[RISCV] make _Complex {integer} consistent with GCC - #216404
Merged
Merged
Conversation
Complex integers should not be passed via FPRs.
llvmorg-github-actions
Bot
added
clang
|
@llvm/pr-subscribers-backend-risc-v @llvm/pr-subscribers-clang Author: Folkert de Vries (folkertdev) ChangesThe Loongarch fix from #215222 but for riscv, which hits the same issue. Clang appears to pass https://godbolt.org/z/dKhPToqcd struct S { _Complex unsigned short c; };
void callee(struct S s);
void caller(void) {
struct S s;
__real__ s.c = 0xC1C0;
__imag__ s.c = 0xD1D0;
callee(s);
}Clang uses both caller:
lui a0, 12
lui a1, 13
addi a0, a0, 448
addi a1, a1, 464
tail calleeGCC bitpacks everything into caller:
addi sp,sp,-16
sd ra,8(sp)
li a0,-774848512
addiw a0,a0,448
call callee@<!-- -->plt
ld ra,8(sp)
addi sp,sp,16
jr raComplex integers are a GNU extension, so clang behavior should match GCC. Full diff: https://github.com/llvm/llvm-project/pull/216404.diff 3 Files Affected:
diff --git a/clang/lib/CodeGen/Targets/RISCV.cpp b/clang/lib/CodeGen/Targets/RISCV.cpp
index ce2352ca76284..4bac5711a2dd6 100644
--- a/clang/lib/CodeGen/Targets/RISCV.cpp
+++ b/clang/lib/CodeGen/Targets/RISCV.cpp
@@ -225,6 +225,12 @@ bool RISCVABIInfo::detectFPCCEligibleStructHelper(QualType Ty, CharUnits CurOff,
if (Field1Ty)
return false;
QualType EltTy = CTy->getElementType();
+ // Only floating-point complex types (e.g. _Complex float/double) are
+ // eligible to be passed in floating-point argument registers. Complex
+ // integer types (a GNU extension) should be treated like a normal
+ // aggregate and packed into GPRs instead.
+ if (!EltTy->isRealFloatingType())
+ return false;
if (getContext().getTypeSize(EltTy) > FLen)
return false;
Field1Ty = CGT.ConvertType(EltTy);
diff --git a/clang/test/CodeGen/RISCV/riscv32-abi.c b/clang/test/CodeGen/RISCV/riscv32-abi.c
index e9f7e6c26a0dc..9fa34a9a75cf2 100644
--- a/clang/test/CodeGen/RISCV/riscv32-abi.c
+++ b/clang/test/CodeGen/RISCV/riscv32-abi.c
@@ -1888,6 +1888,9 @@ struct float16complex_s f_ret_float16complex_s(void) {
return (struct float16complex_s){1.0};
}
+// Complex integer values or structs containing a single complex
+// integer value should be passed as if it were an int+int struct.
+
// CHECK-LABEL: define{{.*}} i64 @f_ucharcomplex(i64 %x.coerce)
// ILP32-ILP32F-ILP32D-LABEL: define dso_local i16 @f_ucharcomplex
// ILP32-ILP32F-ILP32D-SAME: (i16 noundef [[X_COERCE:%.*]]) #[[ATTR0]] {
@@ -1906,13 +1909,9 @@ struct ucharcomplex_s {
unsigned char __complex__ c;
};
// CHECK-LABEL: define{{.*}} i64 @f_ucharcomplex_s(i64 %x.coerce)
-// ILP32-LABEL: define dso_local i16 @f_ucharcomplex_s
-// ILP32-SAME: (i16 [[X_COERCE:%.*]]) #[[ATTR0]] {
-// ILP32: entry:
-//
-// ILP32F-ILP32D-LABEL: define dso_local { i8, i8 } @f_ucharcomplex_s
-// ILP32F-ILP32D-SAME: (i8 [[TMP0:%.*]], i8 [[TMP1:%.*]]) #[[ATTR0]] {
-// ILP32F-ILP32D: entry:
+// ILP32-ILP32F-ILP32D-LABEL: define dso_local i16 @f_ucharcomplex_s
+// ILP32-ILP32F-ILP32D-SAME: (i16 [[X_COERCE:%.*]]) #[[ATTR0]] {
+// ILP32-ILP32F-ILP32D: entry:
//
struct ucharcomplex_s f_ucharcomplex_s(struct ucharcomplex_s x) {
return x;
@@ -1922,13 +1921,9 @@ struct ushortcomplex_s {
unsigned short __complex__ c;
};
// CHECK-LABEL: define{{.*}} i64 @f_ushortcomplex_s(i64 %x.coerce)
-// ILP32-LABEL: define dso_local i32 @f_ushortcomplex_s
-// ILP32-SAME: (i32 [[X_COERCE:%.*]]) #[[ATTR0]] {
-// ILP32: entry:
-//
-// ILP32F-ILP32D-LABEL: define dso_local { i16, i16 } @f_ushortcomplex_s
-// ILP32F-ILP32D-SAME: (i16 [[TMP0:%.*]], i16 [[TMP1:%.*]]) #[[ATTR0]] {
-// ILP32F-ILP32D: entry:
+// ILP32-ILP32F-ILP32D-LABEL: define dso_local i32 @f_ushortcomplex_s
+// ILP32-ILP32F-ILP32D-SAME: (i32 [[X_COERCE:%.*]]) #[[ATTR0]] {
+// ILP32-ILP32F-ILP32D: entry:
//
struct ushortcomplex_s f_ushortcomplex_s(struct ushortcomplex_s x) {
return x;
diff --git a/clang/test/CodeGen/RISCV/riscv64-abi.c b/clang/test/CodeGen/RISCV/riscv64-abi.c
index ac5524720ee31..4a2f89e9ec6b1 100644
--- a/clang/test/CodeGen/RISCV/riscv64-abi.c
+++ b/clang/test/CodeGen/RISCV/riscv64-abi.c
@@ -709,6 +709,9 @@ struct floatcomplex_s f_ret_floatcomplex_s(void) {
return (struct floatcomplex_s){1.0};
}
+// Complex integer values or structs containing a single complex
+// integer value should be passed as if it were an int+int struct.
+
// CHECK-LABEL: define{{.*}} i64 @f_ucharcomplex(i64 %x.coerce)
// LP64-LP64F-LP64D-LABEL: define dso_local i16 @f_ucharcomplex
// LP64-LP64F-LP64D-SAME: (i16 noundef [[X_COERCE:%.*]]) #[[ATTR0]] {
@@ -727,13 +730,9 @@ struct ucharcomplex_s {
unsigned char __complex__ c;
};
// CHECK-LABEL: define{{.*}} i64 @f_ucharcomplex_s(i64 %x.coerce)
-// LP64-LABEL: define dso_local i16 @f_ucharcomplex_s
-// LP64-SAME: (i16 [[X_COERCE:%.*]]) #[[ATTR0]] {
-// LP64: entry:
-//
-// LP64F-LP64D-LABEL: define dso_local { i8, i8 } @f_ucharcomplex_s
-// LP64F-LP64D-SAME: (i8 [[TMP0:%.*]], i8 [[TMP1:%.*]]) #[[ATTR0]] {
-// LP64F-LP64D: entry:
+// LP64-LP64F-LP64D-LABEL: define dso_local i16 @f_ucharcomplex_s
+// LP64-LP64F-LP64D-SAME: (i16 [[X_COERCE:%.*]]) #[[ATTR0]] {
+// LP64-LP64F-LP64D: entry:
//
struct ucharcomplex_s f_ucharcomplex_s(struct ucharcomplex_s x) {
return x;
@@ -743,13 +742,9 @@ struct ushortcomplex_s {
unsigned short __complex__ c;
};
// CHECK-LABEL: define{{.*}} i64 @f_ushortcomplex_s(i64 %x.coerce)
-// LP64-LABEL: define dso_local i32 @f_ushortcomplex_s
-// LP64-SAME: (i32 [[X_COERCE:%.*]]) #[[ATTR0]] {
-// LP64: entry:
-//
-// LP64F-LP64D-LABEL: define dso_local { i16, i16 } @f_ushortcomplex_s
-// LP64F-LP64D-SAME: (i16 [[TMP0:%.*]], i16 [[TMP1:%.*]]) #[[ATTR0]] {
-// LP64F-LP64D: entry:
+// LP64-LP64F-LP64D-LABEL: define dso_local i32 @f_ushortcomplex_s
+// LP64-LP64F-LP64D-SAME: (i32 [[X_COERCE:%.*]]) #[[ATTR0]] {
+// LP64-LP64F-LP64D: entry:
//
struct ushortcomplex_s f_ushortcomplex_s(struct ushortcomplex_s x) {
return x;
|
topperc
reviewed
Aug 14, 2026
| @@ -727,13 +730,9 @@ struct ucharcomplex_s { | |||
| unsigned char __complex__ c; | |||
| }; | |||
| // CHECK-LABEL: define{{.*}} i64 @f_ucharcomplex_s(i64 %x.coerce) | |||
Contributor
There was a problem hiding this comment.
I don't see a RUN line with "CHECK" as a prefix". Is this an unused line?
Contributor
Author
There was a problem hiding this comment.
ah, those didn't get removed when running the update script. fixed.
Member
|
From the psABI (which covers C99 standard
-- §4.4 "C/C++ Type Representations"
-- §2.1 "Integer Calling Convention" I think there are a few different choices here:
|
Contributor
Author
|
Given that it's a GNU extension, GCC is the reference implementation right? Most targets do already match GCC (usually by just treating |
Member
|
Sorry the part from §2.1 i quoted only applies to For
(earlier in §2.1) So this is correct, sorry for the confusion. |
🐧 Linux x64 Test Results
All executed tests passed, but another part of the build failed. Click on a failure below to see the details. libcxx/src/CMakeFiles/cxx_shared.dir/algorithm.cpp.olibcxx/src/CMakeFiles/cxx_static.dir/algorithm.cpp.oIf these failures are unrelated to your changes (for example tests are broken or flaky at HEAD), please open an issue at https://github.com/llvm/llvm-project/issues and add the |
folkertdev
enabled auto-merge (squash)
August 15, 2026 12:55
zhangweize9-cyber
pushed a commit
to zhangweize9-cyber/llvm-project
that referenced
this pull request
Aug 16, 2026
The Loongarch fix from llvm#215222 but for riscv, which hits the same issue. Clang appears to pass `_Complex short` and `_Complex char` like a struct with two `short`/`char` fields, taking up 2 registers. GCC instead packs the values together into a single register. https://godbolt.org/z/dKhPToqcd ```c struct S { _Complex unsigned short c; }; void callee(struct S s); void caller(void) { struct S s; __real__ s.c = 0xC1C0; __imag__ s.c = 0xD1D0; callee(s); } ``` Clang uses both `a0` and `a1`: ```asm caller: lui a0, 12 lui a1, 13 addi a0, a0, 448 addi a1, a1, 464 tail callee ``` GCC bitpacks everything into `a0`: ```asm caller: addi sp,sp,-16 sd ra,8(sp) li a0,-774848512 addiw a0,a0,448 call callee@plt ld ra,8(sp) addi sp,sp,16 jr ra ``` Complex integers are a GNU extension, so clang behavior should match GCC.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The Loongarch fix from #215222 but for riscv, which hits the same issue.
Clang appears to pass
_Complex shortand_Complex charlike a struct with twoshort/charfields, taking up 2 registers. GCC instead packs the values together into a single register.https://godbolt.org/z/dKhPToqcd
Clang uses both
a0anda1:GCC bitpacks everything into
a0:Complex integers are a GNU extension, so clang behavior should match GCC.