[PATCH] D85453: [PowerPC] Implement __int128 vector divide operations

[Albion Fung via Phabricator via cfe-commits]

Conanap created this revision.
Conanap added reviewers: PowerPC, saghir, nemanjai, hfinkel.
Conanap added projects: LLVM, clang, PowerPC.
Herald added a subscriber: kbarton.
Conanap requested review of this revision.
Herald added a subscriber: wuzish.

This patch implements __int128 vector divide operations for ISA3.1.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D85453

Files:
  clang/lib/Headers/altivec.h
  clang/test/CodeGen/builtins-ppc-p10vector.c
  llvm/lib/Target/PowerPC/PPCISelLowering.cpp
  llvm/lib/Target/PowerPC/PPCInstrPrefix.td
  llvm/test/CodeGen/PowerPC/p10-vector-divide.ll


Index: llvm/test/CodeGen/PowerPC/p10-vector-divide.ll
===================================================================
--- llvm/test/CodeGen/PowerPC/p10-vector-divide.ll
+++ llvm/test/CodeGen/PowerPC/p10-vector-divide.ll
@@ -49,3 +49,21 @@
   %div = sdiv <4 x i32> %a, %b
   ret <4 x i32> %div
 }
+
+define <1 x i128> @test_vdivsq(<1 x i128> %x, <1 x i128> %y) nounwind readnone {
+; CHECK-LABEL: test_vdivsq:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vdivsq v2, v2, v3
+; CHECK-NEXT:    blr
+  %tmp = sdiv <1 x i128> %x, %y
+  ret <1 x i128> %tmp
+}
+
+define <1 x i128> @test_vdivuq(<1 x i128> %x, <1 x i128> %y) nounwind readnone {
+; CHECK-LABEL: test_vdivuq:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vdivuq v2, v2, v3
+; CHECK-NEXT:    blr
+  %tmp = udiv <1 x i128> %x, %y
+  ret <1 x i128> %tmp
+}
Index: llvm/lib/Target/PowerPC/PPCInstrPrefix.td
===================================================================
--- llvm/lib/Target/PowerPC/PPCInstrPrefix.td
+++ llvm/lib/Target/PowerPC/PPCInstrPrefix.td
@@ -1124,9 +1124,11 @@
                            (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
                            "vmsumcud $vD, $vA, $vB, $vC", IIC_VecGeneral, []>;
   def VDIVSQ : VXForm_1<267, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
-                        "vdivsq $vD, $vA, $vB", IIC_VecGeneral, []>;
+                        "vdivsq $vD, $vA, $vB", IIC_VecGeneral,
+                        [(set v1i128:$vD, (sdiv v1i128:$vA, v1i128:$vB))]>;
   def VDIVUQ : VXForm_1<11, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
-                        "vdivuq $vD, $vA, $vB", IIC_VecGeneral, []>;
+                        "vdivuq $vD, $vA, $vB", IIC_VecGeneral,
+                        [(set v1i128:$vD, (udiv v1i128:$vA, v1i128:$vB))]>;
   def VDIVESQ : VXForm_1<779, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
                          "vdivesq $vD, $vA, $vB", IIC_VecGeneral, []>;
   def VDIVEUQ : VXForm_1<523, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
Index: llvm/lib/Target/PowerPC/PPCISelLowering.cpp
===================================================================
--- llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -836,6 +836,8 @@
       setOperationAction(ISD::SREM, MVT::v2i64, Legal);
       setOperationAction(ISD::UREM, MVT::v4i32, Legal);
       setOperationAction(ISD::SREM, MVT::v4i32, Legal);
+      setOperationAction(ISD::UDIV, MVT::v1i128, Legal);
+      setOperationAction(ISD::SDIV, MVT::v1i128, Legal);
     }
 
     setOperationAction(ISD::MUL, MVT::v8i16, Legal);
Index: clang/test/CodeGen/builtins-ppc-p10vector.c
===================================================================
--- clang/test/CodeGen/builtins-ppc-p10vector.c
+++ clang/test/CodeGen/builtins-ppc-p10vector.c
@@ -17,6 +17,7 @@
 vector unsigned int vuia, vuib, vuic;
 vector signed long long vslla, vsllb;
 vector unsigned long long vulla, vullb, vullc;
+vector signed __int128 vsi128a, vsi128b;
 vector unsigned __int128 vui128a, vui128b, vui128c;
 vector float vfa, vfb;
 vector double vda, vdb;
@@ -61,6 +62,18 @@
   return vec_div(vulla, vullb);
 }
 
+vector unsigned __int128 test_vec_div_u128(void) {
+  // CHECK: udiv <1 x i128>
+  // CHECK-NEXT: ret <1 x i128>
+  return vec_div(vui128a, vui128b);
+}
+
+vector signed __int128 test_vec_div_s128(void) {
+  // CHECK: sdiv <1 x i128>
+  // CHECK-NEXT: ret <1 x i128>
+  return vec_div(vsi128a, vsi128b);
+}
+
 vector signed int test_vec_mod_si(void) {
   // CHECK: srem <4 x i32>
   // CHECK-NEXT: ret <4 x i32>
Index: clang/lib/Headers/altivec.h
===================================================================
--- clang/lib/Headers/altivec.h
+++ clang/lib/Headers/altivec.h
@@ -3288,6 +3288,18 @@
 }
 #endif
 
+#ifdef __POWER10_VECTOR__
+static __inline__ vector unsigned __int128 __ATTRS_o_ai
+vec_div(vector unsigned __int128 __a, vector unsigned __int128 __b) {
+  return __a / __b;
+}
+
+static __inline__ vector signed __int128 __ATTRS_o_ai
+vec_div(vector signed __int128 __a, vector signed __int128 __b) {
+  return __a / __b;
+}
+#endif __POWER10_VECTOR__
+
 /* vec_dss */
 
 #define vec_dss __builtin_altivec_dss

_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits