Reorganize logical and arithmetic SSE 1 & 2 instructions
llvm-svn: 106557
This commit is contained in:
Bruno Cardoso Lopes
parent
b91af24d3e
commit
3af915f84b
|
|
@ -834,6 +834,10 @@ def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
|
|||
[(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
|
||||
}
|
||||
|
||||
//===----------------------------------------------------------------------===//
|
||||
// SSE 1 & 2 - Logical Instructions
|
||||
//===----------------------------------------------------------------------===//
|
||||
|
||||
/// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
|
||||
///
|
||||
multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
|
||||
|
|
@ -868,6 +872,82 @@ defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
|
|||
let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
|
||||
defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef, 1>;
|
||||
|
||||
/// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
|
||||
///
|
||||
multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
|
||||
SDNode OpNode, int HasPat = 0,
|
||||
list<list<dag>> Pattern = []> {
|
||||
let isAsmParserOnly = 1 in {
|
||||
defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
|
||||
!strconcat(OpcodeStr, "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
|
||||
f128mem,
|
||||
!if(HasPat, Pattern[0], // rr
|
||||
[(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
|
||||
VR128:$src2)))]),
|
||||
!if(HasPat, Pattern[2], // rm
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
|
||||
(memopv2i64 addr:$src2)))])>,
|
||||
VEX_4V;
|
||||
|
||||
defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
|
||||
!strconcat(OpcodeStr, "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
|
||||
f128mem,
|
||||
!if(HasPat, Pattern[1], // rr
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
|
||||
(bc_v2i64 (v2f64
|
||||
VR128:$src2))))]),
|
||||
!if(HasPat, Pattern[3], // rm
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
|
||||
(memopv2i64 addr:$src2)))])>,
|
||||
OpSize, VEX_4V;
|
||||
}
|
||||
let Constraints = "$src1 = $dst" in {
|
||||
defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
|
||||
!strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"), f128mem,
|
||||
!if(HasPat, Pattern[0], // rr
|
||||
[(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
|
||||
VR128:$src2)))]),
|
||||
!if(HasPat, Pattern[2], // rm
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
|
||||
(memopv2i64 addr:$src2)))])>, TB;
|
||||
|
||||
defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
|
||||
!strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"), f128mem,
|
||||
!if(HasPat, Pattern[1], // rr
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
|
||||
(bc_v2i64 (v2f64
|
||||
VR128:$src2))))]),
|
||||
!if(HasPat, Pattern[3], // rm
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
|
||||
(memopv2i64 addr:$src2)))])>,
|
||||
TB, OpSize;
|
||||
}
|
||||
}
|
||||
|
||||
defm AND : sse12_fp_packed_logical<0x54, "and", and>;
|
||||
defm OR : sse12_fp_packed_logical<0x56, "or", or>;
|
||||
defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
|
||||
let isCommutable = 0 in
|
||||
defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
|
||||
// single r+r
|
||||
[(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
|
||||
(bc_v2i64 (v4i32 immAllOnesV))),
|
||||
VR128:$src2)))],
|
||||
// double r+r
|
||||
[(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
|
||||
(bc_v2i64 (v2f64 VR128:$src2))))],
|
||||
// single r+m
|
||||
[(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
|
||||
(bc_v2i64 (v4i32 immAllOnesV))),
|
||||
(memopv2i64 addr:$src2))))],
|
||||
// double r+m
|
||||
[(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
|
||||
(memopv2i64 addr:$src2)))]]>;
|
||||
|
||||
//===----------------------------------------------------------------------===//
|
||||
// SSE 1 & 2 - Arithmetic Instructions
|
||||
//===----------------------------------------------------------------------===//
|
||||
|
||||
/// basic_sse12_fp_binop_rm - SSE 1 & 2 binops come in both scalar and
|
||||
/// vector forms.
|
||||
///
|
||||
|
|
@ -876,8 +956,7 @@ let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
|
|||
/// plain scalar form, in that it takes an entire vector (instead of a scalar)
|
||||
/// and leaves the top elements unmodified (therefore these cannot be commuted).
|
||||
///
|
||||
/// These three forms can each be reg+reg or reg+mem, so there are a total of
|
||||
/// six "instructions".
|
||||
/// These three forms can each be reg+reg or reg+mem.
|
||||
///
|
||||
multiclass basic_sse12_fp_binop_rm<bits<8> opc, string OpcodeStr,
|
||||
SDNode OpNode> {
|
||||
|
|
@ -953,8 +1032,6 @@ let isCommutable = 0 in {
|
|||
/// onto C operators don't use this form since they just use the plain
|
||||
/// vector form instead of having a separate vector intrinsic form.
|
||||
///
|
||||
/// This provides a total of eight "instructions".
|
||||
///
|
||||
multiclass sse12_fp_binop_rm<bits<8> opc, string OpcodeStr,
|
||||
SDNode OpNode> {
|
||||
|
||||
|
|
@ -1210,79 +1287,6 @@ defm RSQRT : sse1_fp_unop_rm<0x52, "rsqrt", X86frsqrt,
|
|||
defm RCP : sse1_fp_unop_rm<0x53, "rcp", X86frcp,
|
||||
int_x86_sse_rcp_ss, int_x86_sse_rcp_ps>;
|
||||
|
||||
/// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
|
||||
///
|
||||
multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
|
||||
SDNode OpNode, int HasPat = 0,
|
||||
list<list<dag>> Pattern = []> {
|
||||
let isAsmParserOnly = 1 in {
|
||||
defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
|
||||
!strconcat(OpcodeStr, "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
|
||||
f128mem,
|
||||
!if(HasPat, Pattern[0], // rr
|
||||
[(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
|
||||
VR128:$src2)))]),
|
||||
!if(HasPat, Pattern[2], // rm
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
|
||||
(memopv2i64 addr:$src2)))])>,
|
||||
VEX_4V;
|
||||
|
||||
defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
|
||||
!strconcat(OpcodeStr, "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
|
||||
f128mem,
|
||||
!if(HasPat, Pattern[1], // rr
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
|
||||
(bc_v2i64 (v2f64
|
||||
VR128:$src2))))]),
|
||||
!if(HasPat, Pattern[3], // rm
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
|
||||
(memopv2i64 addr:$src2)))])>,
|
||||
OpSize, VEX_4V;
|
||||
}
|
||||
let Constraints = "$src1 = $dst" in {
|
||||
defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
|
||||
!strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"), f128mem,
|
||||
!if(HasPat, Pattern[0], // rr
|
||||
[(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
|
||||
VR128:$src2)))]),
|
||||
!if(HasPat, Pattern[2], // rm
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
|
||||
(memopv2i64 addr:$src2)))])>, TB;
|
||||
|
||||
defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
|
||||
!strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"), f128mem,
|
||||
!if(HasPat, Pattern[1], // rr
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
|
||||
(bc_v2i64 (v2f64
|
||||
VR128:$src2))))]),
|
||||
!if(HasPat, Pattern[3], // rm
|
||||
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
|
||||
(memopv2i64 addr:$src2)))])>,
|
||||
TB, OpSize;
|
||||
}
|
||||
}
|
||||
|
||||
// Logical
|
||||
defm AND : sse12_fp_packed_logical<0x54, "and", and>;
|
||||
defm OR : sse12_fp_packed_logical<0x56, "or", or>;
|
||||
defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
|
||||
let isCommutable = 0 in
|
||||
defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
|
||||
// single r+r
|
||||
[(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
|
||||
(bc_v2i64 (v4i32 immAllOnesV))),
|
||||
VR128:$src2)))],
|
||||
// double r+r
|
||||
[(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
|
||||
(bc_v2i64 (v2f64 VR128:$src2))))],
|
||||
// single r+m
|
||||
[(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
|
||||
(bc_v2i64 (v4i32 immAllOnesV))),
|
||||
(memopv2i64 addr:$src2))))],
|
||||
// double r+m
|
||||
[(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
|
||||
(memopv2i64 addr:$src2)))]]>;
|
||||
|
||||
let Constraints = "$src1 = $dst" in {
|
||||
def CMPPSrri : PSIi8<0xC2, MRMSrcReg,
|
||||
(outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc),
|
||||
|
|
|
|||
Loading…
Reference in New Issue