forked from OSchip/llvm-project
Remove more vector_shuffle patterns for unpack. These should be target specific nodes when they get to isel.
llvm-svn: 150363
This commit is contained in:
Craig Topper
parent
6d471c9e49
commit
74650add0e
|
|
@ -2265,11 +2265,6 @@ let Predicates = [HasAVX] in {
|
|||
(VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
|
||||
def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
|
||||
(VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
|
||||
// vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
|
||||
// fall back to this for SSE1)
|
||||
def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
|
||||
(VSHUFPSrri VR128:$src2, VR128:$src1,
|
||||
(SHUFFLE_get_shuf_imm VR128:$src3))>;
|
||||
|
||||
def : Pat<(v2i64 (X86Shufp VR128:$src1,
|
||||
(memopv2i64 addr:$src2), (i8 imm:$imm))),
|
||||
|
|
@ -2297,11 +2292,6 @@ let Predicates = [HasSSE1] in {
|
|||
(SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
|
||||
def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
|
||||
(SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
|
||||
// vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
|
||||
// fall back to this for SSE1)
|
||||
def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
|
||||
(SHUFPSrri VR128:$src2, VR128:$src1,
|
||||
(SHUFFLE_get_shuf_imm VR128:$src3))>;
|
||||
}
|
||||
|
||||
let Predicates = [HasSSE2] in {
|
||||
|
|
@ -2318,7 +2308,7 @@ let Predicates = [HasSSE2] in {
|
|||
//===----------------------------------------------------------------------===//
|
||||
|
||||
/// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
|
||||
multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
|
||||
multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
|
||||
PatFrag mem_frag, RegisterClass RC,
|
||||
X86MemOperand x86memop, string asm,
|
||||
Domain d> {
|
||||
|
|
@ -2335,86 +2325,48 @@ multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
|
|||
IIC_DEFAULT, d>;
|
||||
}
|
||||
|
||||
let AddedComplexity = 10 in {
|
||||
defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
|
||||
VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedSingle>, TB, VEX_4V;
|
||||
defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
|
||||
VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize, VEX_4V;
|
||||
defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
|
||||
VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedSingle>, TB, VEX_4V;
|
||||
defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
|
||||
VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize, VEX_4V;
|
||||
defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
|
||||
VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedSingle>, TB, VEX_4V;
|
||||
defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
|
||||
VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize, VEX_4V;
|
||||
defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
|
||||
VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedSingle>, TB, VEX_4V;
|
||||
defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
|
||||
VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize, VEX_4V;
|
||||
|
||||
defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
|
||||
VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedSingle>, TB, VEX_4V;
|
||||
defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
|
||||
VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize, VEX_4V;
|
||||
defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
|
||||
VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedSingle>, TB, VEX_4V;
|
||||
defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
|
||||
VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize, VEX_4V;
|
||||
defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
|
||||
VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedSingle>, TB, VEX_4V;
|
||||
defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
|
||||
VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize, VEX_4V;
|
||||
defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
|
||||
VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedSingle>, TB, VEX_4V;
|
||||
defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
|
||||
VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize, VEX_4V;
|
||||
|
||||
let Constraints = "$src1 = $dst" in {
|
||||
defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
|
||||
VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
|
||||
SSEPackedSingle>, TB;
|
||||
defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
|
||||
VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize;
|
||||
defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
|
||||
VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
|
||||
SSEPackedSingle>, TB;
|
||||
defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
|
||||
VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize;
|
||||
} // Constraints = "$src1 = $dst"
|
||||
} // AddedComplexity
|
||||
let Constraints = "$src1 = $dst" in {
|
||||
defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
|
||||
VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
|
||||
SSEPackedSingle>, TB;
|
||||
defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
|
||||
VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize;
|
||||
defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
|
||||
VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
|
||||
SSEPackedSingle>, TB;
|
||||
defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
|
||||
VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
|
||||
SSEPackedDouble>, TB, OpSize;
|
||||
} // Constraints = "$src1 = $dst"
|
||||
|
||||
let Predicates = [HasAVX], AddedComplexity = 1 in {
|
||||
def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
|
||||
(VUNPCKLPSrm VR128:$src1, addr:$src2)>;
|
||||
def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
|
||||
(VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
|
||||
def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
|
||||
(VUNPCKHPSrm VR128:$src1, addr:$src2)>;
|
||||
def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
|
||||
(VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
|
||||
|
||||
def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
|
||||
(VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
|
||||
def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
|
||||
(VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
|
||||
def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
|
||||
(VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
|
||||
def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
|
||||
(VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
|
||||
|
||||
def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
|
||||
(VUNPCKLPDrm VR128:$src1, addr:$src2)>;
|
||||
def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
|
||||
(VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
|
||||
def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
|
||||
(VUNPCKHPDrm VR128:$src1, addr:$src2)>;
|
||||
def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
|
||||
(VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
|
||||
|
||||
def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
|
||||
(VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
|
||||
def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
|
||||
(VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
|
||||
def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
|
||||
(VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
|
||||
def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
|
||||
(VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
|
||||
|
||||
// FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
|
||||
// problem is during lowering, where it's not possible to recognize the load
|
||||
// fold cause it has two uses through a bitcast. One use disappears at isel
|
||||
|
|
@ -2423,27 +2375,7 @@ let Predicates = [HasAVX], AddedComplexity = 1 in {
|
|||
(VUNPCKLPDrr VR128:$src, VR128:$src)>;
|
||||
}
|
||||
|
||||
let Predicates = [HasSSE1] in {
|
||||
def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
|
||||
(UNPCKLPSrm VR128:$src1, addr:$src2)>;
|
||||
def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
|
||||
(UNPCKLPSrr VR128:$src1, VR128:$src2)>;
|
||||
def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
|
||||
(UNPCKHPSrm VR128:$src1, addr:$src2)>;
|
||||
def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
|
||||
(UNPCKHPSrr VR128:$src1, VR128:$src2)>;
|
||||
}
|
||||
|
||||
let Predicates = [HasSSE2] in {
|
||||
def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
|
||||
(UNPCKLPDrm VR128:$src1, addr:$src2)>;
|
||||
def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
|
||||
(UNPCKLPDrr VR128:$src1, VR128:$src2)>;
|
||||
def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
|
||||
(UNPCKHPDrm VR128:$src1, addr:$src2)>;
|
||||
def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
|
||||
(UNPCKHPDrr VR128:$src1, VR128:$src2)>;
|
||||
|
||||
// FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
|
||||
// problem is during lowering, where it's not possible to recognize the load
|
||||
// fold cause it has two uses through a bitcast. One use disappears at isel
|
||||
|
|
|
|||
Loading…
Reference in New Issue