Use roundps/pd for llvm.ceil, llvm.trunc, llvm.rint, and llvm.nearbyint of vector...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168141 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index d276353cca6a9e91bfdc572186bb933bbdaae9e6..80dd9ef613bd15addbad7cb4dcc95032418cc045 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -978,7 +978,15 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FNEARBYINT,         MVT::f64,   Legal);
 
     setOperationAction(ISD::FFLOOR,             MVT::v4f32, Legal);
+    setOperationAction(ISD::FCEIL,              MVT::v4f32, Legal);
+    setOperationAction(ISD::FTRUNC,             MVT::v4f32, Legal);
+    setOperationAction(ISD::FRINT,              MVT::v4f32, Legal);
+    setOperationAction(ISD::FNEARBYINT,         MVT::v4f32, Legal);
     setOperationAction(ISD::FFLOOR,             MVT::v2f64, Legal);
+    setOperationAction(ISD::FCEIL,              MVT::v2f64, Legal);
+    setOperationAction(ISD::FTRUNC,             MVT::v2f64, Legal);
+    setOperationAction(ISD::FRINT,              MVT::v2f64, Legal);
+    setOperationAction(ISD::FNEARBYINT,         MVT::v2f64, Legal);
 
     // FIXME: Do we need to handle scalar-to-vector here?
     setOperationAction(ISD::MUL,                MVT::v4i32, Legal);
@@ -1058,6 +1066,10 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FDIV,               MVT::v8f32, Legal);
     setOperationAction(ISD::FSQRT,              MVT::v8f32, Legal);
     setOperationAction(ISD::FFLOOR,             MVT::v8f32, Legal);
+    setOperationAction(ISD::FCEIL,              MVT::v8f32, Legal);
+    setOperationAction(ISD::FTRUNC,             MVT::v8f32, Legal);
+    setOperationAction(ISD::FRINT,              MVT::v8f32, Legal);
+    setOperationAction(ISD::FNEARBYINT,         MVT::v8f32, Legal);
     setOperationAction(ISD::FNEG,               MVT::v8f32, Custom);
     setOperationAction(ISD::FABS,               MVT::v8f32, Custom);
 
@@ -1067,6 +1079,10 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FDIV,               MVT::v4f64, Legal);
     setOperationAction(ISD::FSQRT,              MVT::v4f64, Legal);
     setOperationAction(ISD::FFLOOR,             MVT::v4f64, Legal);
+    setOperationAction(ISD::FCEIL,              MVT::v4f64, Legal);
+    setOperationAction(ISD::FTRUNC,             MVT::v4f64, Legal);
+    setOperationAction(ISD::FRINT,              MVT::v4f64, Legal);
+    setOperationAction(ISD::FNEARBYINT,         MVT::v4f64, Legal);
     setOperationAction(ISD::FNEG,               MVT::v4f64, Custom);
     setOperationAction(ISD::FABS,               MVT::v4f64, Custom);
 

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 6f48d7ed7fe1f6a09fdda4088f1c3d4022b35664..229e8b263f9de31b675f6b094b46b014d7c47d4c 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -6378,12 +6378,47 @@ let Predicates = [HasAVX] in {
 
   def : Pat<(v4f32 (ffloor VR128:$src)),
             (VROUNDPSr VR128:$src, (i32 0x1))>;
+  def : Pat<(v4f32 (fnearbyint VR128:$src)),
+            (VROUNDPSr VR128:$src, (i32 0xC))>;
+  def : Pat<(v4f32 (fceil VR128:$src)),
+            (VROUNDPSr VR128:$src, (i32 0x2))>;
+  def : Pat<(v4f32 (frint VR128:$src)),
+            (VROUNDPSr VR128:$src, (i32 0x4))>;
+  def : Pat<(v4f32 (ftrunc VR128:$src)),
+            (VROUNDPSr VR128:$src, (i32 0x3))>;
+
   def : Pat<(v2f64 (ffloor VR128:$src)),
             (VROUNDPDr VR128:$src, (i32 0x1))>;
+  def : Pat<(v2f64 (fnearbyint VR128:$src)),
+            (VROUNDPDr VR128:$src, (i32 0xC))>;
+  def : Pat<(v2f64 (fceil VR128:$src)),
+            (VROUNDPDr VR128:$src, (i32 0x2))>;
+  def : Pat<(v2f64 (frint VR128:$src)),
+            (VROUNDPDr VR128:$src, (i32 0x4))>;
+  def : Pat<(v2f64 (ftrunc VR128:$src)),
+            (VROUNDPDr VR128:$src, (i32 0x3))>;
+
   def : Pat<(v8f32 (ffloor VR256:$src)),
             (VROUNDYPSr VR256:$src, (i32 0x1))>;
+  def : Pat<(v8f32 (fnearbyint VR256:$src)),
+            (VROUNDYPSr VR256:$src, (i32 0xC))>;
+  def : Pat<(v8f32 (fceil VR256:$src)),
+            (VROUNDYPSr VR256:$src, (i32 0x2))>;
+  def : Pat<(v8f32 (frint VR256:$src)),
+            (VROUNDYPSr VR256:$src, (i32 0x4))>;
+  def : Pat<(v8f32 (ftrunc VR256:$src)),
+            (VROUNDYPSr VR256:$src, (i32 0x3))>;
+
   def : Pat<(v4f64 (ffloor VR256:$src)),
             (VROUNDYPDr VR256:$src, (i32 0x1))>;
+  def : Pat<(v4f64 (fnearbyint VR256:$src)),
+            (VROUNDYPDr VR256:$src, (i32 0xC))>;
+  def : Pat<(v4f64 (fceil VR256:$src)),
+            (VROUNDYPDr VR256:$src, (i32 0x2))>;
+  def : Pat<(v4f64 (frint VR256:$src)),
+            (VROUNDYPDr VR256:$src, (i32 0x4))>;
+  def : Pat<(v4f64 (ftrunc VR256:$src)),
+            (VROUNDYPDr VR256:$src, (i32 0x3))>;
 }
 
 defm ROUND  : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
@@ -6417,8 +6452,25 @@ let Predicates = [UseSSE41] in {
 
   def : Pat<(v4f32 (ffloor VR128:$src)),
             (ROUNDPSr VR128:$src, (i32 0x1))>;
+  def : Pat<(v4f32 (fnearbyint VR128:$src)),
+            (ROUNDPSr VR128:$src, (i32 0xC))>;
+  def : Pat<(v4f32 (fceil VR128:$src)),
+            (ROUNDPSr VR128:$src, (i32 0x2))>;
+  def : Pat<(v4f32 (frint VR128:$src)),
+            (ROUNDPSr VR128:$src, (i32 0x4))>;
+  def : Pat<(v4f32 (ftrunc VR128:$src)),
+            (ROUNDPSr VR128:$src, (i32 0x3))>;
+
   def : Pat<(v2f64 (ffloor VR128:$src)),
             (ROUNDPDr VR128:$src, (i32 0x1))>;
+  def : Pat<(v2f64 (fnearbyint VR128:$src)),
+            (ROUNDPDr VR128:$src, (i32 0xC))>;
+  def : Pat<(v2f64 (fceil VR128:$src)),
+            (ROUNDPDr VR128:$src, (i32 0x2))>;
+  def : Pat<(v2f64 (frint VR128:$src)),
+            (ROUNDPDr VR128:$src, (i32 0x4))>;
+  def : Pat<(v2f64 (ftrunc VR128:$src)),
+            (ROUNDPDr VR128:$src, (i32 0x3))>;
 }
 
 //===----------------------------------------------------------------------===//

diff --git a/test/CodeGen/X86/vec_floor.ll b/test/CodeGen/X86/vec_floor.ll
index 5e0160bd2856c7ef5415a64155bdf9e212183e53..4db68bd18223aeae549056e8b6c05400bf95eed9 100644 (file)
--- a/test/CodeGen/X86/vec_floor.ll
+++ b/test/CodeGen/X86/vec_floor.ll
@@ -36,3 +36,147 @@ define <8 x float> @floor_v8f32(<8 x float> %p)
   ret <8 x float> %t
 }
 declare <8 x float> @llvm.floor.v8f32(<8 x float> %p)
+
+define <2 x double> @ceil_v2f64(<2 x double> %p)
+{
+  ; CHECK: ceil_v2f64
+  ; CHECK: vroundpd
+  %t = call <2 x double> @llvm.ceil.v2f64(<2 x double> %p)
+  ret <2 x double> %t
+}
+declare <2 x double> @llvm.ceil.v2f64(<2 x double> %p)
+
+define <4 x float> @ceil_v4f32(<4 x float> %p)
+{
+  ; CHECK: ceil_v4f32
+  ; CHECK: vroundps
+  %t = call <4 x float> @llvm.ceil.v4f32(<4 x float> %p)
+  ret <4 x float> %t
+}
+declare <4 x float> @llvm.ceil.v4f32(<4 x float> %p)
+
+define <4 x double> @ceil_v4f64(<4 x double> %p)
+{
+  ; CHECK: ceil_v4f64
+  ; CHECK: vroundpd
+  %t = call <4 x double> @llvm.ceil.v4f64(<4 x double> %p)
+  ret <4 x double> %t
+}
+declare <4 x double> @llvm.ceil.v4f64(<4 x double> %p)
+
+define <8 x float> @ceil_v8f32(<8 x float> %p)
+{
+  ; CHECK: ceil_v8f32
+  ; CHECK: vroundps
+  %t = call <8 x float> @llvm.ceil.v8f32(<8 x float> %p)
+  ret <8 x float> %t
+}
+declare <8 x float> @llvm.ceil.v8f32(<8 x float> %p)
+
+define <2 x double> @trunc_v2f64(<2 x double> %p)
+{
+  ; CHECK: trunc_v2f64
+  ; CHECK: vroundpd
+  %t = call <2 x double> @llvm.trunc.v2f64(<2 x double> %p)
+  ret <2 x double> %t
+}
+declare <2 x double> @llvm.trunc.v2f64(<2 x double> %p)
+
+define <4 x float> @trunc_v4f32(<4 x float> %p)
+{
+  ; CHECK: trunc_v4f32
+  ; CHECK: vroundps
+  %t = call <4 x float> @llvm.trunc.v4f32(<4 x float> %p)
+  ret <4 x float> %t
+}
+declare <4 x float> @llvm.trunc.v4f32(<4 x float> %p)
+
+define <4 x double> @trunc_v4f64(<4 x double> %p)
+{
+  ; CHECK: trunc_v4f64
+  ; CHECK: vroundpd
+  %t = call <4 x double> @llvm.trunc.v4f64(<4 x double> %p)
+  ret <4 x double> %t
+}
+declare <4 x double> @llvm.trunc.v4f64(<4 x double> %p)
+
+define <8 x float> @trunc_v8f32(<8 x float> %p)
+{
+  ; CHECK: trunc_v8f32
+  ; CHECK: vroundps
+  %t = call <8 x float> @llvm.trunc.v8f32(<8 x float> %p)
+  ret <8 x float> %t
+}
+declare <8 x float> @llvm.trunc.v8f32(<8 x float> %p)
+
+define <2 x double> @rint_v2f64(<2 x double> %p)
+{
+  ; CHECK: rint_v2f64
+  ; CHECK: vroundpd
+  %t = call <2 x double> @llvm.rint.v2f64(<2 x double> %p)
+  ret <2 x double> %t
+}
+declare <2 x double> @llvm.rint.v2f64(<2 x double> %p)
+
+define <4 x float> @rint_v4f32(<4 x float> %p)
+{
+  ; CHECK: rint_v4f32
+  ; CHECK: vroundps
+  %t = call <4 x float> @llvm.rint.v4f32(<4 x float> %p)
+  ret <4 x float> %t
+}
+declare <4 x float> @llvm.rint.v4f32(<4 x float> %p)
+
+define <4 x double> @rint_v4f64(<4 x double> %p)
+{
+  ; CHECK: rint_v4f64
+  ; CHECK: vroundpd
+  %t = call <4 x double> @llvm.rint.v4f64(<4 x double> %p)
+  ret <4 x double> %t
+}
+declare <4 x double> @llvm.rint.v4f64(<4 x double> %p)
+
+define <8 x float> @rint_v8f32(<8 x float> %p)
+{
+  ; CHECK: rint_v8f32
+  ; CHECK: vroundps
+  %t = call <8 x float> @llvm.rint.v8f32(<8 x float> %p)
+  ret <8 x float> %t
+}
+declare <8 x float> @llvm.rint.v8f32(<8 x float> %p)
+
+define <2 x double> @nearbyint_v2f64(<2 x double> %p)
+{
+  ; CHECK: nearbyint_v2f64
+  ; CHECK: vroundpd
+  %t = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %p)
+  ret <2 x double> %t
+}
+declare <2 x double> @llvm.nearbyint.v2f64(<2 x double> %p)
+
+define <4 x float> @nearbyint_v4f32(<4 x float> %p)
+{
+  ; CHECK: nearbyint_v4f32
+  ; CHECK: vroundps
+  %t = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %p)
+  ret <4 x float> %t
+}
+declare <4 x float> @llvm.nearbyint.v4f32(<4 x float> %p)
+
+define <4 x double> @nearbyint_v4f64(<4 x double> %p)
+{
+  ; CHECK: nearbyint_v4f64
+  ; CHECK: vroundpd
+  %t = call <4 x double> @llvm.nearbyint.v4f64(<4 x double> %p)
+  ret <4 x double> %t
+}
+declare <4 x double> @llvm.nearbyint.v4f64(<4 x double> %p)
+
+define <8 x float> @nearbyint_v8f32(<8 x float> %p)
+{
+  ; CHECK: nearbyint_v8f32
+  ; CHECK: vroundps
+  %t = call <8 x float> @llvm.nearbyint.v8f32(<8 x float> %p)
+  ret <8 x float> %t
+}
+declare <8 x float> @llvm.nearbyint.v8f32(<8 x float> %p)