[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata.

Instead of generating llvm.mem.parallel_loop_access metadata, generate
llvm.access.group on instructions and llvm.loop.parallel_accesses on
loops. There is one access group per generated loop.

This is clang part of D52116/r349725.

Differential Revision: https://reviews.llvm.org/D52117

llvm-svn: 349823

1 files changed, 102 insertions, 102 deletions

diff --git a/clang/test/OpenMP/simd_codegen.cpp b/clang/test/OpenMP/simd_codegen.cpp
index 883273099ad..9ac71b1dbb6 100644
--- a/clang/test/OpenMP/simd_codegen.cpp
+++ b/clang/test/OpenMP/simd_codegen.cpp
@@ -21,23 +21,23 @@ void simple(float *a, float *b, float *c, float *d) {
   #pragma omp simd
 // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP1_ID:[0-9]+]]
+// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], 6
 // CHECK-NEXT: br i1 [[CMP]], label %[[SIMPLE_LOOP1_BODY:.+]], label %[[SIMPLE_LOOP1_END:[^,]+]]
   for (int i = 3; i < 32; i += 5) {
 // CHECK: [[SIMPLE_LOOP1_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP1_ID]]
+// CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK: [[CALC_I_1:%.+]] = mul nsw i32 [[IV1_1]], 5
 // CHECK-NEXT: [[CALC_I_2:%.+]] = add nsw i32 3, [[CALC_I_1]]
-// CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP1_ID]]
+// CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]{{.*}}!llvm.access.group
 // ... loop body ...
 // End of body: store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float* {{%.+}}{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP1_ID]]
+// CHECK: store float [[RESULT:%.+]], float* {{%.+}}{{.*}}!llvm.access.group
     a[i] = b[i] * c[i] * d[i];
-// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP1_ID]]
+// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1
-// CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP1_ID]]
+// CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // br label %{{.+}}, !llvm.loop !{{.+}}
   }
 // CHECK: [[SIMPLE_LOOP1_END]]:
@@ -51,20 +51,20 @@ void simple(float *a, float *b, float *c, float *d) {
 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_VAR]]
 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
 
-// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP2_ID:[0-9]+]]
+// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV2]], 9
 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP2_BODY:.+]], label %[[SIMPLE_LOOP2_END:[^,]+]]
   for (int i = 10; i > 1; i--) {
 // CHECK: [[SIMPLE_LOOP2_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV2_0:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP2_ID]]
+// CHECK: [[IV2_0:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 // CHECK-NEXT: [[IV2_1:%.+]] = mul nsw i32 [[IV2_0]], 1
 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV2_1]]
-// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.mem.parallel_loop_access ![[SIMPLE_LOOP2_ID]]
+// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
 //
-// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP2_ID]]
-// CHECK-NEXT: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP2_ID]]
+// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV2_2]], 3
 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
@@ -72,9 +72,9 @@ void simple(float *a, float *b, float *c, float *d) {
 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
     a[k]++;
     k = k + 3;
-// CHECK: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP2_ID]]
+// CHECK: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV2_2]], 1
-// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV2]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP2_ID]]
+// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV2]]{{.*}}!llvm.access.group
 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP2_ID]]
   }
 // CHECK: [[SIMPLE_LOOP2_END]]:
@@ -101,35 +101,35 @@ void simple(float *a, float *b, float *c, float *d) {
 // CHECK: [[GLIN_LOAD:%.+]] = load double*, double** [[GLIN_VAR:@[^,]+]]
 // CHECK-NEXT: store double* [[GLIN_LOAD]], double** [[GLIN_START:%[^,]+]]
 
-// CHECK: [[IV3:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID:[0-9]+]]
+// CHECK: [[IV3:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP3:%.+]] = icmp ult i64 [[IV3]], 4
 // CHECK-NEXT: br i1 [[CMP3]], label %[[SIMPLE_LOOP3_BODY:.+]], label %[[SIMPLE_LOOP3_END:[^,]+]]
   for (unsigned long long it = 2000; it >= 600; it-=400) {
 // CHECK: [[SIMPLE_LOOP3_BODY]]:
 // Start of body: calculate it from IV:
-// CHECK: [[IV3_0:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
+// CHECK: [[IV3_0:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul i64 [[IV3_0]], 400
 // CHECK-NEXT: [[LC_IT_2:%.+]] = sub i64 2000, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* {{.+}}, !llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
+// CHECK-NEXT: store i64 [[LC_IT_2]], i64* {{.+}}, !llvm.access.group
 //
 // Linear start and step are used to calculate current value of the linear variable.
-// CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
-// CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
-// CHECK-NOT: store i32 {{.+}}, i32* [[LIN_VAR]],{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
-// CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
-// CHECK-NEXT: [[IV3_1:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
+// CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]]{{.*}}!llvm.access.group
+// CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]]{{.*}}!llvm.access.group
+// CHECK-NOT: store i32 {{.+}}, i32* [[LIN_VAR]],{{.*}}!llvm.access.group
+// CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV3_1:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[MUL:%.+]] = mul i64 [[IV3_1]], 1
 // CHECK: [[GEP:%.+]] = getelementptr{{.*}}[[GLINSTART]]
-// CHECK-NEXT: store double* [[GEP]], double** [[G_PTR_CUR:%[^,]+]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
+// CHECK-NEXT: store double* [[GEP]], double** [[G_PTR_CUR:%[^,]+]]{{.*}}!llvm.access.group
     *g_ptr++ = 0.0;
-// CHECK: [[GEP_VAL:%.+]] = load double{{.*}}[[G_PTR_CUR]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
-// CHECK: store double{{.*}}[[GEP_VAL]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
+// CHECK: [[GEP_VAL:%.+]] = load double{{.*}}[[G_PTR_CUR]]{{.*}}!llvm.access.group
+// CHECK: store double{{.*}}[[GEP_VAL]]{{.*}}!llvm.access.group
     a[it + lin]++;
 // CHECK: [[FLT_INC:%.+]] = fadd float
-// CHECK-NEXT: store float [[FLT_INC]],{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
-// CHECK: [[IV3_2:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
+// CHECK-NEXT: store float [[FLT_INC]],{{.*}}!llvm.access.group
+// CHECK: [[IV3_2:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD3_2:%.+]] = add i64 [[IV3_2]], 1
-// CHECK-NEXT: store i64 [[ADD3_2]], i64* [[OMP_IV3]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP3_ID]]
+// CHECK-NEXT: store i64 [[ADD3_2]], i64* [[OMP_IV3]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP3_END]]:
 //
@@ -143,42 +143,42 @@ void simple(float *a, float *b, float *c, float *d) {
   #pragma omp simd
 // CHECK: store i32 0, i32* [[OMP_IV4:%[^,]+]]
 
-// CHECK: [[IV4:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP4_ID:[0-9]+]]
+// CHECK: [[IV4:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP4:%.+]] = icmp slt i32 [[IV4]], 4
 // CHECK-NEXT: br i1 [[CMP4]], label %[[SIMPLE_LOOP4_BODY:.+]], label %[[SIMPLE_LOOP4_END:[^,]+]]
   for (short it = 6; it <= 20; it-=-4) {
 // CHECK: [[SIMPLE_LOOP4_BODY]]:
 // Start of body: calculate it from IV:
-// CHECK: [[IV4_0:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP4_ID]]
+// CHECK: [[IV4_0:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i32 [[IV4_0]], 4
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i32 6, [[LC_IT_1]]
 // CHECK-NEXT: [[LC_IT_3:%.+]] = trunc i32 [[LC_IT_2]] to i16
-// CHECK-NEXT: store i16 [[LC_IT_3]], i16* {{.+}}, !llvm.mem.parallel_loop_access ![[SIMPLE_LOOP4_ID]]
+// CHECK-NEXT: store i16 [[LC_IT_3]], i16* {{.+}}, !llvm.access.group
 
-// CHECK: [[IV4_2:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP4_ID]]
+// CHECK: [[IV4_2:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD4_2:%.+]] = add nsw i32 [[IV4_2]], 1
-// CHECK-NEXT: store i32 [[ADD4_2]], i32* [[OMP_IV4]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP4_ID]]
+// CHECK-NEXT: store i32 [[ADD4_2]], i32* [[OMP_IV4]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP4_END]]:
 
   #pragma omp simd
 // CHECK: store i32 0, i32* [[OMP_IV5:%[^,]+]]
 
-// CHECK: [[IV5:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP5_ID:[0-9]+]]
+// CHECK: [[IV5:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP5:%.+]] = icmp slt i32 [[IV5]], 26
 // CHECK-NEXT: br i1 [[CMP5]], label %[[SIMPLE_LOOP5_BODY:.+]], label %[[SIMPLE_LOOP5_END:[^,]+]]
   for (unsigned char it = 'z'; it >= 'a'; it+=-1) {
 // CHECK: [[SIMPLE_LOOP5_BODY]]:
 // Start of body: calculate it from IV:
-// CHECK: [[IV5_0:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP5_ID]]
+// CHECK: [[IV5_0:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[IV5_1:%.+]] = mul nsw i32 [[IV5_0]], 1
 // CHECK-NEXT: [[LC_IT_1:%.+]] = sub nsw i32 122, [[IV5_1]]
 // CHECK-NEXT: [[LC_IT_2:%.+]] = trunc i32 [[LC_IT_1]] to i8
-// CHECK-NEXT: store i8 [[LC_IT_2]], i8* {{.+}}, !llvm.mem.parallel_loop_access ![[SIMPLE_LOOP5_ID]]
+// CHECK-NEXT: store i8 [[LC_IT_2]], i8* {{.+}}, !llvm.access.group
 
-// CHECK: [[IV5_2:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP5_ID]]
+// CHECK: [[IV5_2:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD5_2:%.+]] = add nsw i32 [[IV5_2]], 1
-// CHECK-NEXT: store i32 [[ADD5_2]], i32* [[OMP_IV5]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP5_ID]]
+// CHECK-NEXT: store i32 [[ADD5_2]], i32* [[OMP_IV5]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP5_END]]:
 
@@ -194,23 +194,23 @@ void simple(float *a, float *b, float *c, float *d) {
 // CHECK: store i64 0, i64* [[OMP_IV7:%[^,]+]]
 // CHECK: br label %[[SIMD_LOOP7_COND:[^,]+]]
 // CHECK: [[SIMD_LOOP7_COND]]:
-// CHECK-NEXT: [[IV7:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID:[0-9]+]]
+// CHECK-NEXT: [[IV7:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP7:%.+]] = icmp slt i64 [[IV7]], 7
 // CHECK-NEXT: br i1 [[CMP7]], label %[[SIMPLE_LOOP7_BODY:.+]], label %[[SIMPLE_LOOP7_END:[^,]+]]
   for (long long i = -10; i < 10; i += 3) {
 // CHECK: [[SIMPLE_LOOP7_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV7_0:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
+// CHECK: [[IV7_0:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV7_0]], 3
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
-// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
+// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group
+// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[CONV:%.+]] = trunc i64 [[LC_VAL]] to i32
-// CHECK-NEXT: store i32 [[CONV]], i32* [[A_PRIV:%[^,]+]],{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
+// CHECK-NEXT: store i32 [[CONV]], i32* [[A_PRIV:%[^,]+]],{{.+}}!llvm.access.group
     A = i;
-// CHECK: [[IV7_2:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
+// CHECK: [[IV7_2:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD7_2:%.+]] = add nsw i64 [[IV7_2]], 1
-// CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
+// CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP7_END]]:
 // CHECK-NEXT: store i64 11, i64*
@@ -224,22 +224,22 @@ void simple(float *a, float *b, float *c, float *d) {
   #pragma omp simd reduction(*:R)
 // CHECK: br label %[[SIMD_LOOP8_COND:[^,]+]]
 // CHECK: [[SIMD_LOOP8_COND]]:
-// CHECK-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID:[0-9]+]]
+// CHECK-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7
 // CHECK-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
   for (long long i = -10; i < 10; i += 3) {
 // CHECK: [[SIMPLE_LOOP8_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+// CHECK: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
-// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
-// CHECK: store i32 %{{.+}}, i32* [[R_PRIV]],{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group
+// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group
+// CHECK: store i32 %{{.+}}, i32* [[R_PRIV]],{{.+}}!llvm.access.group
     R *= i;
-// CHECK: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+// CHECK: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
-// CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP8_ID]]
+// CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]]{{.*}}!llvm.access.group
   }
 // CHECK: [[SIMPLE_LOOP8_END]]:
 // CHECK-DAG: [[R_VAL:%.+]] = load i32, i32* [[R]],
@@ -266,26 +266,26 @@ int templ1(T a, T *z) {
 // CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float {{.+}}, float* {{.+}})
 // CHECK: store i64 0, i64* [[T1_OMP_IV:[^,]+]]
 // ...
-// CHECK: [[IV:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[T1_ID:[0-9]+]]
+// CHECK: [[IV:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP1:%.+]] = icmp slt i64 [[IV]], 16
 // CHECK-NEXT: br i1 [[CMP1]], label %[[T1_BODY:.+]], label %[[T1_END:[^,]+]]
 // CHECK: [[T1_BODY]]:
 // Loop counters i and j updates:
-// CHECK: [[IV1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[T1_ID]]
+// CHECK: [[IV1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[I_1:%.+]] = sdiv i64 [[IV1]], 4
 // CHECK-NEXT: [[I_1_MUL1:%.+]] = mul nsw i64 [[I_1]], 1
 // CHECK-NEXT: [[I_1_ADD0:%.+]] = add nsw i64 0, [[I_1_MUL1]]
 // CHECK-NEXT: [[I_2:%.+]] = trunc i64 [[I_1_ADD0]] to i32
-// CHECK-NEXT: store i32 [[I_2]], i32* {{%.+}}{{.*}}!llvm.mem.parallel_loop_access ![[T1_ID]]
-// CHECK: [[IV2:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[T1_ID]]
+// CHECK-NEXT: store i32 [[I_2]], i32* {{%.+}}{{.*}}!llvm.access.group
+// CHECK: [[IV2:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[J_1:%.+]] = srem i64 [[IV2]], 4
 // CHECK-NEXT: [[J_2:%.+]] = mul nsw i64 [[J_1]], 2
 // CHECK-NEXT: [[J_2_ADD0:%.+]] = add nsw i64 0, [[J_2]]
-// CHECK-NEXT: store i64 [[J_2_ADD0]], i64* {{%.+}}{{.*}}!llvm.mem.parallel_loop_access ![[T1_ID]]
+// CHECK-NEXT: store i64 [[J_2_ADD0]], i64* {{%.+}}{{.*}}!llvm.access.group
 // simd.for.inc:
-// CHECK: [[IV3:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[T1_ID]]
+// CHECK: [[IV3:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[INC:%.+]] = add nsw i64 [[IV3]], 1
-// CHECK-NEXT: store i64 [[INC]], i64* [[T1_OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[T1_ID]]
+// CHECK-NEXT: store i64 [[INC]], i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: br label {{%.+}}
 // CHECK: [[T1_END]]:
 // CHECK: ret i32 0
@@ -338,15 +338,15 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) {
 // CHECK: store i32 0, i32* [[IT_OMP_IV:%[^,]+]]
   #pragma omp simd
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}} !llvm.mem.parallel_loop_access ![[ITER_LOOP_ID:[0-9]+]]
-// CHECK-NEXT: [[LAST_IT:%.+]] = load i32, i32* [[OMP_LAST_IT]]{{.+}}!llvm.mem.parallel_loop_access ![[ITER_LOOP_ID]]
+// CHECK: [[IV:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}} !llvm.access.group
+// CHECK-NEXT: [[LAST_IT:%.+]] = load i32, i32* [[OMP_LAST_IT]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[NUM_IT:%.+]] = add nsw i32 [[LAST_IT]], 1
 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], [[NUM_IT]]
 // CHECK-NEXT: br i1 [[CMP]], label %[[IT_BODY:[^,]+]], label %[[IT_END:[^,]+]]
   for (IterDouble i = ia; i < ib; ++i) {
 // CHECK: [[IT_BODY]]:
 // Start of body: calculate i from index:
-// CHECK: [[IV1:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[ITER_LOOP_ID]]
+// CHECK: [[IV1:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
 // Call of operator+ (i, IV).
 // CHECK: {{%.+}} = invoke {{.+}} @{{.*}}IterDouble{{.*}}
 // ... loop body ...
@@ -354,12 +354,12 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) {
 // Float multiply and save result.
 // CHECK: [[MULR:%.+]] = fmul double {{%.+}}, 5.000000e-01
 // CHECK-NEXT: invoke {{.+}} @{{.*}}IterDouble{{.*}}
-// CHECK: store double [[MULR:%.+]], double* [[RESULT_ADDR:%.+]], !llvm.mem.parallel_loop_access ![[ITER_LOOP_ID]]
+// CHECK: store double [[MULR:%.+]], double* [[RESULT_ADDR:%.+]], !llvm.access.group
    ++ic;
 //
-// CHECK: [[IV2:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[ITER_LOOP_ID]]
+// CHECK: [[IV2:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i32 [[IV2]], 1
-// CHECK-NEXT: store i32 [[ADD2]], i32* [[IT_OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[ITER_LOOP_ID]]
+// CHECK-NEXT: store i32 [[ADD2]], i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
 // br label %{{.*}}, !llvm.loop ![[ITER_LOOP_ID]]
   }
 // CHECK: [[IT_END]]:
@@ -377,7 +377,7 @@ void collapsed(float *a, float *b, float *c, float *d) {
 //
   #pragma omp simd collapse(4)
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[COLL1_LOOP_ID:[0-9]+]]
+// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[CMP:%.+]] = icmp ult i32 [[IV]], 120
 // CHECK-NEXT: br i1 [[CMP]], label %[[COLL1_BODY:[^,]+]], label %[[COLL1_END:[^,]+]]
   for (i = 1; i < 3; i++) // 2 iterations
@@ -387,25 +387,25 @@ void collapsed(float *a, float *b, float *c, float *d) {
         {
 // CHECK: [[COLL1_BODY]]:
 // Start of body: calculate i from index:
-// CHECK: [[IV1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[COLL1_LOOP_ID]]
+// CHECK: [[IV1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
 // Calculation of the loop counters values.
 // CHECK: [[CALC_I_1:%.+]] = udiv i32 [[IV1]], 60
 // CHECK-NEXT: [[CALC_I_1_MUL1:%.+]] = mul i32 [[CALC_I_1]], 1
 // CHECK-NEXT: [[CALC_I_2:%.+]] = add i32 1, [[CALC_I_1_MUL1]]
 // CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]
-// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[COLL1_LOOP_ID]]
+// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[CALC_J_1:%.+]] = udiv i32 [[IV1_2]], 20
 // CHECK-NEXT: [[CALC_J_2:%.+]] = urem i32 [[CALC_J_1]], 3
 // CHECK-NEXT: [[CALC_J_2_MUL1:%.+]] = mul i32 [[CALC_J_2]], 1
 // CHECK-NEXT: [[CALC_J_3:%.+]] = add i32 2, [[CALC_J_2_MUL1]]
 // CHECK-NEXT: store i32 [[CALC_J_3]], i32* [[LC_J:.+]]
-// CHECK: [[IV1_3:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[COLL1_LOOP_ID]]
+// CHECK: [[IV1_3:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[CALC_K_1:%.+]] = udiv i32 [[IV1_3]], 5
 // CHECK-NEXT: [[CALC_K_2:%.+]] = urem i32 [[CALC_K_1]], 4
 // CHECK-NEXT: [[CALC_K_2_MUL1:%.+]] = mul i32 [[CALC_K_2]], 1
 // CHECK-NEXT: [[CALC_K_3:%.+]] = add i32 3, [[CALC_K_2_MUL1]]
 // CHECK-NEXT: store i32 [[CALC_K_3]], i32* [[LC_K:.+]]
-// CHECK: [[IV1_4:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[COLL1_LOOP_ID]]
+// CHECK: [[IV1_4:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[CALC_L_1:%.+]] = urem i32 [[IV1_4]], 5
 // CHECK-NEXT: [[CALC_L_1_MUL1:%.+]] = mul i32 [[CALC_L_1]], 1
 // CHECK-NEXT: [[CALC_L_2:%.+]] = add i32 4, [[CALC_L_1_MUL1]]
@@ -413,12 +413,12 @@ void collapsed(float *a, float *b, float *c, float *d) {
 // CHECK-NEXT: store i16 [[CALC_L_3]], i16* [[LC_L:.+]]
 // ... loop body ...
 // End of body: store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.mem.parallel_loop_access ![[COLL1_LOOP_ID]]
+// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
     float res = b[j] * c[k];
     a[i] = res * d[l];
-// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[COLL1_LOOP_ID]]
+// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2:%.+]] = add i32 [[IV2]], 1
-// CHECK-NEXT: store i32 [[ADD2]], i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[COLL1_LOOP_ID]]
+// CHECK-NEXT: store i32 [[ADD2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // br label %{{[^,]+}}, !llvm.loop ![[COLL1_LOOP_ID]]
 // CHECK: [[COLL1_END]]:
   }
@@ -445,8 +445,8 @@ void widened(float *a, float *b, float *c, float *d) {
 //
   #pragma omp simd collapse(2) private(globalfloat, localint)
 
-// CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[WIDE1_LOOP_ID:[0-9]+]]
-// CHECK-NEXT: [[LI:%.+]] = load i64, i64* [[OMP_LI:%[^,]+]]{{.+}}!llvm.mem.parallel_loop_access ![[WIDE1_LOOP_ID]]
+// CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
+// CHECK-NEXT: [[LI:%.+]] = load i64, i64* [[OMP_LI:%[^,]+]]{{.+}}!llvm.access.group
 // CHECK-NEXT: [[NUMIT:%.+]] = add nsw i64 [[LI]], 1
 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i64 [[IV]], [[NUMIT]]
 // CHECK-NEXT: br i1 [[CMP]], label %[[WIDE1_BODY:[^,]+]], label %[[WIDE1_END:[^,]+]]
@@ -455,10 +455,10 @@ void widened(float *a, float *b, float *c, float *d) {
   {
 // CHECK: [[WIDE1_BODY]]:
 // Start of body: calculate i from index:
-// CHECK: [[IV1:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[WIDE1_LOOP_ID]]
+// CHECK: [[IV1:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
 // Calculation of the loop counters values...
 // CHECK: store i32 {{[^,]+}}, i32* [[LC_I:.+]]
-// CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.mem.parallel_loop_access ![[WIDE1_LOOP_ID]]
+// CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
 // CHECK: store i16 {{[^,]+}}, i16* [[LC_J:.+]]
 // ... loop body ...
 //
@@ -467,14 +467,14 @@ void widened(float *a, float *b, float *c, float *d) {
     globalfloat = (float)j/i;
     float res = b[j] * c[j];
 // Store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.mem.parallel_loop_access ![[WIDE1_LOOP_ID]]
+// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
     a[i] = res * d[i];
 // Then there's a store into private var localint:
-// CHECK: store i32 {{.+}}, i32* [[LOCALINT:%[^,]+]]{{.+}}!llvm.mem.parallel_loop_access ![[WIDE1_LOOP_ID]]
+// CHECK: store i32 {{.+}}, i32* [[LOCALINT:%[^,]+]]{{.+}}!llvm.access.group
     localint = (int)j;
-// CHECK: [[IV2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[WIDE1_LOOP_ID]]
+// CHECK: [[IV2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i64 [[IV2]], 1
-// CHECK-NEXT: store i64 [[ADD2]], i64* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[WIDE1_LOOP_ID]]
+// CHECK-NEXT: store i64 [[ADD2]], i64* [[OMP_IV]]{{.*}}!llvm.access.group
 //
 // br label %{{[^,]+}}, !llvm.loop ![[WIDE1_LOOP_ID]]
 // CHECK: [[WIDE1_END]]:
@@ -505,20 +505,20 @@ void linear(float *a) {
 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]]
 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID:[0-9]+]]
+// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
   for (int i = 10; i > 1; i--) {
 // CHECK: [[SIMPLE_LOOP_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
-// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
 //
-// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
-// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
@@ -526,9 +526,9 @@ void linear(float *a) {
 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
     a[k]++;
     k = k + 3;
-// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
-// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
   }
 // CHECK: [[SIMPLE_LOOP_END]]:
@@ -550,20 +550,20 @@ void linear(float *a) {
 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]]
 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID:[0-9]+]]
+// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
   for (int i = 10; i > 1; i--) {
 // CHECK: [[SIMPLE_LOOP_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
-// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
 //
-// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
-// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
@@ -571,9 +571,9 @@ void linear(float *a) {
 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
     a[k]++;
     k = k + 3;
-// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
-// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
   }
 // CHECK: [[SIMPLE_LOOP_END]]:
@@ -591,20 +591,20 @@ void linear(float *a) {
 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[VAL_ADDR]]
 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
 
-// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID:[0-9]+]]
+// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
   for (int i = 10; i > 1; i--) {
 // CHECK: [[SIMPLE_LOOP_BODY]]:
 // Start of body: calculate i from IV:
-// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
-// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
 //
-// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
-// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
+// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
@@ -612,9 +612,9 @@ void linear(float *a) {
 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
     a[k]++;
     k = k + 3;
-// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
-// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP_ID]]
+// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
   }
 // CHECK: [[SIMPLE_LOOP_END]]:
@@ -635,7 +635,7 @@ void parallel_simd(float *a) {
 #pragma omp simd
   // TERM_DEBUG-NOT: __kmpc_global_thread_num
   // TERM_DEBUG:     invoke i32 {{.*}}bar{{.*}}()
-  // TERM_DEBUG:     unwind label %[[TERM_LPAD:.+]],
+  // TERM_DEBUG:     unwind label %[[TERM_LPAD:[^,]+]],
   // TERM_DEBUG-NOT: __kmpc_global_thread_num
   // TERM_DEBUG:     [[TERM_LPAD]]
   // TERM_DEBUG:     call void @__clang_call_terminate