//===----------------------------------------------------------------------===//
// Load zero.
-let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isMoveImm = 1 in {
+let hasSideEffects = 0, isAsCheapAsAMove = 1, isMoveImm = 1 in {
def LZER : InherentRRE<"lzer", 0xB374, FP32, (fpimm0)>;
def LZDR : InherentRRE<"lzdr", 0xB375, FP64, (fpimm0)>;
def LZXR : InherentRRE<"lzxr", 0xB376, FP128, (fpimm0)>;
}
// Moves between two floating-point registers.
-let neverHasSideEffects = 1 in {
+let hasSideEffects = 0 in {
def LER : UnaryRR <"le", 0x38, null_frag, FP32, FP32>;
def LDR : UnaryRR <"ld", 0x28, null_frag, FP64, FP64>;
def LXR : UnaryRRE<"lx", 0xB365, null_frag, FP128, FP128>;
defm LTDBR : LoadAndTestRRE<"ltdb", 0xB312, FP64>;
defm LTXBR : LoadAndTestRRE<"ltxb", 0xB342, FP128>;
}
-def : CompareZeroFP<LTEBRCompare, FP32>;
-def : CompareZeroFP<LTDBRCompare, FP64>;
-def : CompareZeroFP<LTXBRCompare, FP128>;
+// Note that the comparison against zero operation is not available if we
+// have vector support, since load-and-test instructions will partially
+// clobber the target (vector) register.
+let Predicates = [FeatureNoVector] in {
+ defm : CompareZeroFP<LTEBRCompare, FP32>;
+ defm : CompareZeroFP<LTDBRCompare, FP64>;
+ defm : CompareZeroFP<LTXBRCompare, FP128>;
+}
// Moves between 64-bit integer and floating-point registers.
def LGDR : UnaryRRE<"lgd", 0xB3CD, bitconvert, GR64, FP64>;
// The sign of an FP128 is in the high register.
def : Pat<(fcopysign FP32:$src1, FP128:$src2),
- (CPSDRsd FP32:$src1, (EXTRACT_SUBREG FP128:$src2, subreg_high))>;
+ (CPSDRsd FP32:$src1, (EXTRACT_SUBREG FP128:$src2, subreg_h64))>;
// fcopysign with an FP64 result.
let isCodeGenOnly = 1 in
// The sign of an FP128 is in the high register.
def : Pat<(fcopysign FP64:$src1, FP128:$src2),
- (CPSDRdd FP64:$src1, (EXTRACT_SUBREG FP128:$src2, subreg_high))>;
+ (CPSDRdd FP64:$src1, (EXTRACT_SUBREG FP128:$src2, subreg_h64))>;
// fcopysign with an FP128 result. Use "upper" as the high half and leave
// the low half as-is.
class CopySign128<RegisterOperand cls, dag upper>
: Pat<(fcopysign FP128:$src1, cls:$src2),
- (INSERT_SUBREG FP128:$src1, upper, subreg_high)>;
+ (INSERT_SUBREG FP128:$src1, upper, subreg_h64)>;
-def : CopySign128<FP32, (CPSDRds (EXTRACT_SUBREG FP128:$src1, subreg_high),
+def : CopySign128<FP32, (CPSDRds (EXTRACT_SUBREG FP128:$src1, subreg_h64),
FP32:$src2)>;
-def : CopySign128<FP64, (CPSDRdd (EXTRACT_SUBREG FP128:$src1, subreg_high),
+def : CopySign128<FP64, (CPSDRdd (EXTRACT_SUBREG FP128:$src1, subreg_h64),
FP64:$src2)>;
-def : CopySign128<FP128, (CPSDRdd (EXTRACT_SUBREG FP128:$src1, subreg_high),
- (EXTRACT_SUBREG FP128:$src2, subreg_high))>;
+def : CopySign128<FP128, (CPSDRdd (EXTRACT_SUBREG FP128:$src1, subreg_h64),
+ (EXTRACT_SUBREG FP128:$src2, subreg_h64))>;
-defm LoadStoreF32 : MVCLoadStore<load, store, f32, MVCWrapper, 4>;
-defm LoadStoreF64 : MVCLoadStore<load, store, f64, MVCWrapper, 8>;
-defm LoadStoreF128 : MVCLoadStore<load, store, f128, MVCWrapper, 16>;
+defm LoadStoreF32 : MVCLoadStore<load, f32, MVCSequence, 4>;
+defm LoadStoreF64 : MVCLoadStore<load, f64, MVCSequence, 8>;
+defm LoadStoreF128 : MVCLoadStore<load, f128, MVCSequence, 16>;
//===----------------------------------------------------------------------===//
// Load instructions
defm LE : UnaryRXPair<"le", 0x78, 0xED64, load, FP32, 4>;
defm LD : UnaryRXPair<"ld", 0x68, 0xED65, load, FP64, 8>;
+ // For z13 we prefer LDE over LE to avoid partial register dependencies.
+ def LDE32 : UnaryRXE<"lde", 0xED24, null_frag, FP32, 4>;
+
// These instructions are split after register allocation, so we don't
// want a custom inserter.
let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
def LEXBR : UnaryRRE<"lexb", 0xB346, null_frag, FP128, FP128>;
def LDXBR : UnaryRRE<"ldxb", 0xB345, null_frag, FP128, FP128>;
+def LEDBRA : UnaryRRF4<"ledbra", 0xB344, FP32, FP64>,
+ Requires<[FeatureFPExtension]>;
+def LEXBRA : UnaryRRF4<"lexbra", 0xB346, FP128, FP128>,
+ Requires<[FeatureFPExtension]>;
+def LDXBRA : UnaryRRF4<"ldxbra", 0xB345, FP128, FP128>,
+ Requires<[FeatureFPExtension]>;
+
def : Pat<(f32 (fround FP128:$src)),
- (EXTRACT_SUBREG (LEXBR FP128:$src), subreg_32bit)>;
+ (EXTRACT_SUBREG (LEXBR FP128:$src), subreg_hr32)>;
def : Pat<(f64 (fround FP128:$src)),
- (EXTRACT_SUBREG (LDXBR FP128:$src), subreg_high)>;
+ (EXTRACT_SUBREG (LDXBR FP128:$src), subreg_h64)>;
// Extend register floating-point values to wider representations.
def LDEBR : UnaryRRE<"ldeb", 0xB304, fextend, FP64, FP32>;
def CDGBR : UnaryRRE<"cdgb", 0xB3A5, sint_to_fp, FP64, GR64>;
def CXGBR : UnaryRRE<"cxgb", 0xB3A6, sint_to_fp, FP128, GR64>;
+// Convert am unsigned integer register value to a floating-point one.
+let Predicates = [FeatureFPExtension] in {
+ def CELFBR : UnaryRRF4<"celfbr", 0xB390, FP32, GR32>;
+ def CDLFBR : UnaryRRF4<"cdlfbr", 0xB391, FP64, GR32>;
+ def CXLFBR : UnaryRRF4<"cxlfbr", 0xB392, FP128, GR32>;
+
+ def CELGBR : UnaryRRF4<"celgbr", 0xB3A0, FP32, GR64>;
+ def CDLGBR : UnaryRRF4<"cdlgbr", 0xB3A1, FP64, GR64>;
+ def CXLGBR : UnaryRRF4<"cxlgbr", 0xB3A2, FP128, GR64>;
+
+ def : Pat<(f32 (uint_to_fp GR32:$src)), (CELFBR 0, GR32:$src, 0)>;
+ def : Pat<(f64 (uint_to_fp GR32:$src)), (CDLFBR 0, GR32:$src, 0)>;
+ def : Pat<(f128 (uint_to_fp GR32:$src)), (CXLFBR 0, GR32:$src, 0)>;
+
+ def : Pat<(f32 (uint_to_fp GR64:$src)), (CELGBR 0, GR64:$src, 0)>;
+ def : Pat<(f64 (uint_to_fp GR64:$src)), (CDLGBR 0, GR64:$src, 0)>;
+ def : Pat<(f128 (uint_to_fp GR64:$src)), (CXLGBR 0, GR64:$src, 0)>;
+}
+
// Convert a floating-point register value to a signed integer value,
// with the second operand (modifier M3) specifying the rounding mode.
let Defs = [CC] in {
def : Pat<(i64 (fp_to_sint FP64:$src)), (CGDBR 5, FP64:$src)>;
def : Pat<(i64 (fp_to_sint FP128:$src)), (CGXBR 5, FP128:$src)>;
+// Convert a floating-point register value to an unsigned integer value.
+let Predicates = [FeatureFPExtension] in {
+ let Defs = [CC] in {
+ def CLFEBR : UnaryRRF4<"clfebr", 0xB39C, GR32, FP32>;
+ def CLFDBR : UnaryRRF4<"clfdbr", 0xB39D, GR32, FP64>;
+ def CLFXBR : UnaryRRF4<"clfxbr", 0xB39E, GR32, FP128>;
+
+ def CLGEBR : UnaryRRF4<"clgebr", 0xB3AC, GR64, FP32>;
+ def CLGDBR : UnaryRRF4<"clgdbr", 0xB3AD, GR64, FP64>;
+ def CLGXBR : UnaryRRF4<"clgxbr", 0xB3AE, GR64, FP128>;
+ }
+
+ def : Pat<(i32 (fp_to_uint FP32:$src)), (CLFEBR 5, FP32:$src, 0)>;
+ def : Pat<(i32 (fp_to_uint FP64:$src)), (CLFDBR 5, FP64:$src, 0)>;
+ def : Pat<(i32 (fp_to_uint FP128:$src)), (CLFXBR 5, FP128:$src, 0)>;
+
+ def : Pat<(i64 (fp_to_uint FP32:$src)), (CLGEBR 5, FP32:$src, 0)>;
+ def : Pat<(i64 (fp_to_uint FP64:$src)), (CLGDBR 5, FP64:$src, 0)>;
+ def : Pat<(i64 (fp_to_uint FP128:$src)), (CLGXBR 5, FP128:$src, 0)>;
+}
+
+
//===----------------------------------------------------------------------===//
// Unary arithmetic
//===----------------------------------------------------------------------===//
def SQDB : UnaryRXE<"sqdb", 0xED15, loadu<fsqrt>, FP64, 8>;
// Round to an integer, with the second operand (modifier M3) specifying
-// the rounding mode.
-//
-// These forms always check for inexact conditions. z196 added versions
-// that allow this to suppressed (as for fnearbyint), but we don't yet
-// support -march=z196.
+// the rounding mode. These forms always check for inexact conditions.
def FIEBR : UnaryRRF<"fieb", 0xB357, FP32, FP32>;
def FIDBR : UnaryRRF<"fidb", 0xB35F, FP64, FP64>;
def FIXBR : UnaryRRF<"fixb", 0xB347, FP128, FP128>;
def : Pat<(frint FP64:$src), (FIDBR 0, FP64:$src)>;
def : Pat<(frint FP128:$src), (FIXBR 0, FP128:$src)>;
+let Predicates = [FeatureFPExtension] in {
+ // Extended forms of the FIxBR instructions. M4 can be set to 4
+ // to suppress detection of inexact conditions.
+ def FIEBRA : UnaryRRF4<"fiebra", 0xB357, FP32, FP32>;
+ def FIDBRA : UnaryRRF4<"fidbra", 0xB35F, FP64, FP64>;
+ def FIXBRA : UnaryRRF4<"fixbra", 0xB347, FP128, FP128>;
+
+ // fnearbyint is like frint but does not detect inexact conditions.
+ def : Pat<(fnearbyint FP32:$src), (FIEBRA 0, FP32:$src, 4)>;
+ def : Pat<(fnearbyint FP64:$src), (FIDBRA 0, FP64:$src, 4)>;
+ def : Pat<(fnearbyint FP128:$src), (FIXBRA 0, FP128:$src, 4)>;
+
+ // floor is no longer allowed to raise an inexact condition,
+ // so restrict it to the cases where the condition can be suppressed.
+ // Mode 7 is round towards -inf.
+ def : Pat<(ffloor FP32:$src), (FIEBRA 7, FP32:$src, 4)>;
+ def : Pat<(ffloor FP64:$src), (FIDBRA 7, FP64:$src, 4)>;
+ def : Pat<(ffloor FP128:$src), (FIXBRA 7, FP128:$src, 4)>;
+
+ // Same idea for ceil, where mode 6 is round towards +inf.
+ def : Pat<(fceil FP32:$src), (FIEBRA 6, FP32:$src, 4)>;
+ def : Pat<(fceil FP64:$src), (FIDBRA 6, FP64:$src, 4)>;
+ def : Pat<(fceil FP128:$src), (FIXBRA 6, FP128:$src, 4)>;
+
+ // Same idea for trunc, where mode 5 is round towards zero.
+ def : Pat<(ftrunc FP32:$src), (FIEBRA 5, FP32:$src, 4)>;
+ def : Pat<(ftrunc FP64:$src), (FIDBRA 5, FP64:$src, 4)>;
+ def : Pat<(ftrunc FP128:$src), (FIXBRA 5, FP128:$src, 4)>;
+
+ // Same idea for round, where mode 1 is round towards nearest with
+ // ties away from zero.
+ def : Pat<(frnd FP32:$src), (FIEBRA 1, FP32:$src, 4)>;
+ def : Pat<(frnd FP64:$src), (FIDBRA 1, FP64:$src, 4)>;
+ def : Pat<(frnd FP128:$src), (FIXBRA 1, FP128:$src, 4)>;
+}
+
//===----------------------------------------------------------------------===//
// Binary arithmetic
//===----------------------------------------------------------------------===//
def MDEBR : BinaryRRE<"mdeb", 0xB30C, null_frag, FP64, FP32>;
def : Pat<(fmul (f64 (fextend FP32:$src1)), (f64 (fextend FP32:$src2))),
(MDEBR (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
- FP32:$src1, subreg_32bit), FP32:$src2)>;
+ FP32:$src1, subreg_r32), FP32:$src2)>;
// f64 multiplication of an FP32 register and an f32 memory.
def MDEB : BinaryRXE<"mdeb", 0xED0C, null_frag, FP64, load, 4>;
def : Pat<(fmul (f64 (fextend FP32:$src1)),
(f64 (extloadf32 bdxaddr12only:$addr))),
- (MDEB (INSERT_SUBREG (f64 (IMPLICIT_DEF)), FP32:$src1, subreg_32bit),
+ (MDEB (INSERT_SUBREG (f64 (IMPLICIT_DEF)), FP32:$src1, subreg_r32),
bdxaddr12only:$addr)>;
// f128 multiplication of two FP64 registers.
def MXDBR : BinaryRRE<"mxdb", 0xB307, null_frag, FP128, FP64>;
def : Pat<(fmul (f128 (fextend FP64:$src1)), (f128 (fextend FP64:$src2))),
(MXDBR (INSERT_SUBREG (f128 (IMPLICIT_DEF)),
- FP64:$src1, subreg_high), FP64:$src2)>;
+ FP64:$src1, subreg_h64), FP64:$src2)>;
// f128 multiplication of an FP64 register and an f64 memory.
def MXDB : BinaryRXE<"mxdb", 0xED07, null_frag, FP128, load, 8>;
def : Pat<(fmul (f128 (fextend FP64:$src1)),
(f128 (extloadf64 bdxaddr12only:$addr))),
- (MXDB (INSERT_SUBREG (f128 (IMPLICIT_DEF)), FP64:$src1, subreg_high),
+ (MXDB (INSERT_SUBREG (f128 (IMPLICIT_DEF)), FP64:$src1, subreg_h64),
bdxaddr12only:$addr)>;
// Fused multiply-add.
//===----------------------------------------------------------------------===//
let Defs = [CC], CCValues = 0xF in {
- def CEBR : CompareRRE<"ceb", 0xB309, z_cmp, FP32, FP32>;
- def CDBR : CompareRRE<"cdb", 0xB319, z_cmp, FP64, FP64>;
- def CXBR : CompareRRE<"cxb", 0xB349, z_cmp, FP128, FP128>;
+ def CEBR : CompareRRE<"ceb", 0xB309, z_fcmp, FP32, FP32>;
+ def CDBR : CompareRRE<"cdb", 0xB319, z_fcmp, FP64, FP64>;
+ def CXBR : CompareRRE<"cxb", 0xB349, z_fcmp, FP128, FP128>;
- def CEB : CompareRXE<"ceb", 0xED09, z_cmp, FP32, load, 4>;
- def CDB : CompareRXE<"cdb", 0xED19, z_cmp, FP64, load, 8>;
+ def CEB : CompareRXE<"ceb", 0xED09, z_fcmp, FP32, load, 4>;
+ def CDB : CompareRXE<"cdb", 0xED19, z_fcmp, FP64, load, 8>;
}
//===----------------------------------------------------------------------===//
OSDN Git Service
