Correct the instruction name and add floating point instructions

This commit is contained in:
SparrowLii 2021-03-04 23:54:37 +08:00 committed by Amanieu d'Antras
parent 3bfa7593a7
commit d4952b2084
3 changed files with 175 additions and 87 deletions

View file

@ -73,148 +73,202 @@ pub unsafe fn vceqq_f64(a: float64x2_t, b: float64x2_t) -> uint64x2_t {
simd_eq(a, b)
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqz_s8(a: int8x8_t) -> uint8x8_t {
simd_eq(a, transmute(i8x8::new(0, 0, 0, 0, 0, 0, 0, 0)))
let b: i8x8 = i8x8::new(0, 0, 0, 0, 0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqzq_s8(a: int8x16_t) -> uint8x16_t {
simd_eq(a, transmute(i8x16::new(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)))
let b: i8x16 = i8x16::new(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqz_s16(a: int16x4_t) -> uint16x4_t {
simd_eq(a, transmute(i16x4::new(0, 0, 0, 0)))
let b: i16x4 = i16x4::new(0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqzq_s16(a: int16x8_t) -> uint16x8_t {
simd_eq(a, transmute(i16x8::new(0, 0, 0, 0, 0, 0, 0, 0)))
let b: i16x8 = i16x8::new(0, 0, 0, 0, 0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqz_s32(a: int32x2_t) -> uint32x2_t {
simd_eq(a, transmute(i32x2::new(0, 0)))
let b: i32x2 = i32x2::new(0, 0);
simd_eq(a, transmute(b))
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqzq_s32(a: int32x4_t) -> uint32x4_t {
simd_eq(a, transmute(i32x4::new(0, 0, 0, 0)))
let b: i32x4 = i32x4::new(0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqz_s64(a: int64x1_t) -> uint64x1_t {
simd_eq(a, transmute(i64x1::new(0)))
let b: i64x1 = i64x1::new(0);
simd_eq(a, transmute(b))
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqzq_s64(a: int64x2_t) -> uint64x2_t {
simd_eq(a, transmute(i64x2::new(0, 0)))
let b: i64x2 = i64x2::new(0, 0);
simd_eq(a, transmute(b))
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqz_p64(a: poly64x1_t) -> uint64x1_t {
simd_eq(a, transmute(i64x1::new(0)))
let b: i64x1 = i64x1::new(0);
simd_eq(a, transmute(b))
}
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqzq_p64(a: poly64x2_t) -> uint64x2_t {
simd_eq(a, transmute(i64x2::new(0, 0)))
let b: i64x2 = i64x2::new(0, 0);
simd_eq(a, transmute(b))
}
/// Unsigned Compare bitwise equal to zero
/// Unsigned compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqz_u8(a: uint8x8_t) -> uint8x8_t {
simd_eq(a, transmute(u8x8::new(0, 0, 0, 0, 0, 0, 0, 0)))
let b: u8x8 = u8x8::new(0, 0, 0, 0, 0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Unsigned Compare bitwise equal to zero
/// Unsigned compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqzq_u8(a: uint8x16_t) -> uint8x16_t {
simd_eq(a, transmute(u8x16::new(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)))
let b: u8x16 = u8x16::new(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Unsigned Compare bitwise equal to zero
/// Unsigned compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqz_u16(a: uint16x4_t) -> uint16x4_t {
simd_eq(a, transmute(u16x4::new(0, 0, 0, 0)))
let b: u16x4 = u16x4::new(0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Unsigned Compare bitwise equal to zero
/// Unsigned compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqzq_u16(a: uint16x8_t) -> uint16x8_t {
simd_eq(a, transmute(u16x8::new(0, 0, 0, 0, 0, 0, 0, 0)))
let b: u16x8 = u16x8::new(0, 0, 0, 0, 0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Unsigned Compare bitwise equal to zero
/// Unsigned compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqz_u32(a: uint32x2_t) -> uint32x2_t {
simd_eq(a, transmute(u32x2::new(0, 0)))
let b: u32x2 = u32x2::new(0, 0);
simd_eq(a, transmute(b))
}
/// Unsigned Compare bitwise equal to zero
/// Unsigned compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqzq_u32(a: uint32x4_t) -> uint32x4_t {
simd_eq(a, transmute(u32x4::new(0, 0, 0, 0)))
let b: u32x4 = u32x4::new(0, 0, 0, 0);
simd_eq(a, transmute(b))
}
/// Unsigned Compare bitwise equal to zero
/// Unsigned compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqz_u64(a: uint64x1_t) -> uint64x1_t {
simd_eq(a, transmute(u64x1::new(0)))
let b: u64x1 = u64x1::new(0);
simd_eq(a, transmute(b))
}
/// Unsigned Compare bitwise equal to zero
/// Unsigned compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(vceqz))]
#[cfg_attr(test, assert_instr(cmeq))]
pub unsafe fn vceqzq_u64(a: uint64x2_t) -> uint64x2_t {
simd_eq(a, transmute(u64x2::new(0, 0)))
let b: u64x2 = u64x2::new(0, 0);
simd_eq(a, transmute(b))
}
/// Floating-point compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(fcmeq))]
pub unsafe fn vceqz_f32(a: float32x2_t) -> uint32x2_t {
let b: f32x2 = f32x2::new(0.0, 0.0);
simd_eq(a, transmute(b))
}
/// Floating-point compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(fcmeq))]
pub unsafe fn vceqzq_f32(a: float32x4_t) -> uint32x4_t {
let b: f32x4 = f32x4::new(0.0, 0.0, 0.0, 0.0);
simd_eq(a, transmute(b))
}
/// Floating-point compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(fcmeq))]
pub unsafe fn vceqz_f64(a: float64x1_t) -> uint64x1_t {
let b: f64 = 0.0;
simd_eq(a, transmute(b))
}
/// Floating-point compare bitwise equal to zero
#[inline]
#[target_feature(enable = "neon")]
#[cfg_attr(test, assert_instr(fcmeq))]
pub unsafe fn vceqzq_f64(a: float64x2_t) -> uint64x2_t {
let b: f64x2 = f64x2::new(0.0, 0.0);
simd_eq(a, transmute(b))
}
/// Compare signed greater than
@ -752,6 +806,38 @@ mod test {
assert_eq!(r, e);
}
#[simd_test(enable = "neon")]
unsafe fn test_vceqz_f32() {
let a: f32x2 = f32x2::new(0.0, 1.2);
let e: u32x2 = u32x2::new(0xFF_FF_FF_FF, 0);
let r: u32x2 = transmute(vceqz_f32(transmute(a)));
assert_eq!(r, e);
}
#[simd_test(enable = "neon")]
unsafe fn test_vceqzq_f32() {
let a: f32x4 = f32x4::new(0.0, 1.2, 3.4, 5.6);
let e: u32x4 = u32x4::new(0xFF_FF_FF_FF, 0, 0, 0);
let r: u32x4 = transmute(vceqzq_f32(transmute(a)));
assert_eq!(r, e);
}
#[simd_test(enable = "neon")]
unsafe fn test_vceqz_f64() {
let a: f64 = 0.0;
let e: u64x1 = u64x1::new(0xFF_FF_FF_FF_FF_FF_FF_FF);
let r: u64x1 = transmute(vceqz_f64(transmute(a)));
assert_eq!(r, e);
}
#[simd_test(enable = "neon")]
unsafe fn test_vceqzq_f64() {
let a: f64x2 = f64x2::new(0.0, 1.2);
let e: u64x2 = u64x2::new(0xFF_FF_FF_FF_FF_FF_FF_FF, 0);
let r: u64x2 = transmute(vceqzq_f64(transmute(a)));
assert_eq!(r, e);
}
#[simd_test(enable = "neon")]
unsafe fn test_vcgt_s64() {
let a: i64x1 = i64x1::new(1);

View file

@ -136,26 +136,36 @@ arm = vceq.
// we are missing float16x4_t:uint16x4_t, float16x8_t:uint16x8_t
generate float32x2_t:uint32x2_t, float32x4_t:uint32x4_t
/// Signed Compare bitwise equal to zero
/// Signed compare bitwise equal to zero
name = vceqz
fn = simd_eq
a = MIN, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, MAX
fixed = 0
fixed = 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
validate FALSE, TRUE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE
aarch64 = vceqz
aarch64 = cmeq
generate int8x8_t:uint8x8_t, int8x16_t:uint8x16_t, int16x4_t:uint16x4_t, int16x8_t:uint16x8_t, int32x2_t:uint32x2_t, int32x4_t:uint32x4_t, int64x1_t:uint64x1_t, int64x2_t:uint64x2_t, poly64x1_t:uint64x1_t, poly64x2_t:uint64x2_t
/// Unsigned Compare bitwise equal to zero
/// Unsigned compare bitwise equal to zero
name = vceqz
fn = simd_eq
a = MIN, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, MAX
fixed = 0
fixed = 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
validate TRUE, TRUE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE
aarch64 = vceqz
aarch64 = cmeq
generate uint*_t, uint64x*_t
/// Floating-point compare bitwise equal to zero
name = vceqz
fn = simd_eq
a = 0.0, 1.2, 3.4, 5.6
fixed = 0.0, 0.0, 0.0, 0.0
validate TRUE, FALSE, FALSE, FALSE
aarch64 = fcmeq
generate float32x2_t:uint32x2_t, float32x4_t:uint32x4_t, float64x1_t:uint64x1_t, float64x2_t:uint64x2_t
////////////////////
// greater then
////////////////////

View file

@ -300,7 +300,7 @@ fn gen_aarch64(
out_t: &str,
current_tests: &[(Vec<String>, Vec<String>, Vec<String>)],
single_para: bool,
fixed: &Option<String>,
fixed: &Vec<String>,
) -> (String, String) {
let _global_t = type_to_global_type(in_t);
let _global_ret_t = type_to_global_type(out_t);
@ -342,23 +342,19 @@ fn gen_aarch64(
}}"#,
name, in_t, in_t, out_t, ext_c, current_fn,
)
} else if let Some(fixed_val) = fixed {
let mut fixed_vals = fixed_val.clone();
for _i in 1..type_len(in_t) {
fixed_vals.push_str(", ");
fixed_vals.push_str(fixed_val);
}
} else if fixed.len() != 0 {
let fixed: Vec<String> = fixed.iter().take(type_len(in_t)).cloned().collect();
format!(
r#"pub unsafe fn {}(a: {}) -> {} {{
{}{}(a, transmute({}::new({})))
let b{};
{}{}(a, transmute(b))
}}"#,
name,
in_t,
out_t,
values(in_t, &fixed),
ext_c,
current_fn,
type_to_global_type(in_t),
fixed_vals,
)
} else {
String::new()
@ -451,7 +447,7 @@ fn gen_arm(
out_t: &str,
current_tests: &[(Vec<String>, Vec<String>, Vec<String>)],
single_para: bool,
fixed: &Option<String>,
fixed: &Vec<String>,
) -> (String, String) {
let _global_t = type_to_global_type(in_t);
let _global_ret_t = type_to_global_type(out_t);
@ -506,23 +502,19 @@ fn gen_arm(
}}"#,
name, in_t, in_t, out_t, ext_c, current_fn,
)
} else if let Some(fixed_val) = fixed {
let mut fixed_vals = fixed_val.clone();
for _i in 1..type_len(in_t) {
fixed_vals.push_str(", ");
fixed_vals.push_str(fixed_val);
}
} else if fixed.len() != 0 {
let fixed: Vec<String> = fixed.iter().take(type_len(in_t)).cloned().collect();
format!(
r#"pub unsafe fn {}(a: {}) -> {} {{
{}{}(a, transmute({}::new({})))
let b{};
{}{}(a, transmute(b))
}}"#,
name,
in_t,
out_t,
values(in_t, &fixed),
ext_c,
current_fn,
type_to_global_type(in_t),
fixed_vals,
)
} else {
String::new()
@ -638,7 +630,8 @@ fn main() -> io::Result<()> {
let mut link_aarch64: Option<String> = None;
let mut a: Vec<String> = Vec::new();
let mut b: Vec<String> = Vec::new();
let mut fixed: Option<String> = None;
let mut fixed: Vec<String> = Vec::new();
let mut single_para: bool = true;
let mut current_tests: Vec<(Vec<String>, Vec<String>, Vec<String>)> = Vec::new();
//
@ -709,9 +702,7 @@ mod test {
link_aarch64 = None;
link_arm = None;
current_tests = Vec::new();
a = Vec::new();
b = Vec::new();
fixed = None;
single_para = true;
} else if line.starts_with("//") {
} else if line.starts_with("name = ") {
current_name = Some(String::from(&line[7..]));
@ -725,8 +716,9 @@ mod test {
a = line[4..].split(',').map(|v| v.trim().to_string()).collect();
} else if line.starts_with("b = ") {
b = line[4..].split(',').map(|v| v.trim().to_string()).collect();
single_para = false;
} else if line.starts_with("fixed = ") {
fixed = Some(String::from(&line[8..]));
fixed = line[8..].split(',').map(|v| v.trim().to_string()).collect();
} else if line.starts_with("validate ") {
let e = line[9..].split(',').map(|v| v.trim().to_string()).collect();
current_tests.push((a.clone(), b.clone(), e));
@ -778,7 +770,7 @@ mod test {
&in_t,
&out_t,
&current_tests,
b.len() == 0,
single_para,
&fixed,
);
out_arm.push_str(&function);
@ -793,7 +785,7 @@ mod test {
&in_t,
&out_t,
&current_tests,
b.len() == 0,
single_para,
&fixed,
);
out_aarch64.push_str(&function);