RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 03:33:20 +01:00
Code Issues Projects Releases Wiki Activity

[AVX512] Two new attributes in X86VectorVTInfo for subvector insert

Browse Source 
The new attributes are NumElts and the CD8TupleForm.  This prepares the code
to enable x8 and x2 inserts.

NFC, no change in X86.td.expanded except for the new attributes.

llvm-svn: 219871
This commit is contained in:
Adam Nemet 2014-10-15 23:42:09 +00:00
parent ea53faaf7d
commit a3b1e47840
2 changed files with 14 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
lib/Target/X86/X86InstrAVX512.td
View File

@ -2,9 +2,10 @@
// EltVT). These are things like the register class for the writemask, etc. // EltVT). These are things like the register class for the writemask, etc.
// The idea is to pass one of these as the template argument rather than the // The idea is to pass one of these as the template argument rather than the
// individual arguments. // individual arguments.
class X86VectorVTInfo<int NumElts, ValueType EltVT, RegisterClass rc, class X86VectorVTInfo<int numelts, ValueType EltVT, RegisterClass rc,
string suffix = ""> { string suffix = ""> {
RegisterClass RC = rc; RegisterClass RC = rc;
int NumElts = numelts;
// Corresponding mask register class. // Corresponding mask register class.
RegisterClass KRC = !cast<RegisterClass>("VK" # NumElts); RegisterClass KRC = !cast<RegisterClass>("VK" # NumElts);
@ -73,6 +74,11 @@ class X86VectorVTInfo<int NumElts, ValueType EltVT, RegisterClass rc,
// The string to specify embedded broadcast in assembly. // The string to specify embedded broadcast in assembly.
string BroadcastStr = "{1to" # NumElts # "}"; string BroadcastStr = "{1to" # NumElts # "}";
// 8-bit compressed displacement tuple/subvector format. This is only
// defined for NumElts <= 8.
CD8VForm CD8TupleForm = !if (!eq (!srl(NumElts, 4), 0),
!cast<CD8VForm>("CD8VT" # NumElts), ?);
SubRegIndex SubRegIdx = !if (!eq (Size, 128), sub_xmm, SubRegIndex SubRegIdx = !if (!eq (Size, 128), sub_xmm,
!if (!eq (Size, 256), sub_ymm, ?)); !if (!eq (Size, 256), sub_ymm, ?));
@ -349,7 +355,8 @@ multiclass vinsert_for_size<int Opcode,
let hasSideEffects = 0, ExeDomain = To.ExeDomain in { let hasSideEffects = 0, ExeDomain = To.ExeDomain in {
def rr : AVX512AIi8<Opcode, MRMSrcReg, (outs VR512:$dst), def rr : AVX512AIi8<Opcode, MRMSrcReg, (outs VR512:$dst),
(ins VR512:$src1, From.RC:$src2, i8imm:$src3), (ins VR512:$src1, From.RC:$src2, i8imm:$src3),
"vinsert" # From.EltTypeName # "x4\t{$src3, $src2, $src1, $dst|" "vinsert" # From.EltTypeName # "x" # From.NumElts #
"\t{$src3, $src2, $src1, $dst|"
"$dst, $src1, $src2, $src3}", "$dst, $src1, $src2, $src3}",
[(set To.RC:$dst, (vinsert_insert:$src3 (To.VT VR512:$src1), [(set To.RC:$dst, (vinsert_insert:$src3 (To.VT VR512:$src1),
(From.VT From.RC:$src2), (From.VT From.RC:$src2),
@ -359,9 +366,11 @@ multiclass vinsert_for_size<int Opcode,
let mayLoad = 1 in let mayLoad = 1 in
def rm : AVX512AIi8<Opcode, MRMSrcMem, (outs VR512:$dst), def rm : AVX512AIi8<Opcode, MRMSrcMem, (outs VR512:$dst),
(ins VR512:$src1, From.MemOp:$src2, i8imm:$src3), (ins VR512:$src1, From.MemOp:$src2, i8imm:$src3),
"vinsert" # From.EltTypeName # "x4\t{$src3, $src2, $src1, $dst|" "vinsert" # From.EltTypeName # "x" # From.NumElts #
"\t{$src3, $src2, $src1, $dst|"
"$dst, $src1, $src2, $src3}", "$dst, $src1, $src2, $src3}",
[]>, EVEX_4V, EVEX_V512, EVEX_CD8<From.EltSize, CD8VT4>; []>,
EVEX_4V, EVEX_V512, EVEX_CD8<From.EltSize, From.CD8TupleForm>;
} }
// Codegen pattern with the alternative types, e.g. v2i64 -> v8i64 for // Codegen pattern with the alternative types, e.g. v2i64 -> v8i64 for

1
lib/Target/X86/X86InstrFormats.td
View File

@ -101,6 +101,7 @@ def CD8VF : CD8VForm<0>; // v := VL
def CD8VH : CD8VForm<1>; // v := VL/2 def CD8VH : CD8VForm<1>; // v := VL/2
def CD8VQ : CD8VForm<2>; // v := VL/4 def CD8VQ : CD8VForm<2>; // v := VL/4
def CD8VO : CD8VForm<3>; // v := VL/8 def CD8VO : CD8VForm<3>; // v := VL/8
// The tuple (subvector) forms.
def CD8VT1 : CD8VForm<4>; // v := 1 def CD8VT1 : CD8VForm<4>; // v := 1
def CD8VT2 : CD8VForm<5>; // v := 2 def CD8VT2 : CD8VForm<5>; // v := 2
def CD8VT4 : CD8VForm<6>; // v := 4 def CD8VT4 : CD8VForm<6>; // v := 4