Add x86 SIMD band_not

This commit is contained in:
Andrew Brown
2019-10-08 20:32:52 -07:00
parent fad6bb1a5c
commit 8f74333662
4 changed files with 36 additions and 0 deletions

View File

@@ -1982,6 +1982,10 @@ pub(crate) fn define<'defs>(
let band = band.bind(vector(ty, sse_vector_size)); let band = band.bind(vector(ty, sse_vector_size));
e.enc_32_64(band, rec_fa.opcodes(&PAND)); e.enc_32_64(band, rec_fa.opcodes(&PAND));
// and not (note flipped recipe operands to match band_not order)
let band_not = band_not.bind(vector(ty, sse_vector_size));
e.enc_32_64(band_not, rec_fax.opcodes(&PANDN));
// or // or
let bor = bor.bind(vector(ty, sse_vector_size)); let bor = bor.bind(vector(ty, sse_vector_size));
e.enc_32_64(bor, rec_fa.opcodes(&POR)); e.enc_32_64(bor, rec_fa.opcodes(&POR));

View File

@@ -266,6 +266,9 @@ pub static PADDUSW: [u8; 3] = [0x66, 0x0f, 0xdd];
/// Bitwise AND of xmm2/m128 and xmm1 (SSE2). /// Bitwise AND of xmm2/m128 and xmm1 (SSE2).
pub static PAND: [u8; 3] = [0x66, 0x0f, 0xdb]; pub static PAND: [u8; 3] = [0x66, 0x0f, 0xdb];
/// Bitwise AND NOT of xmm2/m128 and xmm1 (SSE2).
pub static PANDN: [u8; 3] = [0x66, 0x0f, 0xdf];
/// Compare packed data for equal (SSE2). /// Compare packed data for equal (SSE2).
pub static PCMPEQB: [u8; 3] = [0x66, 0x0f, 0x74]; pub static PCMPEQB: [u8; 3] = [0x66, 0x0f, 0x74];

View File

@@ -19,3 +19,9 @@ ebb0(v0: b32x4 [%xmm4], v1: b32x4 [%xmm0]):
[-, %xmm4] v2 = bxor v0, v1 ; bin: 66 0f ef e0 [-, %xmm4] v2 = bxor v0, v1 ; bin: 66 0f ef e0
return v2 return v2
} }
function %band_not_b64x2(b64x2, b64x2) -> b64x2 {
ebb0(v0: b64x2 [%xmm6], v1: b64x2 [%xmm3]):
[-, %xmm3] v2 = band_not v0, v1 ; bin: 66 0f df de
return v2
}

View File

@@ -0,0 +1,23 @@
test run
set enable_simd
target x86_64 skylake
function %bnot() -> b32 {
ebb0:
v0 = vconst.b32x4 [true true true false]
v1 = bnot v0
v2 = extractlane v1, 3
return v2
}
; run
function %band_not() -> b1 {
ebb0:
v0 = vconst.i16x8 [1 0 0 0 0 0 0 0]
v1 = vconst.i16x8 [0 0 0 0 0 0 0 0]
v2 = band_not v0, v1
v3 = extractlane v2, 0
v4 = icmp_imm eq v3, 1
return v4
}
; run