[SystemZ] Add vector intrinsics

This adds intrinsics to allow access to all of the z13 vector instructions.
Note that instructions whose semantics can be described by standard LLVM IR
do not get any intrinsics.

For each instructions whose semantics *cannot* (fully) be described, we
define an LLVM IR target-specific intrinsic that directly maps to this
instruction.

For instructions that also set the condition code, the LLVM IR intrinsic
returns the post-instruction CC value as a second result.  Instruction
selection will attempt to detect code that compares that CC value against
constants and use the condition code directly instead.

Based on a patch by Richard Sandiford.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236527 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/IR/IntrinsicsSystemZ.td

@@ -11,6 +11,185 @@
 //
 //===----------------------------------------------------------------------===//
 
+class SystemZUnaryConv<string name, LLVMType result, LLVMType arg>
+  : GCCBuiltin<"__builtin_s390_" ## name>,
+    Intrinsic<[result], [arg], [IntrNoMem]>;
+
+class SystemZUnary<string name, LLVMType type>
+  : SystemZUnaryConv<name, type, type>;
+
+class SystemZUnaryConvCC<LLVMType result, LLVMType arg>
+  : Intrinsic<[result, llvm_i32_ty], [arg], [IntrNoMem]>;
+
+class SystemZUnaryCC<LLVMType type>
+  : SystemZUnaryConvCC<type, type>;
+
+class SystemZBinaryConv<string name, LLVMType result, LLVMType arg>
+  : GCCBuiltin<"__builtin_s390_" ## name>,
+    Intrinsic<[result], [arg, arg], [IntrNoMem]>;
+
+class SystemZBinary<string name, LLVMType type>
+  : SystemZBinaryConv<name, type, type>;
+
+class SystemZBinaryInt<string name, LLVMType type>
+  : GCCBuiltin<"__builtin_s390_" ## name>,
+    Intrinsic<[type], [type, llvm_i32_ty], [IntrNoMem]>;
+
+class SystemZBinaryConvCC<LLVMType result, LLVMType arg>
+  : Intrinsic<[result, llvm_i32_ty], [arg, arg], [IntrNoMem]>;
+
+class SystemZBinaryConvIntCC<LLVMType result, LLVMType arg>
+  : Intrinsic<[result, llvm_i32_ty], [arg, llvm_i32_ty], [IntrNoMem]>;
+
+class SystemZBinaryCC<LLVMType type>
+  : SystemZBinaryConvCC<type, type>;
+
+class SystemZTernaryConv<string name, LLVMType result, LLVMType arg>
+  : GCCBuiltin<"__builtin_s390_" ## name>,
+    Intrinsic<[result], [arg, arg, result], [IntrNoMem]>;
+
+class SystemZTernary<string name, LLVMType type>
+  : SystemZTernaryConv<name, type, type>;
+
+class SystemZTernaryInt<string name, LLVMType type>
+  : GCCBuiltin<"__builtin_s390_" ## name>,
+    Intrinsic<[type], [type, type, llvm_i32_ty], [IntrNoMem]>;
+
+class SystemZTernaryIntCC<LLVMType type>
+  : Intrinsic<[type, llvm_i32_ty], [type, type, llvm_i32_ty], [IntrNoMem]>;
+
+class SystemZQuaternaryInt<string name, LLVMType type>
+  : GCCBuiltin<"__builtin_s390_" ## name>,
+    Intrinsic<[type], [type, type, type, llvm_i32_ty], [IntrNoMem]>;
+
+class SystemZQuaternaryIntCC<LLVMType type>
+  : Intrinsic<[type, llvm_i32_ty], [type, type, type, llvm_i32_ty],
+              [IntrNoMem]>;
+
+multiclass SystemZUnaryExtBHF<string name> {
+  def b : SystemZUnaryConv<name##"b", llvm_v8i16_ty, llvm_v16i8_ty>;
+  def h : SystemZUnaryConv<name##"h", llvm_v4i32_ty, llvm_v8i16_ty>;
+  def f : SystemZUnaryConv<name##"f", llvm_v2i64_ty, llvm_v4i32_ty>;
+}
+
+multiclass SystemZUnaryExtBHWF<string name> {
+  def b  : SystemZUnaryConv<name##"b",  llvm_v8i16_ty, llvm_v16i8_ty>;
+  def hw : SystemZUnaryConv<name##"hw", llvm_v4i32_ty, llvm_v8i16_ty>;
+  def f  : SystemZUnaryConv<name##"f",  llvm_v2i64_ty, llvm_v4i32_ty>;
+}
+
+multiclass SystemZUnaryBHF<string name> {
+  def b : SystemZUnary<name##"b", llvm_v16i8_ty>;
+  def h : SystemZUnary<name##"h", llvm_v8i16_ty>;
+  def f : SystemZUnary<name##"f", llvm_v4i32_ty>;
+}
+
+multiclass SystemZUnaryBHFG<string name> : SystemZUnaryBHF<name> {
+  def g : SystemZUnary<name##"g", llvm_v2i64_ty>;
+}
+
+multiclass SystemZUnaryCCBHF {
+  def bs : SystemZUnaryCC<llvm_v16i8_ty>;
+  def hs : SystemZUnaryCC<llvm_v8i16_ty>;
+  def fs : SystemZUnaryCC<llvm_v4i32_ty>;
+}
+
+multiclass SystemZBinaryTruncHFG<string name> {
+  def h : SystemZBinaryConv<name##"h", llvm_v16i8_ty, llvm_v8i16_ty>;
+  def f : SystemZBinaryConv<name##"f", llvm_v8i16_ty, llvm_v4i32_ty>;
+  def g : SystemZBinaryConv<name##"g", llvm_v4i32_ty, llvm_v2i64_ty>;
+}
+
+multiclass SystemZBinaryTruncCCHFG {
+  def hs : SystemZBinaryConvCC<llvm_v16i8_ty, llvm_v8i16_ty>;
+  def fs : SystemZBinaryConvCC<llvm_v8i16_ty, llvm_v4i32_ty>;
+  def gs : SystemZBinaryConvCC<llvm_v4i32_ty, llvm_v2i64_ty>;
+}
+
+multiclass SystemZBinaryExtBHF<string name> {
+  def b : SystemZBinaryConv<name##"b", llvm_v8i16_ty, llvm_v16i8_ty>;
+  def h : SystemZBinaryConv<name##"h", llvm_v4i32_ty, llvm_v8i16_ty>;
+  def f : SystemZBinaryConv<name##"f", llvm_v2i64_ty, llvm_v4i32_ty>;
+}
+
+multiclass SystemZBinaryExtBHFG<string name> : SystemZBinaryExtBHF<name> {
+  def g : SystemZBinaryConv<name##"g", llvm_v16i8_ty, llvm_v2i64_ty>;
+}
+
+multiclass SystemZBinaryBHF<string name> {
+  def b : SystemZBinary<name##"b", llvm_v16i8_ty>;
+  def h : SystemZBinary<name##"h", llvm_v8i16_ty>;
+  def f : SystemZBinary<name##"f", llvm_v4i32_ty>;
+}
+
+multiclass SystemZBinaryBHFG<string name> : SystemZBinaryBHF<name> {
+  def g : SystemZBinary<name##"g", llvm_v2i64_ty>;
+}
+
+multiclass SystemZBinaryIntBHFG<string name> {
+  def b : SystemZBinaryInt<name##"b", llvm_v16i8_ty>;
+  def h : SystemZBinaryInt<name##"h", llvm_v8i16_ty>;
+  def f : SystemZBinaryInt<name##"f", llvm_v4i32_ty>;
+  def g : SystemZBinaryInt<name##"g", llvm_v2i64_ty>;
+}
+
+multiclass SystemZBinaryCCBHF {
+  def bs : SystemZBinaryCC<llvm_v16i8_ty>;
+  def hs : SystemZBinaryCC<llvm_v8i16_ty>;
+  def fs : SystemZBinaryCC<llvm_v4i32_ty>;
+}
+
+multiclass SystemZCompareBHFG<string name> {
+  def bs : SystemZBinaryCC<llvm_v16i8_ty>;
+  def hs : SystemZBinaryCC<llvm_v8i16_ty>;
+  def fs : SystemZBinaryCC<llvm_v4i32_ty>;
+  def gs : SystemZBinaryCC<llvm_v2i64_ty>;
+}
+
+multiclass SystemZTernaryExtBHF<string name> {
+  def b : SystemZTernaryConv<name##"b", llvm_v8i16_ty, llvm_v16i8_ty>;
+  def h : SystemZTernaryConv<name##"h", llvm_v4i32_ty, llvm_v8i16_ty>;
+  def f : SystemZTernaryConv<name##"f", llvm_v2i64_ty, llvm_v4i32_ty>;
+}
+
+multiclass SystemZTernaryExtBHFG<string name> : SystemZTernaryExtBHF<name> {
+  def g : SystemZTernaryConv<name##"g", llvm_v16i8_ty, llvm_v2i64_ty>;
+}
+
+multiclass SystemZTernaryBHF<string name> {
+  def b : SystemZTernary<name##"b", llvm_v16i8_ty>;
+  def h : SystemZTernary<name##"h", llvm_v8i16_ty>;
+  def f : SystemZTernary<name##"f", llvm_v4i32_ty>;
+}
+
+multiclass SystemZTernaryIntBHF<string name> {
+  def b : SystemZTernaryInt<name##"b", llvm_v16i8_ty>;
+  def h : SystemZTernaryInt<name##"h", llvm_v8i16_ty>;
+  def f : SystemZTernaryInt<name##"f", llvm_v4i32_ty>;
+}
+
+multiclass SystemZTernaryIntCCBHF {
+  def bs : SystemZTernaryIntCC<llvm_v16i8_ty>;
+  def hs : SystemZTernaryIntCC<llvm_v8i16_ty>;
+  def fs : SystemZTernaryIntCC<llvm_v4i32_ty>;
+}
+
+multiclass SystemZQuaternaryIntBHF<string name> {
+  def b : SystemZQuaternaryInt<name##"b", llvm_v16i8_ty>;
+  def h : SystemZQuaternaryInt<name##"h", llvm_v8i16_ty>;
+  def f : SystemZQuaternaryInt<name##"f", llvm_v4i32_ty>;
+}
+
+multiclass SystemZQuaternaryIntBHFG<string name> : SystemZQuaternaryIntBHF<name> {
+  def g : SystemZQuaternaryInt<name##"g", llvm_v2i64_ty>;
+}
+
+multiclass SystemZQuaternaryIntCCBHF {
+  def bs : SystemZQuaternaryIntCC<llvm_v16i8_ty>;
+  def hs : SystemZQuaternaryIntCC<llvm_v8i16_ty>;
+  def fs : SystemZQuaternaryIntCC<llvm_v4i32_ty>;
+}
+
 //===----------------------------------------------------------------------===//
 //
 // Transactional-execution intrinsics
@@ -44,3 +223,154 @@ let TargetPrefix = "s390" in {
                               Intrinsic<[], [llvm_i32_ty]>;
 }
 
+//===----------------------------------------------------------------------===//
+//
+// Vector intrinsics
+//
+//===----------------------------------------------------------------------===//
+
+let TargetPrefix = "s390" in {
+  def int_s390_lcbb : GCCBuiltin<"__builtin_s390_lcbb">,
+                      Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty],
+                                [IntrNoMem]>;
+
+  def int_s390_vlbb : GCCBuiltin<"__builtin_s390_vlbb">,
+                      Intrinsic<[llvm_v16i8_ty], [llvm_ptr_ty, llvm_i32_ty],
+                                [IntrReadArgMem]>;
+
+  def int_s390_vll : GCCBuiltin<"__builtin_s390_vll">,
+                     Intrinsic<[llvm_v16i8_ty], [llvm_i32_ty, llvm_ptr_ty],
+                               [IntrReadArgMem]>;
+
+  def int_s390_vpdi : GCCBuiltin<"__builtin_s390_vpdi">,
+                      Intrinsic<[llvm_v2i64_ty],
+                                [llvm_v2i64_ty, llvm_v2i64_ty, llvm_i32_ty],
+                                [IntrNoMem]>;
+
+  def int_s390_vperm : GCCBuiltin<"__builtin_s390_vperm">,
+                       Intrinsic<[llvm_v16i8_ty],
+                                 [llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty],
+                                 [IntrNoMem]>;
+
+  defm int_s390_vpks : SystemZBinaryTruncHFG<"vpks">;
+  defm int_s390_vpks : SystemZBinaryTruncCCHFG;
+
+  defm int_s390_vpkls : SystemZBinaryTruncHFG<"vpkls">;
+  defm int_s390_vpkls : SystemZBinaryTruncCCHFG;
+
+  def int_s390_vstl : GCCBuiltin<"__builtin_s390_vstl">,
+                      Intrinsic<[], [llvm_v16i8_ty, llvm_i32_ty, llvm_ptr_ty],
+                                // In fact write-only but there's no property
+                                // for that.
+                                [IntrReadWriteArgMem]>;
+
+  defm int_s390_vupl  : SystemZUnaryExtBHWF<"vupl">;
+  defm int_s390_vupll : SystemZUnaryExtBHF<"vupll">;
+
+  defm int_s390_vuph  : SystemZUnaryExtBHF<"vuph">;
+  defm int_s390_vuplh : SystemZUnaryExtBHF<"vuplh">;
+
+  defm int_s390_vacc : SystemZBinaryBHFG<"vacc">;
+
+  def int_s390_vaq    : SystemZBinary<"vaq",     llvm_v16i8_ty>;
+  def int_s390_vacq   : SystemZTernary<"vacq",   llvm_v16i8_ty>;
+  def int_s390_vaccq  : SystemZBinary<"vaccq",   llvm_v16i8_ty>;
+  def int_s390_vacccq : SystemZTernary<"vacccq", llvm_v16i8_ty>;
+
+  defm int_s390_vavg  : SystemZBinaryBHFG<"vavg">;
+  defm int_s390_vavgl : SystemZBinaryBHFG<"vavgl">;
+
+  def int_s390_vcksm : SystemZBinary<"vcksm", llvm_v4i32_ty>;
+
+  defm int_s390_vgfm  : SystemZBinaryExtBHFG<"vgfm">;
+  defm int_s390_vgfma : SystemZTernaryExtBHFG<"vgfma">;
+
+  defm int_s390_vmah  : SystemZTernaryBHF<"vmah">;
+  defm int_s390_vmalh : SystemZTernaryBHF<"vmalh">;
+  defm int_s390_vmae  : SystemZTernaryExtBHF<"vmae">;
+  defm int_s390_vmale : SystemZTernaryExtBHF<"vmale">;
+  defm int_s390_vmao  : SystemZTernaryExtBHF<"vmao">;
+  defm int_s390_vmalo : SystemZTernaryExtBHF<"vmalo">;
+
+  defm int_s390_vmh  : SystemZBinaryBHF<"vmh">;
+  defm int_s390_vmlh : SystemZBinaryBHF<"vmlh">;
+  defm int_s390_vme  : SystemZBinaryExtBHF<"vme">;
+  defm int_s390_vmle : SystemZBinaryExtBHF<"vmle">;
+  defm int_s390_vmo  : SystemZBinaryExtBHF<"vmo">;
+  defm int_s390_vmlo : SystemZBinaryExtBHF<"vmlo">;
+
+  defm int_s390_verllv : SystemZBinaryBHFG<"verllv">;
+  defm int_s390_verll  : SystemZBinaryIntBHFG<"verll">;
+  defm int_s390_verim  : SystemZQuaternaryIntBHFG<"verim">;
+
+  def int_s390_vsl   : SystemZBinary<"vsl",   llvm_v16i8_ty>;
+  def int_s390_vslb  : SystemZBinary<"vslb",  llvm_v16i8_ty>;
+  def int_s390_vsra  : SystemZBinary<"vsra",  llvm_v16i8_ty>;
+  def int_s390_vsrab : SystemZBinary<"vsrab", llvm_v16i8_ty>;
+  def int_s390_vsrl  : SystemZBinary<"vsrl",  llvm_v16i8_ty>;
+  def int_s390_vsrlb : SystemZBinary<"vsrlb", llvm_v16i8_ty>;
+
+  def int_s390_vsldb : GCCBuiltin<"__builtin_s390_vsldb">,
+                       Intrinsic<[llvm_v16i8_ty],
+                                 [llvm_v16i8_ty, llvm_v16i8_ty, llvm_i32_ty],
+                                 [IntrNoMem]>;
+
+  defm int_s390_vscbi : SystemZBinaryBHFG<"vscbi">;
+
+  def int_s390_vsq     : SystemZBinary<"vsq",      llvm_v16i8_ty>;
+  def int_s390_vsbiq   : SystemZTernary<"vsbiq",   llvm_v16i8_ty>;
+  def int_s390_vscbiq  : SystemZBinary<"vscbiq",   llvm_v16i8_ty>;
+  def int_s390_vsbcbiq : SystemZTernary<"vsbcbiq", llvm_v16i8_ty>;
+
+  def int_s390_vsumb : SystemZBinaryConv<"vsumb", llvm_v4i32_ty, llvm_v16i8_ty>;
+  def int_s390_vsumh : SystemZBinaryConv<"vsumh", llvm_v4i32_ty, llvm_v8i16_ty>;
+
+  def int_s390_vsumgh : SystemZBinaryConv<"vsumgh", llvm_v2i64_ty,
+                                          llvm_v8i16_ty>;
+  def int_s390_vsumgf : SystemZBinaryConv<"vsumgf", llvm_v2i64_ty,
+                                          llvm_v4i32_ty>;
+
+  def int_s390_vsumqf : SystemZBinaryConv<"vsumqf", llvm_v16i8_ty,
+                                          llvm_v4i32_ty>;
+  def int_s390_vsumqg : SystemZBinaryConv<"vsumqg", llvm_v16i8_ty,
+                                          llvm_v2i64_ty>;
+
+  def int_s390_vtm : SystemZBinaryConv<"vtm", llvm_i32_ty, llvm_v16i8_ty>;
+
+  defm int_s390_vceq : SystemZCompareBHFG<"vceq">;
+  defm int_s390_vch  : SystemZCompareBHFG<"vch">;
+  defm int_s390_vchl : SystemZCompareBHFG<"vchl">;
+
+  defm int_s390_vfae  : SystemZTernaryIntBHF<"vfae">;
+  defm int_s390_vfae  : SystemZTernaryIntCCBHF;
+  defm int_s390_vfaez : SystemZTernaryIntBHF<"vfaez">;
+  defm int_s390_vfaez : SystemZTernaryIntCCBHF;
+
+  defm int_s390_vfee  : SystemZBinaryBHF<"vfee">;
+  defm int_s390_vfee  : SystemZBinaryCCBHF;
+  defm int_s390_vfeez : SystemZBinaryBHF<"vfeez">;
+  defm int_s390_vfeez : SystemZBinaryCCBHF;
+
+  defm int_s390_vfene  : SystemZBinaryBHF<"vfene">;
+  defm int_s390_vfene  : SystemZBinaryCCBHF;
+  defm int_s390_vfenez : SystemZBinaryBHF<"vfenez">;
+  defm int_s390_vfenez : SystemZBinaryCCBHF;
+
+  defm int_s390_vistr : SystemZUnaryBHF<"vistr">;
+  defm int_s390_vistr : SystemZUnaryCCBHF;
+
+  defm int_s390_vstrc  : SystemZQuaternaryIntBHF<"vstrc">;
+  defm int_s390_vstrc  : SystemZQuaternaryIntCCBHF;
+  defm int_s390_vstrcz : SystemZQuaternaryIntBHF<"vstrcz">;
+  defm int_s390_vstrcz : SystemZQuaternaryIntCCBHF;
+
+  def int_s390_vfcedbs  : SystemZBinaryConvCC<llvm_v2i64_ty, llvm_v2f64_ty>;
+  def int_s390_vfchdbs  : SystemZBinaryConvCC<llvm_v2i64_ty, llvm_v2f64_ty>;
+  def int_s390_vfchedbs : SystemZBinaryConvCC<llvm_v2i64_ty, llvm_v2f64_ty>;
+
+  def int_s390_vftcidb : SystemZBinaryConvIntCC<llvm_v2i64_ty, llvm_v2f64_ty>;
+
+  def int_s390_vfidb : Intrinsic<[llvm_v2f64_ty],
+                                 [llvm_v2f64_ty, llvm_i32_ty, llvm_i32_ty],
+                                 [IntrNoMem]>;
+}

lib/Target/SystemZ/SystemZ.h

@@ -80,6 +80,13 @@ const unsigned CCMASK_TEND_TX   = CCMASK_0;
 const unsigned CCMASK_TEND_NOTX = CCMASK_2;
 const unsigned CCMASK_TEND      = CCMASK_TEND_TX | CCMASK_TEND_NOTX;
 
+// Condition-code mask assignments for vector comparisons (and similar
+// operations).
+const unsigned CCMASK_VCMP_ALL       = CCMASK_0;
+const unsigned CCMASK_VCMP_MIXED     = CCMASK_1;
+const unsigned CCMASK_VCMP_NONE      = CCMASK_3;
+const unsigned CCMASK_VCMP           = CCMASK_0 | CCMASK_1 | CCMASK_3;
+
 // The position of the low CC bit in an IPM result.
 const unsigned IPM_CC = 28;