[clang] clang/OpenCL: Add baseline test showing broken codegen (PR #138862)
Matt Arsenault via cfe-commits
cfe-commits at lists.llvm.org
Wed May 7 22:38:06 PDT 2025
https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/138862
>From decbf5b3a2cbb4f8844b9556fbfafec73c117687 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Wed, 7 May 2025 08:38:49 +0200
Subject: [PATCH 1/2] clang/OpenCL: Add baseline test showing broken codegen
---
.../CodeGenCXX/amdgcn-automatic-variable.cpp | 24 ++++++++++
.../amdgcn-automatic-variable.cl | 45 +++++++++++++++++++
2 files changed, 69 insertions(+)
diff --git a/clang/test/CodeGenCXX/amdgcn-automatic-variable.cpp b/clang/test/CodeGenCXX/amdgcn-automatic-variable.cpp
index c1f9310141000..0f425d78c3332 100644
--- a/clang/test/CodeGenCXX/amdgcn-automatic-variable.cpp
+++ b/clang/test/CodeGenCXX/amdgcn-automatic-variable.cpp
@@ -133,3 +133,27 @@ void func7() {
use(&x);
}
+#define __private __attribute__((opencl_private))
+
+// CHECK-LABEL: @_Z34explicit_private_address_space_ptrv(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[PLONG:%.*]] = alloca i64, align 8, addrspace(5)
+// CHECK-NEXT: [[PLONGP:%.*]] = alloca ptr, align 8, addrspace(5)
+// CHECK-NEXT: [[PLONGP_B:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
+// CHECK-NEXT: [[PLONG_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONG]] to ptr
+// CHECK-NEXT: [[PLONGP_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONGP]] to ptr
+// CHECK-NEXT: [[PLONGP_B_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONGP_B]] to ptr
+// CHECK-NEXT: store ptr [[PLONG_ASCAST]], ptr [[PLONGP_ASCAST]], align 8
+// CHECK-NEXT: [[PLONG_ASCAST_ASCAST:%.*]] = addrspacecast ptr [[PLONG_ASCAST]] to ptr addrspace(5)
+// CHECK-NEXT: store ptr addrspace(5) [[PLONG_ASCAST_ASCAST]], ptr [[PLONGP_B_ASCAST]], align 4
+// CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[PLONGP_ASCAST]], align 8
+// CHECK-NEXT: store i64 8, ptr [[TMP0]], align 8
+// CHECK-NEXT: ret void
+//
+void explicit_private_address_space_ptr() {
+ long plong;
+ long *plongp = &plong;
+
+ __private long *plongp_b = (__private long *)&plong;
+ *plongp = 8;
+}
diff --git a/clang/test/CodeGenOpenCL/amdgcn-automatic-variable.cl b/clang/test/CodeGenOpenCL/amdgcn-automatic-variable.cl
index dba6519966eb5..b252f1041d68c 100644
--- a/clang/test/CodeGenOpenCL/amdgcn-automatic-variable.cl
+++ b/clang/test/CodeGenOpenCL/amdgcn-automatic-variable.cl
@@ -109,3 +109,48 @@ void func2(void) {
void func3(void) {
float a[16][1] = {{0.}};
}
+
+// CL12-LABEL: define dso_local void @wrong_store_type_private_pointer_alloca(
+// CL12-SAME: ) #[[ATTR0]] {
+// CL12-NEXT: [[ENTRY:.*:]]
+// CL12-NEXT: [[PLONG:%.*]] = alloca i64, align 8, addrspace(5)
+// CL12-NEXT: [[PLONGP:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
+// CL12-NEXT: [[GLONGP:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
+// CL12-NEXT: store i64 5, ptr addrspace(5) [[PLONG]], align 8
+// CL12-NEXT: store ptr addrspace(5) [[PLONG]], ptr addrspace(5) [[PLONGP]], align 4
+// CL12-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[PLONGP]], align 4
+// CL12-NEXT: store i64 8, ptr addrspace(5) [[TMP0]], align 8
+// CL12-NEXT: store ptr addrspace(5) [[PLONG]], ptr addrspace(5) [[GLONGP]], align 4
+// CL12-NEXT: [[TMP1:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[GLONGP]], align 4
+// CL12-NEXT: store i64 9, ptr addrspace(5) [[TMP1]], align 8
+// CL12-NEXT: ret void
+//
+// CL20-LABEL: define dso_local void @wrong_store_type_private_pointer_alloca(
+// CL20-SAME: ) #[[ATTR0]] {
+// CL20-NEXT: [[ENTRY:.*:]]
+// CL20-NEXT: [[PLONG:%.*]] = alloca i64, align 8, addrspace(5)
+// CL20-NEXT: [[PLONGP:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
+// CL20-NEXT: [[GLONGP:%.*]] = alloca ptr, align 8, addrspace(5)
+// CL20-NEXT: [[PLONG_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONG]] to ptr
+// CL20-NEXT: [[PLONGP_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONGP]] to ptr
+// CL20-NEXT: [[GLONGP_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[GLONGP]] to ptr
+// CL20-NEXT: store i64 5, ptr [[PLONG_ASCAST]], align 8
+// CL20-NEXT: store ptr [[PLONG_ASCAST]], ptr [[PLONGP_ASCAST]], align 4
+// CL20-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr [[PLONGP_ASCAST]], align 4
+// CL20-NEXT: store i64 8, ptr addrspace(5) [[TMP0]], align 8
+// CL20-NEXT: store ptr [[PLONG_ASCAST]], ptr [[GLONGP_ASCAST]], align 8
+// CL20-NEXT: [[TMP1:%.*]] = load ptr, ptr [[GLONGP_ASCAST]], align 8
+// CL20-NEXT: store i64 9, ptr [[TMP1]], align 8
+// CL20-NEXT: ret void
+//
+void wrong_store_type_private_pointer_alloca() {
+ long plong = 5;
+
+ // This needs to write an addrspace(5) pointer to the temporary alloca
+ __private long *plongp = &plong;
+ *plongp = 8;
+
+ // This needs to write an addrspace(0) pointer to the temporary alloca in CL2.0
+ long *glongp = &plong;
+ *glongp = 9;
+}
>From e82e94d968394900b2e0f30294b619ac79cd4162 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Wed, 7 May 2025 20:27:43 +0200
Subject: [PATCH 2/2] Split and rename test
---
.../CodeGenCXX/amdgcn-automatic-variable.cpp | 30 ++++----
.../amdgcn-automatic-variable.cl | 77 ++++++++++++-------
2 files changed, 64 insertions(+), 43 deletions(-)
diff --git a/clang/test/CodeGenCXX/amdgcn-automatic-variable.cpp b/clang/test/CodeGenCXX/amdgcn-automatic-variable.cpp
index 0f425d78c3332..3c2a624bd4f95 100644
--- a/clang/test/CodeGenCXX/amdgcn-automatic-variable.cpp
+++ b/clang/test/CodeGenCXX/amdgcn-automatic-variable.cpp
@@ -137,23 +137,23 @@ void func7() {
// CHECK-LABEL: @_Z34explicit_private_address_space_ptrv(
// CHECK-NEXT: entry:
-// CHECK-NEXT: [[PLONG:%.*]] = alloca i64, align 8, addrspace(5)
-// CHECK-NEXT: [[PLONGP:%.*]] = alloca ptr, align 8, addrspace(5)
-// CHECK-NEXT: [[PLONGP_B:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
-// CHECK-NEXT: [[PLONG_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONG]] to ptr
-// CHECK-NEXT: [[PLONGP_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONGP]] to ptr
-// CHECK-NEXT: [[PLONGP_B_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONGP_B]] to ptr
-// CHECK-NEXT: store ptr [[PLONG_ASCAST]], ptr [[PLONGP_ASCAST]], align 8
-// CHECK-NEXT: [[PLONG_ASCAST_ASCAST:%.*]] = addrspacecast ptr [[PLONG_ASCAST]] to ptr addrspace(5)
-// CHECK-NEXT: store ptr addrspace(5) [[PLONG_ASCAST_ASCAST]], ptr [[PLONGP_B_ASCAST]], align 4
-// CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[PLONGP_ASCAST]], align 8
-// CHECK-NEXT: store i64 8, ptr [[TMP0]], align 8
+// CHECK-NEXT: [[VAR:%.*]] = alloca i64, align 8, addrspace(5)
+// CHECK-NEXT: [[ALLOCA_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
+// CHECK-NEXT: [[ALLOCA_ADDR_AS_PRIVATE:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
+// CHECK-NEXT: [[VAR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VAR]] to ptr
+// CHECK-NEXT: [[ALLOCA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[ALLOCA_ADDR]] to ptr
+// CHECK-NEXT: [[ALLOCA_ADDR_AS_PRIVATE_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[ALLOCA_ADDR_AS_PRIVATE]] to ptr
+// CHECK-NEXT: store ptr [[VAR_ASCAST]], ptr [[ALLOCA_ADDR_ASCAST]], align 8
+// CHECK-NEXT: [[VAR_ASCAST_ASCAST:%.*]] = addrspacecast ptr [[VAR_ASCAST]] to ptr addrspace(5)
+// CHECK-NEXT: store ptr addrspace(5) [[VAR_ASCAST_ASCAST]], ptr [[ALLOCA_ADDR_AS_PRIVATE_ASCAST]], align 4
+// CHECK-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr [[ALLOCA_ADDR_AS_PRIVATE_ASCAST]], align 4
+// CHECK-NEXT: store i64 8, ptr addrspace(5) [[TMP0]], align 8
// CHECK-NEXT: ret void
//
void explicit_private_address_space_ptr() {
- long plong;
- long *plongp = &plong;
+ long var;
+ long *alloca_addr = &var;
- __private long *plongp_b = (__private long *)&plong;
- *plongp = 8;
+ __private long *alloca_addr_as_private = (__private long *)&var;
+ *alloca_addr_as_private = 8;
}
diff --git a/clang/test/CodeGenOpenCL/amdgcn-automatic-variable.cl b/clang/test/CodeGenOpenCL/amdgcn-automatic-variable.cl
index b252f1041d68c..e28120adc0364 100644
--- a/clang/test/CodeGenOpenCL/amdgcn-automatic-variable.cl
+++ b/clang/test/CodeGenOpenCL/amdgcn-automatic-variable.cl
@@ -113,44 +113,65 @@ void func3(void) {
// CL12-LABEL: define dso_local void @wrong_store_type_private_pointer_alloca(
// CL12-SAME: ) #[[ATTR0]] {
// CL12-NEXT: [[ENTRY:.*:]]
-// CL12-NEXT: [[PLONG:%.*]] = alloca i64, align 8, addrspace(5)
-// CL12-NEXT: [[PLONGP:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
-// CL12-NEXT: [[GLONGP:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
-// CL12-NEXT: store i64 5, ptr addrspace(5) [[PLONG]], align 8
-// CL12-NEXT: store ptr addrspace(5) [[PLONG]], ptr addrspace(5) [[PLONGP]], align 4
-// CL12-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[PLONGP]], align 4
+// CL12-NEXT: [[VAR:%.*]] = alloca i64, align 8, addrspace(5)
+// CL12-NEXT: [[ALLOCA_ADDR:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
+// CL12-NEXT: store i64 5, ptr addrspace(5) [[VAR]], align 8
+// CL12-NEXT: store ptr addrspace(5) [[VAR]], ptr addrspace(5) [[ALLOCA_ADDR]], align 4
+// CL12-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[ALLOCA_ADDR]], align 4
// CL12-NEXT: store i64 8, ptr addrspace(5) [[TMP0]], align 8
-// CL12-NEXT: store ptr addrspace(5) [[PLONG]], ptr addrspace(5) [[GLONGP]], align 4
-// CL12-NEXT: [[TMP1:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[GLONGP]], align 4
-// CL12-NEXT: store i64 9, ptr addrspace(5) [[TMP1]], align 8
// CL12-NEXT: ret void
//
// CL20-LABEL: define dso_local void @wrong_store_type_private_pointer_alloca(
// CL20-SAME: ) #[[ATTR0]] {
// CL20-NEXT: [[ENTRY:.*:]]
-// CL20-NEXT: [[PLONG:%.*]] = alloca i64, align 8, addrspace(5)
-// CL20-NEXT: [[PLONGP:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
-// CL20-NEXT: [[GLONGP:%.*]] = alloca ptr, align 8, addrspace(5)
-// CL20-NEXT: [[PLONG_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONG]] to ptr
-// CL20-NEXT: [[PLONGP_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[PLONGP]] to ptr
-// CL20-NEXT: [[GLONGP_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[GLONGP]] to ptr
-// CL20-NEXT: store i64 5, ptr [[PLONG_ASCAST]], align 8
-// CL20-NEXT: store ptr [[PLONG_ASCAST]], ptr [[PLONGP_ASCAST]], align 4
-// CL20-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr [[PLONGP_ASCAST]], align 4
+// CL20-NEXT: [[VAR:%.*]] = alloca i64, align 8, addrspace(5)
+// CL20-NEXT: [[ALLOCA_ADDR:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
+// CL20-NEXT: [[VAR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VAR]] to ptr
+// CL20-NEXT: [[ALLOCA_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[ALLOCA_ADDR]] to ptr
+// CL20-NEXT: store i64 5, ptr [[VAR_ASCAST]], align 8
+// CL20-NEXT: store ptr [[VAR_ASCAST]], ptr [[ALLOCA_ADDR_ASCAST]], align 4
+// CL20-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr [[ALLOCA_ADDR_ASCAST]], align 4
// CL20-NEXT: store i64 8, ptr addrspace(5) [[TMP0]], align 8
-// CL20-NEXT: store ptr [[PLONG_ASCAST]], ptr [[GLONGP_ASCAST]], align 8
-// CL20-NEXT: [[TMP1:%.*]] = load ptr, ptr [[GLONGP_ASCAST]], align 8
-// CL20-NEXT: store i64 9, ptr [[TMP1]], align 8
// CL20-NEXT: ret void
//
void wrong_store_type_private_pointer_alloca() {
- long plong = 5;
+ long var = 5;
- // This needs to write an addrspace(5) pointer to the temporary alloca
- __private long *plongp = &plong;
- *plongp = 8;
+ // This needs to write an addrspace(5) pointer to the temporary alloca,
+ // which should be allocated with the correct size.
+ __private long *alloca_addr = &var;
+ *alloca_addr = 8;
+}
+
+// CL12-LABEL: define dso_local void @wrong_store_type_private_pointer_as_generic_alloca(
+// CL12-SAME: ) #[[ATTR0]] {
+// CL12-NEXT: [[ENTRY:.*:]]
+// CL12-NEXT: [[VAR:%.*]] = alloca i64, align 8, addrspace(5)
+// CL12-NEXT: [[ALLOCA_ADDR_AS_GENERIC:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
+// CL12-NEXT: store i64 5, ptr addrspace(5) [[VAR]], align 8
+// CL12-NEXT: store ptr addrspace(5) [[VAR]], ptr addrspace(5) [[ALLOCA_ADDR_AS_GENERIC]], align 4
+// CL12-NEXT: [[TMP0:%.*]] = load ptr addrspace(5), ptr addrspace(5) [[ALLOCA_ADDR_AS_GENERIC]], align 4
+// CL12-NEXT: store i64 9, ptr addrspace(5) [[TMP0]], align 8
+// CL12-NEXT: ret void
+//
+// CL20-LABEL: define dso_local void @wrong_store_type_private_pointer_as_generic_alloca(
+// CL20-SAME: ) #[[ATTR0]] {
+// CL20-NEXT: [[ENTRY:.*:]]
+// CL20-NEXT: [[VAR:%.*]] = alloca i64, align 8, addrspace(5)
+// CL20-NEXT: [[ALLOCA_ADDR_AS_GENERIC:%.*]] = alloca ptr, align 8, addrspace(5)
+// CL20-NEXT: [[VAR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VAR]] to ptr
+// CL20-NEXT: [[ALLOCA_ADDR_AS_GENERIC_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[ALLOCA_ADDR_AS_GENERIC]] to ptr
+// CL20-NEXT: store i64 5, ptr [[VAR_ASCAST]], align 8
+// CL20-NEXT: store ptr [[VAR_ASCAST]], ptr [[ALLOCA_ADDR_AS_GENERIC_ASCAST]], align 8
+// CL20-NEXT: [[TMP0:%.*]] = load ptr, ptr [[ALLOCA_ADDR_AS_GENERIC_ASCAST]], align 8
+// CL20-NEXT: store i64 9, ptr [[TMP0]], align 8
+// CL20-NEXT: ret void
+//
+void wrong_store_type_private_pointer_as_generic_alloca() {
+ long var = 5;
- // This needs to write an addrspace(0) pointer to the temporary alloca in CL2.0
- long *glongp = &plong;
- *glongp = 9;
+ // This needs to write an addrspace(0) pointer to the temporary alloca in
+ // CL2.0, which should be allocated with the correct size.
+ long *alloca_addr_as_generic = &var;
+ *alloca_addr_as_generic = 9;
}
More information about the cfe-commits
mailing list