[Clang - ARM] - Make fp16 NEON intrinsics available in AArch32.

[LukeCheeseman]

Summary

This patch makes the NEON intrinsics vget_lane_f16, vgetq_lane_f16, vset_lane_16 and vset1_lane_f16 available in AArch32.

[LukeCheeseman]

(dummy comment to get phabricator to send emails)

[t.p.northover]

I think this looks reasonable.

Tim.

cfe/trunk/include/clang/Basic/arm_neon.td

 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Vector fused multiply-add operations
 
 def VFMA : SInst<"vfma", "dddd", "fQf">;
 
 ////////////////////////////////////////////////////////////////////////////////
+// fp16 vector operations
+def SCALAR_HALF_GET_LANE : IOpInst<"vget_lane", "sdi", "h", OP_SCALAR_HALF_GET_LN>;
+def SCALAR_HALF_SET_LANE : IOpInst<"vset_lane", "dsdi", "h", OP_SCALAR_HALF_SET_LN>;
+def SCALAR_HALF_GET_LANEQ : IOpInst<"vget_lane", "sdi", "Qh", OP_SCALAR_HALF_GET_LNQ>;
+def SCALAR_HALF_SET_LANEQ : IOpInst<"vset_lane", "dsdi", "Qh", OP_SCALAR_HALF_SET_LNQ>;
+
+////////////////////////////////////////////////////////////////////////////////
 // AArch64 Intrinsics
 
 let ArchGuard = "defined(__aarch64__)" in {
 
 ////////////////////////////////////////////////////////////////////////////////
 // Load/Store
 def LD1 : WInst<"vld1", "dc", "dQdPlQPl">;
 def LD2 : WInst<"vld2", "2c", "QUlQldQdPlQPl">;

 def SCALAR_SQDMULH_LANEQ : SOpInst<"vqdmulh_laneq", "ssji", "SsSi", OP_SCALAR_QDMULH_LN>;
 
 // Scalar Integer Saturating Rounding Doubling Multiply Half High
 def SCALAR_SQRDMULH_LANE : SOpInst<"vqrdmulh_lane", "ssdi", "SsSi", OP_SCALAR_QRDMULH_LN>;
 def SCALAR_SQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "ssji", "SsSi", OP_SCALAR_QRDMULH_LN>;
 
 def SCALAR_VDUP_LANE : IInst<"vdup_lane", "sdi", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
 def SCALAR_VDUP_LANEQ : IInst<"vdup_laneq", "sji", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
-
-// FIXME: Rename so it is obvious this only applies to halfs.
-def SCALAR_HALF_GET_LANE : IOpInst<"vget_lane", "sdi", "h", OP_SCALAR_HALF_GET_LN>;
-def SCALAR_HALF_SET_LANE : IOpInst<"vset_lane", "dsdi", "h", OP_SCALAR_HALF_SET_LN>;
-def SCALAR_HALF_GET_LANEQ : IOpInst<"vget_lane", "sdi", "Qh", OP_SCALAR_HALF_GET_LNQ>;
-def SCALAR_HALF_SET_LANEQ : IOpInst<"vset_lane", "dsdi", "Qh", OP_SCALAR_HALF_SET_LNQ>;
 }

cfe/trunk/test/CodeGen/arm_neon_intrinsics.c

 }
 
 // CHECK-LABEL: test_vget_lane_f32
 // CHECK: vmov 
 float32_t test_vget_lane_f32(float32x2_t a) {
   return vget_lane_f32(a, 1);
 }
 
+// CHECK-LABEL: test_vget_lane_f16
+// CHECK: vmov 
+float32_t test_vget_lane_f16(float16x4_t a) {
+  return vget_lane_f16(a, 1);
+}
+
 // CHECK-LABEL: test_vgetq_lane_u8
 // CHECK: vmov 
 uint8_t test_vgetq_lane_u8(uint8x16_t a) {
   return vgetq_lane_u8(a, 15);
 }
 
 // CHECK-LABEL: test_vgetq_lane_u16
 // CHECK: vmov 

 }
 
 // CHECK-LABEL: test_vgetq_lane_f32
 // CHECK: vmov 
 float32_t test_vgetq_lane_f32(float32x4_t a) {
   return vgetq_lane_f32(a, 3);
 }
 
+// CHECK-LABEL: test_vgetq_lane_f16
+// CHECK: vmov 
+float32_t test_vgetq_lane_f16(float16x8_t a) {
+  return vgetq_lane_f16(a, 3);
+}
+
 // CHECK-LABEL: test_vget_lane_s64
 // The optimizer is able to remove all moves now.
 int64_t test_vget_lane_s64(int64x1_t a) {
   return vget_lane_s64(a, 0);
 }
 
 // CHECK-LABEL: test_vget_lane_u64
 // The optimizer is able to remove all moves now.

 }
 
 // CHECK-LABEL: test_vset_lane_f32
 // CHECK: mov 
 float32x2_t test_vset_lane_f32(float32_t a, float32x2_t b) {
   return vset_lane_f32(a, b, 1);
 }
 
+// CHECK-LABEL: test_vset_lane_f16
+// CHECK: mov 
+float16x4_t test_vset_lane_f16(float16_t *a, float16x4_t b) {
+  return vset_lane_f16(*a, b, 1);
+}
+
 // CHECK-LABEL: test_vsetq_lane_u8
 // CHECK: vmov 
 uint8x16_t test_vsetq_lane_u8(uint8_t a, uint8x16_t b) {
   return vsetq_lane_u8(a, b, 15);
 }
 
 // CHECK-LABEL: test_vsetq_lane_u16
 // CHECK: vmov 

 }
 
 // CHECK-LABEL: test_vsetq_lane_f32
 // CHECK: vmov 
 float32x4_t test_vsetq_lane_f32(float32_t a, float32x4_t b) {
   return vsetq_lane_f32(a, b, 3);
 }
 
+// CHECK-LABEL: test_vsetq_lane_f16
+// CHECK: vmov 
+float16x8_t test_vsetq_lane_f16(float16_t *a, float16x8_t b) {
+  return vsetq_lane_f16(*a, b, 3);
+}
+
 // CHECK-LABEL: test_vset_lane_s64
 // The optimizer is able to get rid of all moves now.
 int64x1_t test_vset_lane_s64(int64_t a, int64x1_t b) {
   return vset_lane_s64(a, b, 0);
 }
 
 // CHECK-LABEL: test_vset_lane_u64
 // The optimizer is able to get rid of all moves now.