r196153 - [AArch64] Implemented vcopy_lane patterns using scalar DUP instruction.

[Chad Rosier]

Author: mcrosier
Date: Mon Dec  2 15:07:27 2013
New Revision: 196153

URL: http://llvm.org/viewvc/llvm-project?rev=196153&view=rev
Log:
[AArch64] Implemented vcopy_lane patterns using scalar DUP instruction.
Patch by Ana Pazos!

Modified:
    cfe/trunk/include/clang/Basic/arm_neon.td
    cfe/trunk/test/CodeGen/aarch64-neon-copy.c

Modified: cfe/trunk/include/clang/Basic/arm_neon.td
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/arm_neon.td?rev=196153&r1=196152&r2=196153&view=diff
==============================================================================
--- cfe/trunk/include/clang/Basic/arm_neon.td (original)
+++ cfe/trunk/include/clang/Basic/arm_neon.td Mon Dec  2 15:07:27 2013
@@ -836,13 +836,13 @@ def GET_LANE : IInst<"vget_lane", "sdi",
 def SET_LANE : IInst<"vset_lane", "dsdi",
                      "csilPcPsUcUsUiUlQcQsQiQlQUcQUsQUiQUlPcPsQPcQPsfdQfQdPlQPl">;
 def COPY_LANE : IOpInst<"vcopy_lane", "ddidi",
-                        "csiPcPsUcUsUiPcPsfPl", OP_COPY_LN>;
+                        "csilPcPsUcUsUiUlPcPsPlfd", OP_COPY_LN>;
 def COPYQ_LANE : IOpInst<"vcopy_lane", "ddigi",
                         "QcQsQiQlQUcQUsQUiQUlQPcQPsQfQdQPl", OP_COPYQ_LN>;
 def COPY_LANEQ : IOpInst<"vcopy_laneq", "ddiki",
-                     "csiPcPsUcUsUif", OP_COPY_LNQ>;
+                     "csilPcPsPlUcUsUiUlf", OP_COPY_LNQ>;
 def COPYQ_LANEQ : IOpInst<"vcopy_laneq", "ddidi",
-                     "QcQsQiQlQUcQUsQUiQUlQPcQPsQfdQPl", OP_COPY_LN>;
+                     "QcQsQiQlQUcQUsQUiQUlQPcQPsQfdQdQPl", OP_COPY_LN>;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Set all lanes to same value

Modified: cfe/trunk/test/CodeGen/aarch64-neon-copy.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/aarch64-neon-copy.c?rev=196153&r1=196152&r2=196153&view=diff
==============================================================================
--- cfe/trunk/test/CodeGen/aarch64-neon-copy.c (original)
+++ cfe/trunk/test/CodeGen/aarch64-neon-copy.c Mon Dec  2 15:07:27 2013
@@ -1192,3 +1192,56 @@ float64x2_t test_vmovq_n_f64(float64_t v
   return vmovq_n_f64(v1);
   // CHECK: dup {{v[0-9]+}}.2d, {{v[0-9]+}}.d[0]
 }
+
+// CHECK: test_vcopy_lane_s64
+int64x1_t test_vcopy_lane_s64(int64x1_t a, int64x1_t c) {
+  return vcopy_lane_s64(a, 0, c, 0);
+// CHECK: fmov {{d[0-9]+}}, {{d[0-9]+}}
+// CHECK-NOT: dup {{d[0-9]+}}, {{v[0-9]+}}.d[0]
+}
+
+// CHECK: test_vcopy_lane_u64
+uint64x1_t test_vcopy_lane_u64(uint64x1_t a, uint64x1_t c) {
+  return vcopy_lane_u64(a, 0, c, 0);
+// CHECK: fmov {{d[0-9]+}}, {{d[0-9]+}}
+// CHECK-NOT: dup {{d[0-9]+}}, {{v[0-9]+}}.d[0]
+}
+
+// CHECK: test_vcopy_lane_f64
+float64x1_t test_vcopy_lane_f64(float64x1_t a, float64x1_t c) {
+  return vcopy_lane_f64(a, 0, c, 0);
+// CHECK: fmov {{d[0-9]+}}, {{d[0-9]+}}
+// CHECK-NOT: dup {{d[0-9]+}}, {{v[0-9]+}}.d[0]
+}
+
+// CHECK: test_vcopy_laneq_s64
+int64x1_t test_vcopy_laneq_s64(int64x1_t a, int64x2_t c) {
+  return vcopy_laneq_s64(a, 0, c, 1);
+// CHECK: dup {{d[0-9]+}}, {{v[0-9]+}}.d[1]
+}
+
+// CHECK: test_vcopy_laneq_u64
+uint64x1_t test_vcopy_laneq_u64(uint64x1_t a, uint64x2_t c) {
+  return vcopy_laneq_u64(a, 0, c, 1);
+// CHECK: dup {{d[0-9]+}}, {{v[0-9]+}}.d[1]
+}
+
+// CHECK: test_vcopy_laneq_f64
+float64x1_t test_vcopy_laneq_f64(float64x1_t a, float64x1_t c) {
+  return vcopy_laneq_f64(a, 0, c, 0);
+// CHECK: fmov {{d[0-9]+}}, {{d[0-9]+}}
+// CHECK-NOT: dup {{d[0-9]+}}, {{v[0-9]+}}.d[0]
+}
+
+// CHECK: test_vcopy_laneq_p64
+poly64x1_t test_vcopy_laneq_p64(poly64x1_t a, poly64x2_t c) {
+  return vcopy_laneq_p64(a, 0, c, 1);
+// CHECK: dup {{d[0-9]+}}, {{v[0-9]+}}.d[1]
+}
+
+// CHECK: test_vcopyq_laneq_f64
+float64x2_t test_vcopyq_laneq_f64(float64x2_t a, float64x2_t c) {
+// CHECK: ins  {{v[0-9]+}}.d[1], {{v[0-9]+}}.d[1]
+  return vcopyq_laneq_f64(a, 1, c, 1);
+}
+