Skip to content

Commit

Permalink
[RISCV] Reduce alignment for __attribute__((riscv_rvv_vector_bits)) f…
Browse files Browse the repository at this point in the history 
…or LMUL<1 types.

Don't use an alignment larger than the vector size.
  • Loading branch information
@topperc
topperc committed Jun 16, 2023
1 parent b3137d5 commit 3ab9d4f
Show file tree
Hide file tree
Showing 3 changed files with 131 additions and 14 deletions.
2 changes: 1 addition & 1 deletion clang/lib/AST/ASTContext.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1998,7 +1998,7 @@ TypeInfo ASTContext::getTypeInfoImpl(const Type *T) const {
Align = 16;
else if (VT->getVectorKind() == VectorType::RVVFixedLengthDataVector)
// Adjust the alignment for fixed-length RVV vectors.
Align = 64;
Align = std::min<unsigned>(64, Width);
break;
}

Expand Down
117 changes: 117 additions & 0 deletions clang/test/CodeGen/attr-riscv-rvv-vector-bits-types.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,22 @@

#include <stdint.h>

typedef __rvv_int8mf8_t vint8mf8_t;
typedef __rvv_uint8mf8_t vuint8mf8_t;

typedef __rvv_int8mf4_t vint8mf4_t;
typedef __rvv_uint8mf4_t vuint8mf4_t;
typedef __rvv_int16mf4_t vint16mf4_t;
typedef __rvv_uint16mf4_t vuint16mf4_t;

typedef __rvv_int8mf2_t vint8mf2_t;
typedef __rvv_uint8mf2_t vuint8mf2_t;
typedef __rvv_int16mf2_t vint16mf2_t;
typedef __rvv_uint16mf2_t vuint16mf2_t;
typedef __rvv_int32mf2_t vint32mf2_t;
typedef __rvv_uint32mf2_t vuint32mf2_t;
typedef __rvv_float32mf2_t vfloat32mf2_t;

typedef __rvv_int8m1_t vint8m1_t;
typedef __rvv_uint8m1_t vuint8m1_t;
typedef __rvv_int16m1_t vint16m1_t;
Expand Down Expand Up @@ -53,6 +69,26 @@ typedef __rvv_float32m8_t vfloat32m8_t;
typedef __rvv_float64m8_t vfloat64m8_t;

// Define valid fixed-width RVV types
typedef vint8mf8_t fixed_int8mf8_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 8)));

typedef vuint8mf8_t fixed_uint8mf8_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 8)));

typedef vint8mf4_t fixed_int8mf4_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 4)));
typedef vint16mf4_t fixed_int16mf4_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 4)));

typedef vuint8mf4_t fixed_uint8mf4_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 4)));
typedef vuint16mf4_t fixed_uint16mf4_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 4)));

typedef vint8mf2_t fixed_int8mf2_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 2)));
typedef vint16mf2_t fixed_int16mf2_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 2)));
typedef vint32mf2_t fixed_int32mf2_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 2)));

typedef vuint8mf2_t fixed_uint8mf2_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 2)));
typedef vuint16mf2_t fixed_uint16mf2_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 2)));
typedef vuint32mf2_t fixed_uint32mf2_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 2)));

typedef vfloat32mf2_t fixed_float32mf2_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen / 2)));

typedef vint8m1_t fixed_int8m1_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen)));
typedef vint16m1_t fixed_int16m1_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen)));
typedef vint32m1_t fixed_int32m1_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen)));
Expand Down Expand Up @@ -409,6 +445,22 @@ void f() {
fixed_uint64m8_t local_arr_u64m8[3];
fixed_float32m8_t local_arr_f32m8[3];
fixed_float64m8_t local_arr_f64m8[3];

fixed_int8mf2_t local_arr_i8mf2[3];
fixed_int16mf2_t local_arr_i16mf2[3];
fixed_int32mf2_t local_arr_i32mf2[3];
fixed_uint8mf2_t local_arr_u8mf2[3];
fixed_uint16mf2_t local_arr_u16mf2[3];
fixed_uint32mf2_t local_arr_u32mf2[3];
fixed_float32mf2_t local_arr_f32mf2[3];

fixed_int8mf4_t local_arr_i8mf4[3];
fixed_int16mf4_t local_arr_i16mf4[3];
fixed_uint8mf4_t local_arr_u8mf4[3];
fixed_uint16mf4_t local_arr_u16mf4[3];

fixed_int8mf8_t local_arr_i8mf8[3];
fixed_uint8mf8_t local_arr_u8mf8[3];
}

//===----------------------------------------------------------------------===//
Expand Down Expand Up @@ -1487,6 +1539,19 @@ void f() {
// CHECK-64-NEXT: %local_arr_u64m8 = alloca [3 x <8 x i64>], align 8
// CHECK-64-NEXT: %local_arr_f32m8 = alloca [3 x <16 x float>], align 8
// CHECK-64-NEXT: %local_arr_f64m8 = alloca [3 x <8 x double>], align 8
// CHECK-64-NEXT: %local_arr_i8mf2 = alloca [3 x <4 x i8>], align 4
// CHECK-64-NEXT: %local_arr_i16mf2 = alloca [3 x <2 x i16>], align 4
// CHECK-64-NEXT: %local_arr_i32mf2 = alloca [3 x <1 x i32>], align 4
// CHECK-64-NEXT: %local_arr_u8mf2 = alloca [3 x <4 x i8>], align 4
// CHECK-64-NEXT: %local_arr_u16mf2 = alloca [3 x <2 x i16>], align 4
// CHECK-64-NEXT: %local_arr_u32mf2 = alloca [3 x <1 x i32>], align 4
// CHECK-64-NEXT: %local_arr_f32mf2 = alloca [3 x <1 x float>], align 4
// CHECK-64-NEXT: %local_arr_i8mf4 = alloca [3 x <2 x i8>], align 2
// CHECK-64-NEXT: %local_arr_i16mf4 = alloca [3 x <1 x i16>], align 2
// CHECK-64-NEXT: %local_arr_u8mf4 = alloca [3 x <2 x i8>], align 2
// CHECK-64-NEXT: %local_arr_u16mf4 = alloca [3 x <1 x i16>], align 2
// CHECK-64-NEXT: %local_arr_i8mf8 = alloca [3 x <1 x i8>], align 1
// CHECK-64-NEXT: %local_arr_u8mf8 = alloca [3 x <1 x i8>], align 1

// CHECK-128: %local_arr_i8 = alloca [3 x <16 x i8>], align 8
// CHECK-128-NEXT: %local_arr_i16 = alloca [3 x <8 x i16>], align 8
Expand Down Expand Up @@ -1528,6 +1593,19 @@ void f() {
// CHECK-128-NEXT: %local_arr_u64m8 = alloca [3 x <16 x i64>], align 8
// CHECK-128-NEXT: %local_arr_f32m8 = alloca [3 x <32 x float>], align 8
// CHECK-128-NEXT: %local_arr_f64m8 = alloca [3 x <16 x double>], align 8
// CHECK-128-NEXT: %local_arr_i8mf2 = alloca [3 x <8 x i8>], align 8
// CHECK-128-NEXT: %local_arr_i16mf2 = alloca [3 x <4 x i16>], align 8
// CHECK-128-NEXT: %local_arr_i32mf2 = alloca [3 x <2 x i32>], align 8
// CHECK-128-NEXT: %local_arr_u8mf2 = alloca [3 x <8 x i8>], align 8
// CHECK-128-NEXT: %local_arr_u16mf2 = alloca [3 x <4 x i16>], align 8
// CHECK-128-NEXT: %local_arr_u32mf2 = alloca [3 x <2 x i32>], align 8
// CHECK-128-NEXT: %local_arr_f32mf2 = alloca [3 x <2 x float>], align 8
// CHECK-128-NEXT: %local_arr_i8mf4 = alloca [3 x <4 x i8>], align 4
// CHECK-128-NEXT: %local_arr_i16mf4 = alloca [3 x <2 x i16>], align 4
// CHECK-128-NEXT: %local_arr_u8mf4 = alloca [3 x <4 x i8>], align 4
// CHECK-128-NEXT: %local_arr_u16mf4 = alloca [3 x <2 x i16>], align 4
// CHECK-128-NEXT: %local_arr_i8mf8 = alloca [3 x <2 x i8>], align 2
// CHECK-128-NEXT: %local_arr_u8mf8 = alloca [3 x <2 x i8>], align 2

// CHECK-256: %local_arr_i8 = alloca [3 x <32 x i8>], align 8
// CHECK-256-NEXT: %local_arr_i16 = alloca [3 x <16 x i16>], align 8
Expand Down Expand Up @@ -1569,6 +1647,19 @@ void f() {
// CHECK-256-NEXT: %local_arr_u64m8 = alloca [3 x <32 x i64>], align 8
// CHECK-256-NEXT: %local_arr_f32m8 = alloca [3 x <64 x float>], align 8
// CHECK-256-NEXT: %local_arr_f64m8 = alloca [3 x <32 x double>], align 8
// CHECK-256-NEXT: %local_arr_i8mf2 = alloca [3 x <16 x i8>], align 8
// CHECK-256-NEXT: %local_arr_i16mf2 = alloca [3 x <8 x i16>], align 8
// CHECK-256-NEXT: %local_arr_i32mf2 = alloca [3 x <4 x i32>], align 8
// CHECK-256-NEXT: %local_arr_u8mf2 = alloca [3 x <16 x i8>], align 8
// CHECK-256-NEXT: %local_arr_u16mf2 = alloca [3 x <8 x i16>], align 8
// CHECK-256-NEXT: %local_arr_u32mf2 = alloca [3 x <4 x i32>], align 8
// CHECK-256-NEXT: %local_arr_f32mf2 = alloca [3 x <4 x float>], align 8
// CHECK-256-NEXT: %local_arr_i8mf4 = alloca [3 x <8 x i8>], align 8
// CHECK-256-NEXT: %local_arr_i16mf4 = alloca [3 x <4 x i16>], align 8
// CHECK-256-NEXT: %local_arr_u8mf4 = alloca [3 x <8 x i8>], align 8
// CHECK-256-NEXT: %local_arr_u16mf4 = alloca [3 x <4 x i16>], align 8
// CHECK-256-NEXT: %local_arr_i8mf8 = alloca [3 x <4 x i8>], align 4
// CHECK-256-NEXT: %local_arr_u8mf8 = alloca [3 x <4 x i8>], align 4

// CHECK-512: %local_arr_i8 = alloca [3 x <64 x i8>], align 8
// CHECK-512-NEXT: %local_arr_i16 = alloca [3 x <32 x i16>], align 8
Expand Down Expand Up @@ -1610,6 +1701,19 @@ void f() {
// CHECK-512-NEXT: %local_arr_u64m8 = alloca [3 x <64 x i64>], align 8
// CHECK-512-NEXT: %local_arr_f32m8 = alloca [3 x <128 x float>], align 8
// CHECK-512-NEXT: %local_arr_f64m8 = alloca [3 x <64 x double>], align 8
// CHECK-512-NEXT: %local_arr_i8mf2 = alloca [3 x <32 x i8>], align 8
// CHECK-512-NEXT: %local_arr_i16mf2 = alloca [3 x <16 x i16>], align 8
// CHECK-512-NEXT: %local_arr_i32mf2 = alloca [3 x <8 x i32>], align 8
// CHECK-512-NEXT: %local_arr_u8mf2 = alloca [3 x <32 x i8>], align 8
// CHECK-512-NEXT: %local_arr_u16mf2 = alloca [3 x <16 x i16>], align 8
// CHECK-512-NEXT: %local_arr_u32mf2 = alloca [3 x <8 x i32>], align 8
// CHECK-512-NEXT: %local_arr_f32mf2 = alloca [3 x <8 x float>], align 8
// CHECK-512-NEXT: %local_arr_i8mf4 = alloca [3 x <16 x i8>], align 8
// CHECK-512-NEXT: %local_arr_i16mf4 = alloca [3 x <8 x i16>], align 8
// CHECK-512-NEXT: %local_arr_u8mf4 = alloca [3 x <16 x i8>], align 8
// CHECK-512-NEXT: %local_arr_u16mf4 = alloca [3 x <8 x i16>], align 8
// CHECK-512-NEXT: %local_arr_i8mf8 = alloca [3 x <8 x i8>], align 8
// CHECK-512-NEXT: %local_arr_u8mf8 = alloca [3 x <8 x i8>], align 8

// CHECK-1024: %local_arr_i8 = alloca [3 x <128 x i8>], align 8
// CHECK-1024-NEXT: %local_arr_i16 = alloca [3 x <64 x i16>], align 8
Expand Down Expand Up @@ -1651,3 +1755,16 @@ void f() {
// CHECK-1024-NEXT: %local_arr_u64m8 = alloca [3 x <128 x i64>], align 8
// CHECK-1024-NEXT: %local_arr_f32m8 = alloca [3 x <256 x float>], align 8
// CHECK-1024-NEXT: %local_arr_f64m8 = alloca [3 x <128 x double>], align 8
// CHECK-1024-NEXT: %local_arr_i8mf2 = alloca [3 x <64 x i8>], align 8
// CHECK-1024-NEXT: %local_arr_i16mf2 = alloca [3 x <32 x i16>], align 8
// CHECK-1024-NEXT: %local_arr_i32mf2 = alloca [3 x <16 x i32>], align 8
// CHECK-1024-NEXT: %local_arr_u8mf2 = alloca [3 x <64 x i8>], align 8
// CHECK-1024-NEXT: %local_arr_u16mf2 = alloca [3 x <32 x i16>], align 8
// CHECK-1024-NEXT: %local_arr_u32mf2 = alloca [3 x <16 x i32>], align 8
// CHECK-1024-NEXT: %local_arr_f32mf2 = alloca [3 x <16 x float>], align 8
// CHECK-1024-NEXT: %local_arr_i8mf4 = alloca [3 x <32 x i8>], align 8
// CHECK-1024-NEXT: %local_arr_i16mf4 = alloca [3 x <16 x i16>], align 8
// CHECK-1024-NEXT: %local_arr_u8mf4 = alloca [3 x <32 x i8>], align 8
// CHECK-1024-NEXT: %local_arr_u16mf4 = alloca [3 x <16 x i16>], align 8
// CHECK-1024-NEXT: %local_arr_i8mf8 = alloca [3 x <16 x i8>], align 8
// CHECK-1024-NEXT: %local_arr_u8mf8 = alloca [3 x <16 x i8>], align 8
26 changes: 13 additions & 13 deletions clang/test/Sema/attr-riscv-rvv-vector-bits.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -403,25 +403,25 @@ _Static_assert(sizeof(fixed_float64m8_t) == VECTOR_SIZE * 8, "");

#define VECTOR_ALIGN 8

_Static_assert(__alignof__(fixed_int8mf8_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_int8mf8_t) == (sizeof(fixed_int8mf8_t) < VECTOR_ALIGN ? sizeof(fixed_int8mf8_t) : VECTOR_ALIGN), "");

_Static_assert(__alignof__(fixed_uint8mf8_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_uint8mf8_t) == (sizeof(fixed_uint8mf8_t) < VECTOR_ALIGN ? sizeof(fixed_int8mf8_t) : VECTOR_ALIGN), "");

_Static_assert(__alignof__(fixed_int8mf4_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_int16mf4_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_int8mf4_t) == (sizeof(fixed_int8mf4_t) < VECTOR_ALIGN ? sizeof(fixed_int8mf4_t) : VECTOR_ALIGN), "");
_Static_assert(__alignof__(fixed_int16mf4_t) == (sizeof(fixed_int16mf4_t) < VECTOR_ALIGN ? sizeof(fixed_int16mf4_t) : VECTOR_ALIGN), "");

_Static_assert(__alignof__(fixed_uint8mf4_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_uint16mf4_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_uint8mf4_t) == (sizeof(fixed_uint8mf4_t) < VECTOR_ALIGN ? sizeof(fixed_uint8mf4_t) : VECTOR_ALIGN), "");
_Static_assert(__alignof__(fixed_uint16mf4_t) == (sizeof(fixed_uint16mf4_t) < VECTOR_ALIGN ? sizeof(fixed_uint16mf4_t) : VECTOR_ALIGN), "");

_Static_assert(__alignof__(fixed_int8mf2_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_int16mf2_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_int32mf2_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_int8mf2_t) == (sizeof(fixed_int8mf2_t) < VECTOR_ALIGN ? sizeof(fixed_int8mf2_t) : VECTOR_ALIGN), "");
_Static_assert(__alignof__(fixed_int16mf2_t) == (sizeof(fixed_int16mf2_t) < VECTOR_ALIGN ? sizeof(fixed_int16mf2_t) : VECTOR_ALIGN), "");
_Static_assert(__alignof__(fixed_int32mf2_t) == (sizeof(fixed_int32mf2_t) < VECTOR_ALIGN ? sizeof(fixed_int32mf2_t) : VECTOR_ALIGN), "");

_Static_assert(__alignof__(fixed_uint8mf2_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_uint16mf2_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_uint32mf2_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_uint8mf2_t) == (sizeof(fixed_uint8mf2_t) < VECTOR_ALIGN ? sizeof(fixed_uint8mf2_t) : VECTOR_ALIGN), "");
_Static_assert(__alignof__(fixed_uint16mf2_t) == (sizeof(fixed_uint16mf2_t) < VECTOR_ALIGN ? sizeof(fixed_uint16mf2_t) : VECTOR_ALIGN), "");
_Static_assert(__alignof__(fixed_uint32mf2_t) == (sizeof(fixed_uint32mf2_t) < VECTOR_ALIGN ? sizeof(fixed_uint32mf2_t) : VECTOR_ALIGN), "");

_Static_assert(__alignof__(fixed_float32mf2_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_float32mf2_t) == (sizeof(fixed_float32mf2_t) < VECTOR_ALIGN ? sizeof(fixed_float32mf2_t) : VECTOR_ALIGN), "");

_Static_assert(__alignof__(fixed_int8m1_t) == VECTOR_ALIGN, "");
_Static_assert(__alignof__(fixed_int16m1_t) == VECTOR_ALIGN, "");
Expand Down

0 comments on commit 3ab9d4f

Please sign in to comment.