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-target-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_offloading_{{.+}}_omp_{{.+}}.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: %[[TMP0:[^ ]+]] = getelementptr inbounds i8, ptr %[[CUR]], i32 15
  37 // OMP-HOST: %[[ALIGN:[^ ]+]] = call ptr @llvm.ptrmask.p0.i64(ptr %[[TMP0]], i64 -16)
  38 // OMP-HOST: %[[V4:[0-9a-zA-Z_.]+]] = load fp128, ptr %[[ALIGN]], align 16
  39 // OMP-HOST: call void @foo_ld(fp128 noundef %[[V4]])
  40 void omp(int n, ...) {
  41   va_list ap;
  42   va_start(ap, n);
  43   foo_ld(va_arg(ap, long double));
  44   #pragma omp target parallel
  45   for (int i = 1; i < n; ++i) {
  46     foo_ld(va_arg(ap, long double));
  47   }
  48   va_end(ap);
  49 }
  50
  51 // IEEE-LABEL: define{{.*}} void @f128
  52 // IEEE: call void @llvm.va_start(ptr %[[AP:[0-9a-zA-Z_.]+]])
  53 // IEEE: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr %[[AP]]
  54 // IEEE: %[[TMP0:[^ ]+]] = getelementptr inbounds i8, ptr %[[CUR]], i32 15
  55 // IEEE: %[[ALIGN:[^ ]+]] = call ptr @llvm.ptrmask.p0.i64(ptr %[[TMP0]], i64 -16)
  56 // IEEE: %[[V4:[0-9a-zA-Z_.]+]] = load fp128, ptr %[[ALIGN]], align 16
  57 // IEEE: call void @foo_fq(fp128 noundef %[[V4]])
  58 // IEEE: call void @llvm.va_end(ptr %[[AP]])
  59 void f128(int n, ...) {
  60   va_list ap;
  61   va_start(ap, n);
  62   foo_fq(va_arg(ap, __float128));
  63   va_end(ap);
  64 }
  65
  66 // IEEE-LABEL: define{{.*}} void @long_double
  67 // IEEE: call void @llvm.va_start(ptr %[[AP:[0-9a-zA-Z_.]+]])
  68 // IEEE: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr %[[AP]]
  69 // IEEE: %[[TMP0:[^ ]+]] = getelementptr inbounds i8, ptr %[[CUR]], i32 15
  70 // IEEE: %[[ALIGN:[^ ]+]] = call ptr @llvm.ptrmask.p0.i64(ptr %[[TMP0]], i64 -16)
  71 // IEEE: %[[V4:[0-9a-zA-Z_.]+]] = load fp128, ptr %[[ALIGN]], align 16
  72 // IEEE: call void @foo_ld(fp128 noundef %[[V4]])
  73 // IEEE: call void @llvm.va_end(ptr %[[AP]])
  74
  75 // IBM-LABEL: define{{.*}} void @long_double
  76 // IBM: call void @llvm.va_start(ptr  %[[AP:[0-9a-zA-Z_.]+]])
  77 // IBM: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr %[[AP]]
  78 // IBM: %[[V4:[0-9a-zA-Z_.]+]] = load ppc_fp128, ptr %[[CUR]], align 8
  79 // IBM: call void @foo_ld(ppc_fp128 noundef %[[V4]])
  80 // IBM: call void @llvm.va_end(ptr %[[AP]])
  81 void long_double(int n, ...) {
  82   va_list ap;
  83   va_start(ap, n);
  84   foo_ld(va_arg(ap, long double));
  85   va_end(ap);
  86 }
  87
  88 // IEEE-LABEL: define{{.*}} void @long_double_struct
  89 // IEEE: call void @llvm.va_start(ptr %[[AP:[0-9a-zA-Z_.]+]])
  90 // IEEE: %[[CUR:[0-9a-zA-Z_.]+]] = load ptr, ptr %[[AP]]
  91 // IEEE: %[[TMP0:[^ ]+]] = getelementptr inbounds i8, ptr %[[CUR]], i32 15
  92 // IEEE: %[[ALIGN:[^ ]+]] = call ptr @llvm.ptrmask.p0.i64(ptr %[[TMP0]], i64 -16)
  93 // IEEE: %[[V0:[0-9a-zA-Z_.]+]] = getelementptr inbounds i8, ptr %[[ALIGN]], i64 16
  94 // IEEE: store ptr %[[V0]], ptr %[[AP]], align 8
  95 // IEEE: call void @llvm.memcpy.p0.p0.i64(ptr align 16 %[[TMP:[0-9a-zA-Z_.]+]], ptr align 16 %[[ALIGN]], i64 16, i1 false)
  96 // IEEE: %[[COERCE:[0-9a-zA-Z_.]+]] = getelementptr inbounds %struct.ldbl128_s, ptr %[[TMP]], i32 0, i32 0
  97 // IEEE: %[[V4:[0-9a-zA-Z_.]+]] = load fp128, ptr %[[COERCE]], align 16
  98 // IEEE: call void @foo_ls(fp128 inreg %[[V4]])
  99 // IEEE: call void @llvm.va_end(ptr %[[AP]])
 100 void long_double_struct(int n, ...) {
 101   va_list ap;
 102   va_start(ap, n);
 103   foo_ls(va_arg(ap, ldbl128_s));
 104   va_end(ap);
 105 }