clang/test/CodeGen/PowerPC/ppc64le-varargs-f128.c

   1 // RUN: %clang_cc1 -triple powerpc64le-unknown-linux-gnu -emit-llvm \
   2 // RUN:   -target-cpu pwr9 -target-feature +float128 -mabi=ieeelongdouble \
   3 // RUN:   -o - %s | FileCheck %s -check-prefix=IEEE
   4 // RUN: %clang_cc1 -triple powerpc64le-unknown-linux-gnu -emit-llvm \
   5 // RUN:   -target-cpu pwr9 -target-feature +float128 \
   6 // RUN:   -o - %s | FileCheck %s -check-prefix=IBM
   7
   8 // RUN: %clang_cc1 -triple ppc64le -emit-llvm-bc %s -target-cpu pwr9 \
   9 // RUN:   -target-feature +float128 -mabi=ieeelongdouble -fopenmp \
  10 // RUN:   -fopenmp-targets=ppc64le -o %t-ppc-host.bc
  11 // RUN: %clang_cc1 -triple ppc64le -aux-triple ppc64le %s -target-cpu pwr9 \
  12 // RUN:   -target-feature +float128 -fopenmp -fopenmp-is-device -emit-llvm \
  13 // RUN:   -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s \
  14 // RUN:   -check-prefix=OMP-TARGET
  15 // RUN: %clang_cc1 -triple ppc64le %t-ppc-host.bc -emit-llvm -o - | FileCheck %s \
  16 // RUN:   -check-prefix=OMP-HOST
  17
  18 #include <stdarg.h>
  19
  20 typedef struct { long double x; } ldbl128_s;
  21
  22 void foo_ld(long double);
  23 void foo_fq(__float128);
  24 void foo_ls(ldbl128_s);
  25
  26 // Verify cases when OpenMP target's and host's long-double semantics differ.
  27
  28 // OMP-TARGET-LABEL: define internal void @.omp_outlined.(
  29 // OMP-TARGET: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr
  30 // OMP-TARGET: %[[V3:[0-9a-zA-Z_.]+]] = load ppc_fp128, ptr %[[CUR]], align 8
  31 // OMP-TARGET: call void @foo_ld(ppc_fp128 noundef %[[V3]])
  32
  33 // OMP-HOST-LABEL: define{{.*}} void @omp(
  34 // OMP-HOST: call void @llvm.va_start(ptr %[[AP:[0-9a-zA-Z_.]+]])
  35 // OMP-HOST: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr %[[AP]], align 8
  36 // OMP-HOST: %[[V0:[0-9a-zA-Z_.]+]] = ptrtoint ptr %[[CUR]] to i64
  37 // OMP-HOST: %[[V1:[0-9a-zA-Z_.]+]] = add i64 %[[V0]], 15
  38 // OMP-HOST: %[[V2:[0-9a-zA-Z_.]+]] = and i64 %[[V1]], -16
  39 // OMP-HOST: %[[ALIGN:[0-9a-zA-Z_.]+]] = inttoptr i64 %[[V2]] to ptr
  40 // OMP-HOST: %[[V4:[0-9a-zA-Z_.]+]] = load fp128, ptr %[[ALIGN]], align 16
  41 // OMP-HOST: call void @foo_ld(fp128 noundef %[[V4]])
  42 void omp(int n, ...) {
  43   va_list ap;
  44   va_start(ap, n);
  45   foo_ld(va_arg(ap, long double));
  46   #pragma omp target parallel
  47   for (int i = 1; i < n; ++i) {
  48     foo_ld(va_arg(ap, long double));
  49   }
  50   va_end(ap);
  51 }
  52
  53 // IEEE-LABEL: define{{.*}} void @f128
  54 // IEEE: call void @llvm.va_start(ptr %[[AP:[0-9a-zA-Z_.]+]])
  55 // IEEE: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr %[[AP]]
  56 // IEEE: %[[V0:[0-9a-zA-Z_.]+]] = ptrtoint ptr %[[CUR]] to i64
  57 // IEEE: %[[V1:[0-9a-zA-Z_.]+]] = add i64 %[[V0]], 15
  58 // IEEE: %[[V2:[0-9a-zA-Z_.]+]] = and i64 %[[V1]], -16
  59 // IEEE: %[[ALIGN:[0-9a-zA-Z_.]+]] = inttoptr i64 %[[V2]] to ptr
  60 // IEEE: %[[V4:[0-9a-zA-Z_.]+]] = load fp128, ptr %[[ALIGN]], align 16
  61 // IEEE: call void @foo_fq(fp128 noundef %[[V4]])
  62 // IEEE: call void @llvm.va_end(ptr %[[AP]])
  63 void f128(int n, ...) {
  64   va_list ap;
  65   va_start(ap, n);
  66   foo_fq(va_arg(ap, __float128));
  67   va_end(ap);
  68 }
  69
  70 // IEEE-LABEL: define{{.*}} void @long_double
  71 // IEEE: call void @llvm.va_start(ptr %[[AP:[0-9a-zA-Z_.]+]])
  72 // IEEE: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr %[[AP]]
  73 // IEEE: %[[V0:[0-9a-zA-Z_.]+]] = ptrtoint ptr %[[CUR]] to i64
  74 // IEEE: %[[V1:[0-9a-zA-Z_.]+]] = add i64 %[[V0]], 15
  75 // IEEE: %[[V2:[0-9a-zA-Z_.]+]] = and i64 %[[V1]], -16
  76 // IEEE: %[[ALIGN:[0-9a-zA-Z_.]+]] = inttoptr i64 %[[V2]] to ptr
  77 // IEEE: %[[V4:[0-9a-zA-Z_.]+]] = load fp128, ptr %[[ALIGN]], align 16
  78 // IEEE: call void @foo_ld(fp128 noundef %[[V4]])
  79 // IEEE: call void @llvm.va_end(ptr %[[AP]])
  80
  81 // IBM-LABEL: define{{.*}} void @long_double
  82 // IBM: call void @llvm.va_start(ptr  %[[AP:[0-9a-zA-Z_.]+]])
  83 // IBM: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr %[[AP]]
  84 // IBM: %[[V4:[0-9a-zA-Z_.]+]] = load ppc_fp128, ptr %[[CUR]], align 8
  85 // IBM: call void @foo_ld(ppc_fp128 noundef %[[V4]])
  86 // IBM: call void @llvm.va_end(ptr %[[AP]])
  87 void long_double(int n, ...) {
  88   va_list ap;
  89   va_start(ap, n);
  90   foo_ld(va_arg(ap, long double));
  91   va_end(ap);
  92 }
  93
  94 // IEEE-LABEL: define{{.*}} void @long_double_struct
  95 // IEEE: call void @llvm.va_start(ptr %[[AP:[0-9a-zA-Z_.]+]])
  96 // IEEE: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr %[[AP]]
  97 // IEEE: %[[P0:[0-9a-zA-Z_.]+]] = ptrtoint ptr %[[CUR]] to i64
  98 // IEEE: %[[P1:[0-9a-zA-Z_.]+]] = add i64 %[[P0]], 15
  99 // IEEE: %[[P2:[0-9a-zA-Z_.]+]] = and i64 %[[P1]], -16
 100 // IEEE: %[[ALIGN:[0-9a-zA-Z_.]+]] = inttoptr i64 %[[P2]] to ptr
 101 // IEEE: %[[V0:[0-9a-zA-Z_.]+]] = getelementptr inbounds i8, ptr %[[ALIGN]], i64 16
 102 // IEEE: store ptr %[[V0]], ptr %[[AP]], align 8
 103 // IEEE: call void @llvm.memcpy.p0.p0.i64(ptr align 16 %[[TMP:[0-9a-zA-Z_.]+]], ptr align 16 %[[ALIGN]], i64 16, i1 false)
 104 // IEEE: %[[COERCE:[0-9a-zA-Z_.]+]] = getelementptr inbounds %struct.ldbl128_s, ptr %[[TMP]], i32 0, i32 0
 105 // IEEE: %[[V4:[0-9a-zA-Z_.]+]] = load fp128, ptr %[[COERCE]], align 16
 106 // IEEE: call void @foo_ls(fp128 inreg %[[V4]])
 107 // IEEE: call void @llvm.va_end(ptr %[[AP]])
 108 void long_double_struct(int n, ...) {
 109   va_list ap;
 110   va_start(ap, n);
 111   foo_ls(va_arg(ap, ldbl128_s));
 112   va_end(ap);
 113 }