[SystemZ] Add support for IBM z14 processor (3/3)
This adds support for the new 128-bit vector float instructions of z14.
Note that these instructions actually only operate on the f128 type,
since only each 128-bit vector register can hold only one 128-bit
float value. However, this is still preferable to the legacy 128-bit
float instructions, since those operate on pairs of floating-point
registers (so we can hold at most 8 values in registers), while the
new instructions use single vector registers (so we hold up to 32
value in registers).
Adding support includes:
- Enabling the instructions for the assembler/disassembler.
- CodeGen for the instructions. This includes allocating the f128
type now to the VR128BitRegClass instead of FP128BitRegClass.
- Scheduler description support for the instructions.
Note that for a small number of operations, we have no new vector
instructions (like integer <-> 128-bit float conversions), and so
we use the legacy instruction and then reformat the operand
(i.e. copy between a pair of floating-point registers and a
vector register).
llvm-svn: 308196
diff --git a/llvm/lib/Target/SystemZ/SystemZInstrVector.td b/llvm/lib/Target/SystemZ/SystemZInstrVector.td
index 9d8f74c..c9a02d9 100644
--- a/llvm/lib/Target/SystemZ/SystemZInstrVector.td
+++ b/llvm/lib/Target/SystemZ/SystemZInstrVector.td
@@ -938,6 +938,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFASB : BinaryVRRc<"vfasb", 0xE7E3, fadd, v128sb, v128sb, 2, 0>;
def WFASB : BinaryVRRc<"wfasb", 0xE7E3, fadd, v32sb, v32sb, 2, 8>;
+ def WFAXB : BinaryVRRc<"wfaxb", 0xE7E3, fadd, v128xb, v128xb, 4, 8>;
}
// Convert from fixed 64-bit.
@@ -973,6 +974,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFDSB : BinaryVRRc<"vfdsb", 0xE7E5, fdiv, v128sb, v128sb, 2, 0>;
def WFDSB : BinaryVRRc<"wfdsb", 0xE7E5, fdiv, v32sb, v32sb, 2, 8>;
+ def WFDXB : BinaryVRRc<"wfdxb", 0xE7E5, fdiv, v128xb, v128xb, 4, 8>;
}
// Load FP integer.
@@ -984,8 +986,10 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFISB : TernaryVRRa<"vfisb", 0xE7C7, int_s390_vfisb, v128sb, v128sb, 2, 0>;
def WFISB : TernaryVRRa<"wfisb", 0xE7C7, null_frag, v32sb, v32sb, 2, 8>;
+ def WFIXB : TernaryVRRa<"wfixb", 0xE7C7, null_frag, v128xb, v128xb, 4, 8>;
defm : VectorRounding<VFISB, v128sb>;
defm : VectorRounding<WFISB, v32sb>;
+ defm : VectorRounding<WFIXB, v128xb>;
}
// Load lengthened.
@@ -998,6 +1002,9 @@
def VFLLS : UnaryVRRa<"vflls", 0xE7C4, null_frag, v128db, v128sb, 2, 0>;
def WFLLS : UnaryVRRa<"wflls", 0xE7C4, null_frag, v64db, v32sb, 2, 8>;
}
+ def WFLLD : UnaryVRRa<"wflld", 0xE7C4, fpextend, v128xb, v64db, 3, 8>;
+ def : Pat<(f128 (fpextend (f32 VR32:$src))),
+ (WFLLD (WLDEB VR32:$src))>;
}
// Load rounded.
@@ -1012,6 +1019,10 @@
def VFLRD : TernaryVRRa<"vflrd", 0xE7C5, null_frag, v128sb, v128db, 3, 0>;
def WFLRD : TernaryVRRa<"wflrd", 0xE7C5, null_frag, v32sb, v64db, 3, 8>;
}
+ def WFLRX : TernaryVRRa<"wflrx", 0xE7C5, null_frag, v64db, v128xb, 4, 8>;
+ def : FPConversion<WFLRX, fpround, v64db, v128xb, 0, 0>;
+ def : Pat<(f32 (fpround (f128 VR128:$src))),
+ (WLEDB (WFLRX VR128:$src, 0, 3), 0, 0)>;
}
// Maximum.
@@ -1029,10 +1040,13 @@
v128sb, v128sb, 2, 0>;
def WFMAXSB : TernaryVRRcFloat<"wfmaxsb", 0xE7EF, null_frag,
v32sb, v32sb, 2, 8>;
+ def WFMAXXB : TernaryVRRcFloat<"wfmaxxb", 0xE7EF, null_frag,
+ v128xb, v128xb, 4, 8>;
defm : VectorMax<VFMAXDB, v128db>;
defm : VectorMax<WFMAXDB, v64db>;
defm : VectorMax<VFMAXSB, v128sb>;
defm : VectorMax<WFMAXSB, v32sb>;
+ defm : VectorMax<WFMAXXB, v128xb>;
}
// Minimum.
@@ -1050,10 +1064,13 @@
v128sb, v128sb, 2, 0>;
def WFMINSB : TernaryVRRcFloat<"wfminsb", 0xE7EE, null_frag,
v32sb, v32sb, 2, 8>;
+ def WFMINXB : TernaryVRRcFloat<"wfminxb", 0xE7EE, null_frag,
+ v128xb, v128xb, 4, 8>;
defm : VectorMin<VFMINDB, v128db>;
defm : VectorMin<WFMINDB, v64db>;
defm : VectorMin<VFMINSB, v128sb>;
defm : VectorMin<WFMINSB, v32sb>;
+ defm : VectorMin<WFMINXB, v128xb>;
}
// Multiply.
@@ -1063,6 +1080,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFMSB : BinaryVRRc<"vfmsb", 0xE7E7, fmul, v128sb, v128sb, 2, 0>;
def WFMSB : BinaryVRRc<"wfmsb", 0xE7E7, fmul, v32sb, v32sb, 2, 8>;
+ def WFMXB : BinaryVRRc<"wfmxb", 0xE7E7, fmul, v128xb, v128xb, 4, 8>;
}
// Multiply and add.
@@ -1072,6 +1090,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFMASB : TernaryVRRe<"vfmasb", 0xE78F, fma, v128sb, v128sb, 0, 2>;
def WFMASB : TernaryVRRe<"wfmasb", 0xE78F, fma, v32sb, v32sb, 8, 2>;
+ def WFMAXB : TernaryVRRe<"wfmaxb", 0xE78F, fma, v128xb, v128xb, 8, 4>;
}
// Multiply and subtract.
@@ -1081,6 +1100,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFMSSB : TernaryVRRe<"vfmssb", 0xE78E, fms, v128sb, v128sb, 0, 2>;
def WFMSSB : TernaryVRRe<"wfmssb", 0xE78E, fms, v32sb, v32sb, 8, 2>;
+ def WFMSXB : TernaryVRRe<"wfmsxb", 0xE78E, fms, v128xb, v128xb, 8, 4>;
}
// Negative multiply and add.
@@ -1090,6 +1110,7 @@
def WFNMADB : TernaryVRRe<"wfnmadb", 0xE79F, fnma, v64db, v64db, 8, 3>;
def VFNMASB : TernaryVRRe<"vfnmasb", 0xE79F, fnma, v128sb, v128sb, 0, 2>;
def WFNMASB : TernaryVRRe<"wfnmasb", 0xE79F, fnma, v32sb, v32sb, 8, 2>;
+ def WFNMAXB : TernaryVRRe<"wfnmaxb", 0xE79F, fnma, v128xb, v128xb, 8, 4>;
}
// Negative multiply and subtract.
@@ -1099,6 +1120,7 @@
def WFNMSDB : TernaryVRRe<"wfnmsdb", 0xE79E, fnms, v64db, v64db, 8, 3>;
def VFNMSSB : TernaryVRRe<"vfnmssb", 0xE79E, fnms, v128sb, v128sb, 0, 2>;
def WFNMSSB : TernaryVRRe<"wfnmssb", 0xE79E, fnms, v32sb, v32sb, 8, 2>;
+ def WFNMSXB : TernaryVRRe<"wfnmsxb", 0xE79E, fnms, v128xb, v128xb, 8, 4>;
}
// Perform sign operation.
@@ -1108,6 +1130,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFPSOSB : BinaryVRRa<"vfpsosb", 0xE7CC, null_frag, v128sb, v128sb, 2, 0>;
def WFPSOSB : BinaryVRRa<"wfpsosb", 0xE7CC, null_frag, v32sb, v32sb, 2, 8>;
+ def WFPSOXB : BinaryVRRa<"wfpsoxb", 0xE7CC, null_frag, v128xb, v128xb, 4, 8>;
}
// Load complement.
@@ -1116,6 +1139,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFLCSB : UnaryVRRa<"vflcsb", 0xE7CC, fneg, v128sb, v128sb, 2, 0, 0>;
def WFLCSB : UnaryVRRa<"wflcsb", 0xE7CC, fneg, v32sb, v32sb, 2, 8, 0>;
+ def WFLCXB : UnaryVRRa<"wflcxb", 0xE7CC, fneg, v128xb, v128xb, 4, 8, 0>;
}
// Load negative.
@@ -1124,6 +1148,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFLNSB : UnaryVRRa<"vflnsb", 0xE7CC, fnabs, v128sb, v128sb, 2, 0, 1>;
def WFLNSB : UnaryVRRa<"wflnsb", 0xE7CC, fnabs, v32sb, v32sb, 2, 8, 1>;
+ def WFLNXB : UnaryVRRa<"wflnxb", 0xE7CC, fnabs, v128xb, v128xb, 4, 8, 1>;
}
// Load positive.
@@ -1132,6 +1157,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFLPSB : UnaryVRRa<"vflpsb", 0xE7CC, fabs, v128sb, v128sb, 2, 0, 2>;
def WFLPSB : UnaryVRRa<"wflpsb", 0xE7CC, fabs, v32sb, v32sb, 2, 8, 2>;
+ def WFLPXB : UnaryVRRa<"wflpxb", 0xE7CC, fabs, v128xb, v128xb, 4, 8, 2>;
}
// Square root.
@@ -1141,6 +1167,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFSQSB : UnaryVRRa<"vfsqsb", 0xE7CE, fsqrt, v128sb, v128sb, 2, 0>;
def WFSQSB : UnaryVRRa<"wfsqsb", 0xE7CE, fsqrt, v32sb, v32sb, 2, 8>;
+ def WFSQXB : UnaryVRRa<"wfsqxb", 0xE7CE, fsqrt, v128xb, v128xb, 4, 8>;
}
// Subtract.
@@ -1150,6 +1177,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFSSB : BinaryVRRc<"vfssb", 0xE7E2, fsub, v128sb, v128sb, 2, 0>;
def WFSSB : BinaryVRRc<"wfssb", 0xE7E2, fsub, v32sb, v32sb, 2, 8>;
+ def WFSXB : BinaryVRRc<"wfsxb", 0xE7E2, fsub, v128xb, v128xb, 4, 8>;
}
// Test data class immediate.
@@ -1160,6 +1188,7 @@
let Predicates = [FeatureVectorEnhancements1] in {
def VFTCISB : BinaryVRIe<"vftcisb", 0xE74A, z_vftci, v128f, v128sb, 2, 0>;
def WFTCISB : BinaryVRIe<"wftcisb", 0xE74A, null_frag, v32f, v32sb, 2, 8>;
+ def WFTCIXB : BinaryVRIe<"wftcixb", 0xE74A, null_frag, v128q, v128xb, 4, 8>;
}
}
}
@@ -1175,6 +1204,7 @@
def WFCDB : CompareVRRa<"wfcdb", 0xE7CB, z_fcmp, v64db, 3>;
let Predicates = [FeatureVectorEnhancements1] in {
def WFCSB : CompareVRRa<"wfcsb", 0xE7CB, z_fcmp, v32sb, 2>;
+ def WFCXB : CompareVRRa<"wfcxb", 0xE7CB, z_fcmp, v128xb, 4>;
}
}
@@ -1184,6 +1214,7 @@
def WFKDB : CompareVRRa<"wfkdb", 0xE7CA, null_frag, v64db, 3>;
let Predicates = [FeatureVectorEnhancements1] in {
def WFKSB : CompareVRRa<"wfksb", 0xE7CA, null_frag, v32sb, 2>;
+ def WFKXB : CompareVRRa<"wfkxb", 0xE7CA, null_frag, v128xb, 4>;
}
}
@@ -1198,6 +1229,8 @@
v128f, v128sb, 2, 0>;
defm WFCESB : BinaryVRRcSPair<"wfcesb", 0xE7E8, null_frag, null_frag,
v32f, v32sb, 2, 8>;
+ defm WFCEXB : BinaryVRRcSPair<"wfcexb", 0xE7E8, null_frag, null_frag,
+ v128q, v128xb, 4, 8>;
}
// Compare and signal equal.
@@ -1210,6 +1243,8 @@
v128f, v128sb, 2, 4>;
defm WFKESB : BinaryVRRcSPair<"wfkesb", 0xE7E8, null_frag, null_frag,
v32f, v32sb, 2, 12>;
+ defm WFKEXB : BinaryVRRcSPair<"wfkexb", 0xE7E8, null_frag, null_frag,
+ v128q, v128xb, 4, 12>;
}
// Compare high.
@@ -1223,6 +1258,8 @@
v128f, v128sb, 2, 0>;
defm WFCHSB : BinaryVRRcSPair<"wfchsb", 0xE7EB, null_frag, null_frag,
v32f, v32sb, 2, 8>;
+ defm WFCHXB : BinaryVRRcSPair<"wfchxb", 0xE7EB, null_frag, null_frag,
+ v128q, v128xb, 4, 8>;
}
// Compare and signal high.
@@ -1235,6 +1272,8 @@
v128f, v128sb, 2, 4>;
defm WFKHSB : BinaryVRRcSPair<"wfkhsb", 0xE7EB, null_frag, null_frag,
v32f, v32sb, 2, 12>;
+ defm WFKHXB : BinaryVRRcSPair<"wfkhxb", 0xE7EB, null_frag, null_frag,
+ v128q, v128xb, 4, 12>;
}
// Compare high or equal.
@@ -1248,6 +1287,8 @@
v128f, v128sb, 2, 0>;
defm WFCHESB : BinaryVRRcSPair<"wfchesb", 0xE7EA, null_frag, null_frag,
v32f, v32sb, 2, 8>;
+ defm WFCHEXB : BinaryVRRcSPair<"wfchexb", 0xE7EA, null_frag, null_frag,
+ v128q, v128xb, 4, 8>;
}
// Compare and signal high or equal.
@@ -1260,6 +1301,8 @@
v128f, v128sb, 2, 4>;
defm WFKHESB : BinaryVRRcSPair<"wfkhesb", 0xE7EA, null_frag, null_frag,
v32f, v32sb, 2, 12>;
+ defm WFKHEXB : BinaryVRRcSPair<"wfkhexb", 0xE7EA, null_frag, null_frag,
+ v128q, v128xb, 4, 12>;
}
}
@@ -1272,36 +1315,49 @@
def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
+def : Pat<(v16i8 (bitconvert (f128 VR128:$src))), (v16i8 VR128:$src)>;
def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
+def : Pat<(v8i16 (bitconvert (f128 VR128:$src))), (v8i16 VR128:$src)>;
def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
+def : Pat<(v4i32 (bitconvert (f128 VR128:$src))), (v4i32 VR128:$src)>;
def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
+def : Pat<(v2i64 (bitconvert (f128 VR128:$src))), (v2i64 VR128:$src)>;
def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
+def : Pat<(v4f32 (bitconvert (f128 VR128:$src))), (v4f32 VR128:$src)>;
def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
+def : Pat<(v2f64 (bitconvert (f128 VR128:$src))), (v2f64 VR128:$src)>;
+
+def : Pat<(f128 (bitconvert (v16i8 VR128:$src))), (f128 VR128:$src)>;
+def : Pat<(f128 (bitconvert (v8i16 VR128:$src))), (f128 VR128:$src)>;
+def : Pat<(f128 (bitconvert (v4i32 VR128:$src))), (f128 VR128:$src)>;
+def : Pat<(f128 (bitconvert (v2i64 VR128:$src))), (f128 VR128:$src)>;
+def : Pat<(f128 (bitconvert (v4f32 VR128:$src))), (f128 VR128:$src)>;
+def : Pat<(f128 (bitconvert (v2f64 VR128:$src))), (f128 VR128:$src)>;
//===----------------------------------------------------------------------===//
// Replicating scalars
@@ -1378,6 +1434,20 @@
}
//===----------------------------------------------------------------------===//
+// Support for 128-bit floating-point values in vector registers
+//===----------------------------------------------------------------------===//
+
+let Predicates = [FeatureVectorEnhancements1] in {
+ def : Pat<(f128 (load bdxaddr12only:$addr)),
+ (VL bdxaddr12only:$addr)>;
+ def : Pat<(store (f128 VR128:$src), bdxaddr12only:$addr),
+ (VST VR128:$src, bdxaddr12only:$addr)>;
+
+ def : Pat<(f128 fpimm0), (VZERO)>;
+ def : Pat<(f128 fpimmneg0), (WFLNXB (VZERO))>;
+}
+
+//===----------------------------------------------------------------------===//
// String instructions
//===----------------------------------------------------------------------===//