
158+
159+
// CHECK-LABEL: define {{[^@]+}}@second_i32_asc
160+
// CHECK-SAME: (i32 noundef [[X:%.*]], ptr nocapture noundef readonly [[Y:%.*]], ptr nocapture noundef writeonly [[R:%.*]]) local_unnamed_addr #[[ATTR1]] {
161+
// CHECK-NEXT: entry:
162+
// CHECK-NEXT: tail call void @llvm.memmove.p0.p0.i32(ptr noundef nonnull align 8 dereferenceable(24) [[R]], ptr noundef nonnull align 1 dereferenceable(24) [[Y]], i32 24, i1 false)
163+
// CHECK-NEXT: ret void
164+
//
165+
void second_i32_asc(int x, asc *y, asc *r) { *r = second<int, asc>(x, *y); }
166+
167+
// CHECK-LABEL: define {{[^@]+}}@first_asc_i32
168+
// CHECK-SAME: (ptr nocapture noundef readonly [[X:%.*]], i32 noundef [[Y:%.*]], ptr nocapture noundef writeonly [[R:%.*]]) local_unnamed_addr #[[ATTR1]] {
169+
// CHECK-NEXT: entry:
170+
// CHECK-NEXT: tail call void @llvm.memmove.p0.p0.i32(ptr noundef nonnull align 8 dereferenceable(24) [[R]], ptr noundef nonnull align 1 dereferenceable(24) [[X]], i32 24, i1 false)
171+
// CHECK-NEXT: ret void
172+
//
173+
void first_asc_i32(asc *x, int y, asc *r) { *r = first<asc, int>(*x, y); }
174+
175+
// CHECK-LABEL: define {{[^@]+}}@second_asc_i32
176+
// CHECK-SAME: (ptr nocapture noundef readonly [[X:%.*]], i32 noundef returned [[Y:%.*]])
177+
// CHECK-NEXT: entry:
178+
// CHECK-NEXT: ret i32 [[Y]]
179+
//
180+
int second_asc_i32(asc *x, int y) { return second<asc, int>(*x, y); }
181+
}

‎libc/config/gpu/entrypoints.txt‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -181,6 +181,10 @@ set(TARGET_LIBC_ENTRYPOINTS
181181
libc.src.stdio.fflush
182182
libc.src.stdio.ftell
183183
libc.src.stdio.clearerr
184+
libc.src.stdio.sprintf
185+
libc.src.stdio.snprintf
186+
libc.src.stdio.vsprintf
187+
libc.src.stdio.vsnprintf
184188
libc.src.stdio.puts
185189
libc.src.stdio.fopen
186190
libc.src.stdio.fclose

‎libc/test/src/__support/CMakeLists.txt‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -86,8 +86,8 @@ add_libc_test(
8686
libc.src.__support.uint128
8787
)
8888

89-
# The GPU does not support varargs currently.
90-
if(NOT LIBC_TARGET_OS_IS_GPU)
89+
# NVPTX does not support varargs currently.
90+
if(NOT LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
9191
add_libc_test(
9292
arg_list_test
9393
SUITE

‎llvm/include/llvm/IR/InstrTypes.h‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2120,6 +2120,15 @@ class CallBase : public Instruction {
21202120
return Attrs.getParamStackAlignment(ArgNo);
21212121
}
21222122

2123+
/// Extract the byref type for a call or parameter.
2124+
Type *getParamByRefType(unsigned ArgNo) const {
2125+
if (auto *Ty = Attrs.getParamByRefType(ArgNo))
2126+
return Ty;
2127+
if (const Function *F = getCalledFunction())
2128+
return F->getAttributes().getParamByRefType(ArgNo);
2129+
return nullptr;
2130+
}
2131+
21232132
/// Extract the byval type for a call or parameter.
21242133
Type *getParamByValType(unsigned ArgNo) const {
21252134
if (auto *Ty = Attrs.getParamByValType(ArgNo))

‎llvm/include/llvm/InitializePasses.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -106,6 +106,7 @@ void initializeExpandLargeDivRemLegacyPassPass(PassRegistry&);
106106
void initializeExpandMemCmpLegacyPassPass(PassRegistry &);
107107
void initializeExpandPostRAPass(PassRegistry&);
108108
void initializeExpandReductionsPass(PassRegistry&);
109+
void initializeExpandVariadicsPass(PassRegistry &);
109110
void initializeExpandVectorPredicationPass(PassRegistry &);
110111
void initializeExternalAAWrapperPassPass(PassRegistry&);
111112
void initializeFEntryInserterPass(PassRegistry&);

‎llvm/include/llvm/Transforms/IPO/ExpandVariadics.h‎

Lines changed: 40 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,40 @@
1+
//===- ExpandVariadics.h - expand variadic functions ------------*- C++ -*-===//
2+
//
3+
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4+
// See https://llvm.org/LICENSE.txt for license information.
5+
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6+
//
7+
//===----------------------------------------------------------------------===//
8+
#ifndef LLVM_TRANSFORMS_IPO_EXPANDVARIADICS_H
9+
#define LLVM_TRANSFORMS_IPO_EXPANDVARIADICS_H
10+
11+
#include "llvm/IR/PassManager.h"
12+
13+
namespace llvm {
14+
15+
class Module;
16+
class ModulePass;
17+
class OptimizationLevel;
18+
19+
enum class ExpandVariadicsMode {
20+
Unspecified, // Use the implementation defaults
21+
Disable, // Disable the pass entirely
22+
Optimize, // Optimise without changing ABI
23+
Lowering, // Change variadic calling convention
24+
};
25+
26+
class ExpandVariadicsPass : public PassInfoMixin<ExpandVariadicsPass> {
27+
const ExpandVariadicsMode Mode;
28+
29+
public:
30+
// Operates under passed mode unless overridden on commandline
31+
ExpandVariadicsPass(ExpandVariadicsMode Mode);
32+
33+
PreservedAnalyses run(Module &M, ModuleAnalysisManager &AM);
34+
};
35+
36+
ModulePass *createExpandVariadicsPass(ExpandVariadicsMode);
37+
38+
} // end namespace llvm
39+
40+
#endif // LLVM_TRANSFORMS_IPO_EXPANDVARIADICS_H

‎llvm/lib/Passes/PassBuilder.cpp‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -138,6 +138,7 @@
138138
#include "llvm/Transforms/IPO/DeadArgumentElimination.h"
139139
#include "llvm/Transforms/IPO/ElimAvailExtern.h"
140140
#include "llvm/Transforms/IPO/EmbedBitcodePass.h"
141+
#include "llvm/Transforms/IPO/ExpandVariadics.h"
141142
#include "llvm/Transforms/IPO/ForceFunctionAttrs.h"
142143
#include "llvm/Transforms/IPO/FunctionAttrs.h"
143144
#include "llvm/Transforms/IPO/FunctionImport.h"

‎llvm/lib/Passes/PassRegistry.def‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -59,6 +59,7 @@ MODULE_PASS("dot-callgraph", CallGraphDOTPrinterPass())
5959
MODULE_PASS("dxil-upgrade", DXILUpgradePass())
6060
MODULE_PASS("elim-avail-extern", EliminateAvailableExternallyPass())
6161
MODULE_PASS("extract-blocks", BlockExtractorPass({}, false))
62+
MODULE_PASS("expand-variadics", ExpandVariadicsPass(ExpandVariadicsMode::Disable))
6263
MODULE_PASS("forceattrs", ForceFunctionAttrsPass())
6364
MODULE_PASS("function-import", FunctionImportPass())
6465
MODULE_PASS("globalopt", GlobalOptPass())

0 commit comments

Comments
 (0)