clang/test/CodeGen/mips-varargs.c

   1 // RUN: %clang_cc1 -triple mips-unknown-linux -o - -emit-llvm %s | FileCheck %s -check-prefixes=ALL,O32 -enable-var-scope
   2 // RUN: %clang_cc1 -triple mipsel-unknown-linux -o - -emit-llvm %s | FileCheck %s -check-prefixes=ALL,O32 -enable-var-scope
   3 // RUN: %clang_cc1 -triple mips64-unknown-linux -o - -emit-llvm  -target-abi n32 %s | FileCheck %s -check-prefixes=ALL,N32,NEW -enable-var-scope
   4 // RUN: %clang_cc1 -triple mips64-unknown-linux -o - -emit-llvm  -target-abi n32 %s | FileCheck %s -check-prefixes=ALL,N32,NEW -enable-var-scope
   5 // RUN: %clang_cc1 -triple mips64-unknown-linux -o - -emit-llvm %s | FileCheck %s -check-prefixes=ALL,N64,NEW -enable-var-scope
   6 // RUN: %clang_cc1 -triple mips64el-unknown-linux -o - -emit-llvm %s | FileCheck %s -check-prefixes=ALL,N64,NEW -enable-var-scope
   7
   8 #include <stdarg.h>
   9
  10 typedef int v4i32 __attribute__ ((__vector_size__ (16)));
  11
  12 int test_i32(char *fmt, ...) {
  13   va_list va;
  14
  15   va_start(va, fmt);
  16   int v = va_arg(va, int);
  17   va_end(va);
  18
  19   return v;
  20 }
  21
  22 // O32-LABEL: define{{.*}} i32 @test_i32(ptr{{.*}} %fmt, ...)
  23 // N32-LABEL: define{{.*}} signext i32 @test_i32(ptr{{.*}} %fmt, ...)
  24 // N64-LABEL: define{{.*}} signext i32 @test_i32(ptr{{.*}} %fmt, ...)
  25 //
  26 // O32:   %va = alloca ptr, align [[$PTRALIGN:4]]
  27 // N32:   %va = alloca ptr, align [[$PTRALIGN:4]]
  28 // N64:   %va = alloca ptr, align [[$PTRALIGN:8]]
  29 // ALL:   [[V:%.*]] = alloca i32, align 4
  30 // NEW:   [[PROMOTION_TEMP:%.*]] = alloca i32, align 4
  31 //
  32 // ALL:   call void @llvm.va_start(ptr %va)
  33 // ALL:   [[AP_CUR:%.+]] = load ptr, ptr %va, align [[$PTRALIGN]]
  34 // O32:   [[AP_NEXT:%.+]] = getelementptr inbounds i8, ptr [[AP_CUR]], [[$INTPTR_T:i32]] [[$CHUNKSIZE:4]]
  35 // NEW:   [[AP_NEXT:%.+]] = getelementptr inbounds i8, ptr [[AP_CUR]], [[$INTPTR_T:i32|i64]] [[$CHUNKSIZE:8]]
  36 //
  37 // ALL:   store ptr [[AP_NEXT]], ptr %va, align [[$PTRALIGN]]
  38 //
  39 // O32:   [[ARG:%.+]] = load i32, ptr [[AP_CUR]], align [[CHUNKALIGN:4]]
  40 //
  41 // N32:   [[TMP:%.+]] = load i64, ptr [[AP_CUR]], align [[CHUNKALIGN:8]]
  42 // N64:   [[TMP:%.+]] = load i64, ptr [[AP_CUR]], align [[CHUNKALIGN:8]]
  43 // NEW:   [[TMP2:%.+]] = trunc i64 [[TMP]] to i32
  44 // NEW:   store i32 [[TMP2]], ptr [[PROMOTION_TEMP]], align 4
  45 // NEW:   [[ARG:%.+]] = load i32, ptr [[PROMOTION_TEMP]], align 4
  46 // ALL:   store i32 [[ARG]], ptr [[V]], align 4
  47 //
  48 // ALL:   call void @llvm.va_end(ptr %va)
  49 // ALL: }
  50
  51 long long test_i64(char *fmt, ...) {
  52   va_list va;
  53
  54   va_start(va, fmt);
  55   long long v = va_arg(va, long long);
  56   va_end(va);
  57
  58   return v;
  59 }
  60
  61 // ALL-LABEL: define{{.*}} i64 @test_i64(ptr{{.*}} %fmt, ...)
  62 //
  63 // ALL:   %va = alloca ptr, align [[$PTRALIGN]]
  64 // ALL:   call void @llvm.va_start(ptr %va)
  65 // ALL:   [[AP_CUR:%.+]] = load ptr, ptr %va, align [[$PTRALIGN]]
  66 //
  67 // i64 is 8-byte aligned, while this is within O32's stack alignment there's no
  68 // guarantee that the offset is still 8-byte aligned after earlier reads.
  69 // O32:   [[TMP1:%.+]] = getelementptr inbounds i8, ptr [[AP_CUR]], i32 7
  70 // O32:   [[AP_CUR:%.+]] = call ptr @llvm.ptrmask.p0.i32(ptr [[TMP1]], i32 -8)
  71 //
  72 // ALL:   [[AP_NEXT:%.+]] = getelementptr inbounds i8, ptr [[AP_CUR]], [[$INTPTR_T]] 8
  73 // ALL:   store ptr [[AP_NEXT]], ptr %va, align [[$PTRALIGN]]
  74 //
  75 // ALL:   [[ARG:%.+]] = load i64, ptr [[AP_CUR]], align 8
  76 //
  77 // ALL:   call void @llvm.va_end(ptr %va)
  78 // ALL: }
  79
  80 char *test_ptr(char *fmt, ...) {
  81   va_list va;
  82
  83   va_start(va, fmt);
  84   char *v = va_arg(va, char *);
  85   va_end(va);
  86
  87   return v;
  88 }
  89
  90 // ALL-LABEL: define{{.*}} ptr @test_ptr(ptr{{.*}} %fmt, ...)
  91 //
  92 // ALL:   %va = alloca ptr, align [[$PTRALIGN]]
  93 // ALL:   [[V:%.*]] = alloca ptr, align [[$PTRALIGN]]
  94 // N32:   [[AP_CAST:%.+]] = alloca ptr, align 4
  95 // ALL:   call void @llvm.va_start(ptr %va)
  96 // ALL:   [[AP_CUR:%.+]] = load ptr, ptr %va, align [[$PTRALIGN]]
  97 // ALL:   [[AP_NEXT:%.+]] = getelementptr inbounds i8, ptr [[AP_CUR]], [[$INTPTR_T]] [[$CHUNKSIZE]]
  98 // ALL:   store ptr [[AP_NEXT]], ptr %va, align [[$PTRALIGN]]
  99 //
 100 // When the chunk size matches the pointer size, this is easy.
 101 // Otherwise we need a promotion temporary.
 102 // N32:   [[TMP2:%.+]] = load i64, ptr [[AP_CUR]], align 8
 103 // N32:   [[TMP3:%.+]] = trunc i64 [[TMP2]] to i32
 104 // N32:   [[PTR:%.+]] = inttoptr i32 [[TMP3]] to ptr
 105 // N32:   store ptr [[PTR]], ptr [[AP_CAST]], align 4
 106 // N32:   [[ARG:%.+]] = load ptr, ptr [[AP_CAST]], align [[$PTRALIGN]]
 107 //
 108 // O32:   [[ARG:%.+]] = load ptr, ptr [[AP_CUR]], align [[$PTRALIGN]]
 109 // N64:   [[ARG:%.+]] = load ptr, ptr [[AP_CUR]], align [[$PTRALIGN]]
 110 // ALL:   store ptr [[ARG]], ptr [[V]], align [[$PTRALIGN]]
 111 //
 112 // ALL:   call void @llvm.va_end(ptr %va)
 113 // ALL: }
 114
 115 int test_v4i32(char *fmt, ...) {
 116   va_list va;
 117
 118   va_start(va, fmt);
 119   v4i32 v = va_arg(va, v4i32);
 120   va_end(va);
 121
 122   return v[0];
 123 }
 124
 125 // O32-LABEL: define{{.*}} i32 @test_v4i32(ptr{{.*}} %fmt, ...)
 126 // N32-LABEL: define{{.*}} signext i32 @test_v4i32(ptr{{.*}} %fmt, ...)
 127 // N64-LABEL: define{{.*}} signext i32 @test_v4i32(ptr{{.*}} %fmt, ...)
 128 //
 129 // ALL:   %va = alloca ptr, align [[$PTRALIGN]]
 130 // ALL:   [[V:%.+]] = alloca <4 x i32>, align 16
 131 // ALL:   call void @llvm.va_start(ptr %va)
 132 // ALL:   [[AP_CUR:%.+]] = load ptr, ptr %va, align [[$PTRALIGN]]
 133 //
 134 // Vectors are 16-byte aligned, however the O32 ABI has a maximum alignment of
 135 // 8-bytes since the base of the stack is 8-byte aligned.
 136
 137 // O32:   [[TMP1:%.+]] = getelementptr inbounds i8, ptr [[AP_CUR]], i32 7
 138 // O32:   [[AP_CUR:%.+]] = call ptr @llvm.ptrmask.p0.i32(ptr [[TMP1]], i32 -8)
 139
 140 // N32:   [[TMP1:%.+]] = getelementptr inbounds i8, ptr [[AP_CUR]], i32 15
 141 // N32:   [[AP_CUR:%.+]] = call ptr @llvm.ptrmask.p0.i32(ptr [[TMP1]], i32 -16)
 142
 143 // N64:   [[TMP1:%.+]] = getelementptr inbounds i8, ptr [[AP_CUR]], i32 15
 144 // N64:   [[AP_CUR:%.+]] = call ptr @llvm.ptrmask.p0.i64(ptr [[TMP1]], i64 -16)
 145
 146 //
 147 // ALL:   [[AP_NEXT:%.+]] = getelementptr inbounds i8, ptr [[AP_CUR]], [[$INTPTR_T]] 16
 148 // ALL:   store ptr [[AP_NEXT]], ptr %va, align [[$PTRALIGN]]
 149 //
 150 // O32:   [[ARG:%.+]] = load <4 x i32>, ptr [[AP_CUR]], align 8
 151 // N64:   [[ARG:%.+]] = load <4 x i32>, ptr [[AP_CUR]], align 16
 152 // N32:   [[ARG:%.+]] = load <4 x i32>, ptr [[AP_CUR]], align 16
 153 // ALL:   store <4 x i32> [[ARG]], ptr [[V]], align 16
 154 //
 155 // ALL:   call void @llvm.va_end(ptr %va)
 156 // ALL:   [[VECEXT:%.+]] = extractelement <4 x i32> {{.*}}, i32 0
 157 // ALL:   ret i32 [[VECEXT]]
 158 // ALL: }