[llvm-branch-commits] [llvm-branch] r195758 - Merging r195713:

[Bill Wendling]

Author: void
Date: Tue Nov 26 06:29:45 2013
New Revision: 195758

URL: http://llvm.org/viewvc/llvm-project?rev=195758&view=rev
Log:
Merging r195713:
------------------------------------------------------------------------
r195713 | kevinqin | 2013-11-25 18:33:42 -0800 (Mon, 25 Nov 2013) | 1 line

[AArch64]Implement 128 bit register copy with NEON.
------------------------------------------------------------------------

Modified:
    llvm/branches/release_34/   (props changed)
    llvm/branches/release_34/lib/Target/AArch64/AArch64InstrInfo.cpp
    llvm/branches/release_34/test/CodeGen/AArch64/neon-perm.ll

Propchange: llvm/branches/release_34/
------------------------------------------------------------------------------
--- svn:mergeinfo (original)
+++ svn:mergeinfo Tue Nov 26 06:29:45 2013
@@ -1,3 +1,3 @@
 /llvm/branches/Apple/Pertwee:110850,110961
 /llvm/branches/type-system-rewrite:133420-134817
-/llvm/trunk:155241,195092-195094,195100,195102-195103,195118,195129,195136,195138,195152,195156-195157,195161-195162,195193,195272,195317-195318,195327,195330,195333,195339,195343,195355,195364,195379,195397-195399,195421,195423,195432,195439,195444,195455-195456,195469,195476-195477,195479,195491-195493,195514,195528,195547,195567,195591,195599,195632,195635-195636,195670,195679,195682,195684,195716
+/llvm/trunk:155241,195092-195094,195100,195102-195103,195118,195129,195136,195138,195152,195156-195157,195161-195162,195193,195272,195317-195318,195327,195330,195333,195339,195343,195355,195364,195379,195397-195399,195421,195423,195432,195439,195444,195455-195456,195469,195476-195477,195479,195491-195493,195514,195528,195547,195567,195591,195599,195632,195635-195636,195670,195679,195682,195684,195713,195716

Modified: llvm/branches/release_34/lib/Target/AArch64/AArch64InstrInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/release_34/lib/Target/AArch64/AArch64InstrInfo.cpp?rev=195758&r1=195757&r2=195758&view=diff
==============================================================================
--- llvm/branches/release_34/lib/Target/AArch64/AArch64InstrInfo.cpp (original)
+++ llvm/branches/release_34/lib/Target/AArch64/AArch64InstrInfo.cpp Tue Nov 26 06:29:45 2013
@@ -114,23 +114,25 @@ void AArch64InstrInfo::copyPhysReg(Machi
   } else if (AArch64::FPR128RegClass.contains(DestReg)) {
     assert(AArch64::FPR128RegClass.contains(SrcReg));
 
-    // FIXME: there's no good way to do this, at least without NEON:
-    //   + There's no single move instruction for q-registers
-    //   + We can't create a spill slot and use normal STR/LDR because stack
-    //     allocation has already happened
-    //   + We can't go via X-registers with FMOV because register allocation has
-    //     already happened.
-    // This may not be efficient, but at least it works.
-    BuildMI(MBB, I, DL, get(AArch64::LSFP128_PreInd_STR), AArch64::XSP)
-      .addReg(SrcReg)
-      .addReg(AArch64::XSP)
-      .addImm(0x1ff & -16);
+    // If NEON is enable, we use ORR to implement this copy.
+    // If NEON isn't available, emit STR and LDR to handle this.
+    if(getSubTarget().hasNEON()) {
+      BuildMI(MBB, I, DL, get(AArch64::ORRvvv_16B), DestReg)
+        .addReg(SrcReg)
+        .addReg(SrcReg);
+      return;
+    } else {
+      BuildMI(MBB, I, DL, get(AArch64::LSFP128_PreInd_STR), AArch64::XSP)
+        .addReg(SrcReg)
+        .addReg(AArch64::XSP)
+        .addImm(0x1ff & -16);
 
-    BuildMI(MBB, I, DL, get(AArch64::LSFP128_PostInd_LDR), DestReg)
-      .addReg(AArch64::XSP, RegState::Define)
-      .addReg(AArch64::XSP)
-      .addImm(16);
-    return;
+      BuildMI(MBB, I, DL, get(AArch64::LSFP128_PostInd_LDR), DestReg)
+        .addReg(AArch64::XSP, RegState::Define)
+        .addReg(AArch64::XSP)
+        .addImm(16);
+      return;
+    }
   } else {
     llvm_unreachable("Unknown register class in copyPhysReg");
   }

Modified: llvm/branches/release_34/test/CodeGen/AArch64/neon-perm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/release_34/test/CodeGen/AArch64/neon-perm.ll?rev=195758&r1=195757&r2=195758&view=diff
==============================================================================
--- llvm/branches/release_34/test/CodeGen/AArch64/neon-perm.ll (original)
+++ llvm/branches/release_34/test/CodeGen/AArch64/neon-perm.ll Tue Nov 26 06:29:45 2013
@@ -238,6 +238,7 @@ entry:
 define <2 x i64> @test_vuzp2q_s64(<2 x i64> %a, <2 x i64> %b) {
 ; CHECK: test_vuzp2q_s64:
 ; CHECK: ins {{v[0-9]+}}.d[0], {{v[0-9]+}}.d[1]
+; CHECK-NEXT: orr {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
 entry:
   %shuffle.i = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 3>
   ret <2 x i64> %shuffle.i
@@ -294,6 +295,7 @@ entry:
 define <2 x i64> @test_vuzp2q_u64(<2 x i64> %a, <2 x i64> %b) {
 ; CHECK: test_vuzp2q_u64:
 ; CHECK: ins {{v[0-9]+}}.d[0], {{v[0-9]+}}.d[1]
+; CHECK-NEXT: orr {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
 entry:
   %shuffle.i = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 3>
   ret <2 x i64> %shuffle.i
@@ -318,6 +320,7 @@ entry:
 define <2 x double> @test_vuzp2q_f64(<2 x double> %a, <2 x double> %b) {
 ; CHECK: test_vuzp2q_f64:
 ; CHECK: ins {{v[0-9]+}}.d[0], {{v[0-9]+}}.d[1]
+; CHECK-NEXT: orr {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
 entry:
   %shuffle.i = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 1, i32 3>
   ret <2 x double> %shuffle.i