Change Neon VLDn intrinsics to return multiple values instead of really

wide vectors.  Likewise, change VSTn intrinsics to take separate arguments
for each vector in a multi-vector struct.  Adjust tests accordingly.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77468 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 114 insertions, 78 deletions

diff --git a/include/llvm/IntrinsicsARM.td b/include/llvm/IntrinsicsARM.td
index e16797ae70..efe5bff437 100644
--- a/include/llvm/IntrinsicsARM.td
+++ b/include/llvm/IntrinsicsARM.td
@@ -291,20 +291,56 @@ def int_arm_neon_vmovlu : Neon_1Arg_Long_Intrinsic;
 let TargetPrefix = "arm" in {
 
   // De-interleaving vector loads from N-element structures.
-  def int_arm_neon_vldi : Intrinsic<[llvm_anyint_ty],
-                                    [llvm_ptr_ty, llvm_i32_ty],
-                                    [IntrReadArgMem]>;
-  def int_arm_neon_vldf : Intrinsic<[llvm_anyfloat_ty],
-                                    [llvm_ptr_ty, llvm_i32_ty],
-                                    [IntrReadArgMem]>;
+  def int_arm_neon_vld1i : Intrinsic<[llvm_anyint_ty],
+                                     [llvm_ptr_ty], [IntrReadArgMem]>;
+  def int_arm_neon_vld1f : Intrinsic<[llvm_anyfloat_ty],
+                                     [llvm_ptr_ty], [IntrReadArgMem]>;
+  def int_arm_neon_vld2i : Intrinsic<[llvm_anyint_ty, LLVMMatchType<0>],
+                                     [llvm_ptr_ty], [IntrReadArgMem]>;
+  def int_arm_neon_vld2f : Intrinsic<[llvm_anyfloat_ty, LLVMMatchType<0>],
+                                     [llvm_ptr_ty], [IntrReadArgMem]>;
+  def int_arm_neon_vld3i : Intrinsic<[llvm_anyint_ty, LLVMMatchType<0>,
+                                      LLVMMatchType<0>],
+                                     [llvm_ptr_ty], [IntrReadArgMem]>;
+  def int_arm_neon_vld3f : Intrinsic<[llvm_anyfloat_ty, LLVMMatchType<0>,
+                                      LLVMMatchType<0>],
+                                     [llvm_ptr_ty], [IntrReadArgMem]>;
+  def int_arm_neon_vld4i : Intrinsic<[llvm_anyint_ty, LLVMMatchType<0>,
+                                      LLVMMatchType<0>, LLVMMatchType<0>],
+                                     [llvm_ptr_ty], [IntrReadArgMem]>;
+  def int_arm_neon_vld4f : Intrinsic<[llvm_anyfloat_ty, LLVMMatchType<0>,
+                                      LLVMMatchType<0>, LLVMMatchType<0>],
+                                     [llvm_ptr_ty], [IntrReadArgMem]>;
 
   // Interleaving vector stores from N-element structures.
-  def int_arm_neon_vsti : Intrinsic<[llvm_void_ty],
-                                    [llvm_ptr_ty, llvm_anyint_ty, llvm_i32_ty],
-                                    [IntrWriteArgMem]>;
-  def int_arm_neon_vstf : Intrinsic<[llvm_void_ty],
-                                    [llvm_ptr_ty, llvm_anyfloat_ty,llvm_i32_ty],
-                                    [IntrWriteArgMem]>;
+  def int_arm_neon_vst1i : Intrinsic<[llvm_void_ty],
+                                     [llvm_ptr_ty, llvm_anyint_ty],
+                                     [IntrWriteArgMem]>;
+  def int_arm_neon_vst1f : Intrinsic<[llvm_void_ty],
+                                     [llvm_ptr_ty, llvm_anyfloat_ty],
+                                     [IntrWriteArgMem]>;
+  def int_arm_neon_vst2i : Intrinsic<[llvm_void_ty],
+                                     [llvm_ptr_ty, llvm_anyint_ty,
+                                      LLVMMatchType<0>], [IntrWriteArgMem]>;
+  def int_arm_neon_vst2f : Intrinsic<[llvm_void_ty],
+                                     [llvm_ptr_ty, llvm_anyfloat_ty,
+                                      LLVMMatchType<0>], [IntrWriteArgMem]>;
+  def int_arm_neon_vst3i : Intrinsic<[llvm_void_ty],
+                                     [llvm_ptr_ty, llvm_anyint_ty,
+                                      LLVMMatchType<0>, LLVMMatchType<0>],
+                                      [IntrWriteArgMem]>;
+  def int_arm_neon_vst3f : Intrinsic<[llvm_void_ty],
+                                     [llvm_ptr_ty, llvm_anyfloat_ty,
+                                      LLVMMatchType<0>, LLVMMatchType<0>],
+                                      [IntrWriteArgMem]>;
+  def int_arm_neon_vst4i : Intrinsic<[llvm_void_ty],
+                                     [llvm_ptr_ty, llvm_anyint_ty,
+                                      LLVMMatchType<0>, LLVMMatchType<0>,
+                                      LLVMMatchType<0>], [IntrWriteArgMem]>;
+  def int_arm_neon_vst4f : Intrinsic<[llvm_void_ty],
+                                     [llvm_ptr_ty, llvm_anyfloat_ty,
+                                      LLVMMatchType<0>, LLVMMatchType<0>,
+                                      LLVMMatchType<0>], [IntrWriteArgMem]>;
 
   // Vector Table Lookup
   def int_arm_neon_vtbl : Intrinsic<[llvm_v8i8_ty],
diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td
index 9415b40e76..8641d6274a 100644
--- a/lib/Target/ARM/ARMInstrNEON.td
+++ b/lib/Target/ARM/ARMInstrNEON.td
@@ -135,45 +135,45 @@ def VSTRQ : NI<(outs), (ins QPR:$src, GPR:$addr),
 class VLD1D<string OpcodeStr, ValueType Ty, Intrinsic IntOp>
   : NLdSt<(outs DPR:$dst), (ins addrmode6:$addr),
           !strconcat(OpcodeStr, "\t${dst:dregsingle}, $addr"),
-          [(set DPR:$dst, (Ty (IntOp addrmode6:$addr, 1)))]>;
+          [(set DPR:$dst, (Ty (IntOp addrmode6:$addr)))]>;
 class VLD1Q<string OpcodeStr, ValueType Ty, Intrinsic IntOp>
   : NLdSt<(outs QPR:$dst), (ins addrmode6:$addr),
           !strconcat(OpcodeStr, "\t${dst:dregpair}, $addr"),
-          [(set QPR:$dst, (Ty (IntOp addrmode6:$addr, 1)))]>;
+          [(set QPR:$dst, (Ty (IntOp addrmode6:$addr)))]>;
 
-def  VLD1d8   : VLD1D<"vld1.8",  v8i8,  int_arm_neon_vldi>;
-def  VLD1d16  : VLD1D<"vld1.16", v4i16, int_arm_neon_vldi>;
-def  VLD1d32  : VLD1D<"vld1.32", v2i32, int_arm_neon_vldi>;
-def  VLD1df   : VLD1D<"vld1.32", v2f32, int_arm_neon_vldf>;
-def  VLD1d64  : VLD1D<"vld1.64", v1i64, int_arm_neon_vldi>;
+def  VLD1d8   : VLD1D<"vld1.8",  v8i8,  int_arm_neon_vld1i>;
+def  VLD1d16  : VLD1D<"vld1.16", v4i16, int_arm_neon_vld1i>;
+def  VLD1d32  : VLD1D<"vld1.32", v2i32, int_arm_neon_vld1i>;
+def  VLD1df   : VLD1D<"vld1.32", v2f32, int_arm_neon_vld1f>;
+def  VLD1d64  : VLD1D<"vld1.64", v1i64, int_arm_neon_vld1i>;
 
-def  VLD1q8   : VLD1Q<"vld1.8",  v16i8, int_arm_neon_vldi>;
-def  VLD1q16  : VLD1Q<"vld1.16", v8i16, int_arm_neon_vldi>;
-def  VLD1q32  : VLD1Q<"vld1.32", v4i32, int_arm_neon_vldi>;
-def  VLD1qf   : VLD1Q<"vld1.32", v4f32, int_arm_neon_vldf>;
-def  VLD1q64  : VLD1Q<"vld1.64", v2i64, int_arm_neon_vldi>;
+def  VLD1q8   : VLD1Q<"vld1.8",  v16i8, int_arm_neon_vld1i>;
+def  VLD1q16  : VLD1Q<"vld1.16", v8i16, int_arm_neon_vld1i>;
+def  VLD1q32  : VLD1Q<"vld1.32", v4i32, int_arm_neon_vld1i>;
+def  VLD1qf   : VLD1Q<"vld1.32", v4f32, int_arm_neon_vld1f>;
+def  VLD1q64  : VLD1Q<"vld1.64", v2i64, int_arm_neon_vld1i>;
 
 //   VST1     : Vector Store (multiple single elements)
 class VST1D<string OpcodeStr, ValueType Ty, Intrinsic IntOp>
   : NLdSt<(outs), (ins addrmode6:$addr, DPR:$src),
           !strconcat(OpcodeStr, "\t${src:dregsingle}, $addr"),
-          [(IntOp addrmode6:$addr, (Ty DPR:$src), 1)]>;
+          [(IntOp addrmode6:$addr, (Ty DPR:$src))]>;
 class VST1Q<string OpcodeStr, ValueType Ty, Intrinsic IntOp>
   : NLdSt<(outs), (ins addrmode6:$addr, QPR:$src),
           !strconcat(OpcodeStr, "\t${src:dregpair}, $addr"),
-          [(IntOp addrmode6:$addr, (Ty QPR:$src), 1)]>;
-
-def  VST1d8   : VST1D<"vst1.8",  v8i8,  int_arm_neon_vsti>;
-def  VST1d16  : VST1D<"vst1.16", v4i16, int_arm_neon_vsti>;
-def  VST1d32  : VST1D<"vst1.32", v2i32, int_arm_neon_vsti>;
-def  VST1df   : VST1D<"vst1.32", v2f32, int_arm_neon_vstf>;
-def  VST1d64  : VST1D<"vst1.64", v1i64, int_arm_neon_vsti>;
-
-def  VST1q8   : VST1Q<"vst1.8",  v16i8, int_arm_neon_vsti>;
-def  VST1q16  : VST1Q<"vst1.16", v8i16, int_arm_neon_vsti>;
-def  VST1q32  : VST1Q<"vst1.32", v4i32, int_arm_neon_vsti>;
-def  VST1qf   : VST1Q<"vst1.32", v4f32, int_arm_neon_vstf>;
-def  VST1q64  : VST1Q<"vst1.64", v2i64, int_arm_neon_vsti>;
+          [(IntOp addrmode6:$addr, (Ty QPR:$src))]>;
+
+def  VST1d8   : VST1D<"vst1.8",  v8i8,  int_arm_neon_vst1i>;
+def  VST1d16  : VST1D<"vst1.16", v4i16, int_arm_neon_vst1i>;
+def  VST1d32  : VST1D<"vst1.32", v2i32, int_arm_neon_vst1i>;
+def  VST1df   : VST1D<"vst1.32", v2f32, int_arm_neon_vst1f>;
+def  VST1d64  : VST1D<"vst1.64", v1i64, int_arm_neon_vst1i>;
+
+def  VST1q8   : VST1Q<"vst1.8",  v16i8, int_arm_neon_vst1i>;
+def  VST1q16  : VST1Q<"vst1.16", v8i16, int_arm_neon_vst1i>;
+def  VST1q32  : VST1Q<"vst1.32", v4i32, int_arm_neon_vst1i>;
+def  VST1qf   : VST1Q<"vst1.32", v4f32, int_arm_neon_vst1f>;
+def  VST1q64  : VST1Q<"vst1.64", v2i64, int_arm_neon_vst1i>;
 
 
 //===----------------------------------------------------------------------===//
diff --git a/test/CodeGen/ARM/vld1.ll b/test/CodeGen/ARM/vld1.ll
index 161cb71d53..fc925f6068 100644
--- a/test/CodeGen/ARM/vld1.ll
+++ b/test/CodeGen/ARM/vld1.ll
@@ -5,63 +5,63 @@
 ; RUN: grep {vld1\\.64} %t | count 2
 
 define <8 x i8> @vld1i8(i8* %A) nounwind {
-	%tmp1 = call <8 x i8> @llvm.arm.neon.vldi.v8i8(i8* %A, i32 1)
+	%tmp1 = call <8 x i8> @llvm.arm.neon.vld1i.v8i8(i8* %A)
 	ret <8 x i8> %tmp1
 }
 
 define <4 x i16> @vld1i16(i16* %A) nounwind {
-	%tmp1 = call <4 x i16> @llvm.arm.neon.vldi.v4i16(i16* %A, i32 1)
+	%tmp1 = call <4 x i16> @llvm.arm.neon.vld1i.v4i16(i16* %A)
 	ret <4 x i16> %tmp1
 }
 
 define <2 x i32> @vld1i32(i32* %A) nounwind {
-	%tmp1 = call <2 x i32> @llvm.arm.neon.vldi.v2i32(i32* %A, i32 1)
+	%tmp1 = call <2 x i32> @llvm.arm.neon.vld1i.v2i32(i32* %A)
 	ret <2 x i32> %tmp1
 }
 
 define <2 x float> @vld1f(float* %A) nounwind {
-	%tmp1 = call <2 x float> @llvm.arm.neon.vldf.v2f32(float* %A, i32 1)
+	%tmp1 = call <2 x float> @llvm.arm.neon.vld1f.v2f32(float* %A)
 	ret <2 x float> %tmp1
 }
 
 define <1 x i64> @vld1i64(i64* %A) nounwind {
-	%tmp1 = call <1 x i64> @llvm.arm.neon.vldi.v1i64(i64* %A, i32 1)
+	%tmp1 = call <1 x i64> @llvm.arm.neon.vld1i.v1i64(i64* %A)
 	ret <1 x i64> %tmp1
 }
 
 define <16 x i8> @vld1Qi8(i8* %A) nounwind {
-	%tmp1 = call <16 x i8> @llvm.arm.neon.vldi.v16i8(i8* %A, i32 1)
+	%tmp1 = call <16 x i8> @llvm.arm.neon.vld1i.v16i8(i8* %A)
 	ret <16 x i8> %tmp1
 }
 
 define <8 x i16> @vld1Qi16(i16* %A) nounwind {
-	%tmp1 = call <8 x i16> @llvm.arm.neon.vldi.v8i16(i16* %A, i32 1)
+	%tmp1 = call <8 x i16> @llvm.arm.neon.vld1i.v8i16(i16* %A)
 	ret <8 x i16> %tmp1
 }
 
 define <4 x i32> @vld1Qi32(i32* %A) nounwind {
-	%tmp1 = call <4 x i32> @llvm.arm.neon.vldi.v4i32(i32* %A, i32 1)
+	%tmp1 = call <4 x i32> @llvm.arm.neon.vld1i.v4i32(i32* %A)
 	ret <4 x i32> %tmp1
 }
 
 define <4 x float> @vld1Qf(float* %A) nounwind {
-	%tmp1 = call <4 x float> @llvm.arm.neon.vldf.v4f32(float* %A, i32 1)
+	%tmp1 = call <4 x float> @llvm.arm.neon.vld1f.v4f32(float* %A)
 	ret <4 x float> %tmp1
 }
 
 define <2 x i64> @vld1Qi64(i64* %A) nounwind {
-	%tmp1 = call <2 x i64> @llvm.arm.neon.vldi.v2i64(i64* %A, i32 1)
+	%tmp1 = call <2 x i64> @llvm.arm.neon.vld1i.v2i64(i64* %A)
 	ret <2 x i64> %tmp1
 }
 
-declare <8 x i8>  @llvm.arm.neon.vldi.v8i8(i8*, i32) nounwind readnone
-declare <4 x i16> @llvm.arm.neon.vldi.v4i16(i16*, i32) nounwind readnone
-declare <2 x i32> @llvm.arm.neon.vldi.v2i32(i32*, i32) nounwind readnone
-declare <2 x float> @llvm.arm.neon.vldf.v2f32(float*, i32) nounwind readnone
-declare <1 x i64> @llvm.arm.neon.vldi.v1i64(i64*, i32) nounwind readnone
+declare <8 x i8>  @llvm.arm.neon.vld1i.v8i8(i8*) nounwind readnone
+declare <4 x i16> @llvm.arm.neon.vld1i.v4i16(i16*) nounwind readnone
+declare <2 x i32> @llvm.arm.neon.vld1i.v2i32(i32*) nounwind readnone
+declare <2 x float> @llvm.arm.neon.vld1f.v2f32(float*) nounwind readnone
+declare <1 x i64> @llvm.arm.neon.vld1i.v1i64(i64*) nounwind readnone
 
-declare <16 x i8> @llvm.arm.neon.vldi.v16i8(i8*, i32) nounwind readnone
-declare <8 x i16> @llvm.arm.neon.vldi.v8i16(i16*, i32) nounwind readnone
-declare <4 x i32> @llvm.arm.neon.vldi.v4i32(i32*, i32) nounwind readnone
-declare <4 x float> @llvm.arm.neon.vldf.v4f32(float*, i32) nounwind readnone
-declare <2 x i64> @llvm.arm.neon.vldi.v2i64(i64*, i32) nounwind readnone
+declare <16 x i8> @llvm.arm.neon.vld1i.v16i8(i8*) nounwind readnone
+declare <8 x i16> @llvm.arm.neon.vld1i.v8i16(i16*) nounwind readnone
+declare <4 x i32> @llvm.arm.neon.vld1i.v4i32(i32*) nounwind readnone
+declare <4 x float> @llvm.arm.neon.vld1f.v4f32(float*) nounwind readnone
+declare <2 x i64> @llvm.arm.neon.vld1i.v2i64(i64*) nounwind readnone
diff --git a/test/CodeGen/ARM/vst1.ll b/test/CodeGen/ARM/vst1.ll
index 70a05fa803..c4099a50ef 100644
--- a/test/CodeGen/ARM/vst1.ll
+++ b/test/CodeGen/ARM/vst1.ll
@@ -6,72 +6,72 @@
 
 define void @vst1i8(i8* %A, <8 x i8>* %B) nounwind {
 	%tmp1 = load <8 x i8>* %B
-	call void @llvm.arm.neon.vsti.v8i8(i8* %A, <8 x i8> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1i.v8i8(i8* %A, <8 x i8> %tmp1)
 	ret void
 }
 
 define void @vst1i16(i16* %A, <4 x i16>* %B) nounwind {
 	%tmp1 = load <4 x i16>* %B
-	call void @llvm.arm.neon.vsti.v4i16(i16* %A, <4 x i16> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1i.v4i16(i16* %A, <4 x i16> %tmp1)
 	ret void
 }
 
 define void @vst1i32(i32* %A, <2 x i32>* %B) nounwind {
 	%tmp1 = load <2 x i32>* %B
-	call void @llvm.arm.neon.vsti.v2i32(i32* %A, <2 x i32> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1i.v2i32(i32* %A, <2 x i32> %tmp1)
 	ret void
 }
 
 define void @vst1f(float* %A, <2 x float>* %B) nounwind {
 	%tmp1 = load <2 x float>* %B
-	call void @llvm.arm.neon.vstf.v2f32(float* %A, <2 x float> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1f.v2f32(float* %A, <2 x float> %tmp1)
 	ret void
 }
 
 define void @vst1i64(i64* %A, <1 x i64>* %B) nounwind {
 	%tmp1 = load <1 x i64>* %B
-	call void @llvm.arm.neon.vsti.v1i64(i64* %A, <1 x i64> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1i.v1i64(i64* %A, <1 x i64> %tmp1)
 	ret void
 }
 
 define void @vst1Qi8(i8* %A, <16 x i8>* %B) nounwind {
 	%tmp1 = load <16 x i8>* %B
-	call void @llvm.arm.neon.vsti.v16i8(i8* %A, <16 x i8> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1i.v16i8(i8* %A, <16 x i8> %tmp1)
 	ret void
 }
 
 define void @vst1Qi16(i16* %A, <8 x i16>* %B) nounwind {
 	%tmp1 = load <8 x i16>* %B
-	call void @llvm.arm.neon.vsti.v8i16(i16* %A, <8 x i16> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1i.v8i16(i16* %A, <8 x i16> %tmp1)
 	ret void
 }
 
 define void @vst1Qi32(i32* %A, <4 x i32>* %B) nounwind {
 	%tmp1 = load <4 x i32>* %B
-	call void @llvm.arm.neon.vsti.v4i32(i32* %A, <4 x i32> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1i.v4i32(i32* %A, <4 x i32> %tmp1)
 	ret void
 }
 
 define void @vst1Qf(float* %A, <4 x float>* %B) nounwind {
 	%tmp1 = load <4 x float>* %B
-	call void @llvm.arm.neon.vstf.v4f32(float* %A, <4 x float> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1f.v4f32(float* %A, <4 x float> %tmp1)
 	ret void
 }
 
 define void @vst1Qi64(i64* %A, <2 x i64>* %B) nounwind {
 	%tmp1 = load <2 x i64>* %B
-	call void @llvm.arm.neon.vsti.v2i64(i64* %A, <2 x i64> %tmp1, i32 1)
+	call void @llvm.arm.neon.vst1i.v2i64(i64* %A, <2 x i64> %tmp1)
 	ret void
 }
 
-declare void @llvm.arm.neon.vsti.v8i8(i8*, <8 x i8>, i32) nounwind readnone
-declare void @llvm.arm.neon.vsti.v4i16(i16*, <4 x i16>, i32) nounwind readnone
-declare void @llvm.arm.neon.vsti.v2i32(i32*, <2 x i32>, i32) nounwind readnone
-declare void @llvm.arm.neon.vstf.v2f32(float*, <2 x float>, i32) nounwind readnone
-declare void @llvm.arm.neon.vsti.v1i64(i64*, <1 x i64>, i32) nounwind readnone
+declare void @llvm.arm.neon.vst1i.v8i8(i8*, <8 x i8>) nounwind readnone
+declare void @llvm.arm.neon.vst1i.v4i16(i16*, <4 x i16>) nounwind readnone
+declare void @llvm.arm.neon.vst1i.v2i32(i32*, <2 x i32>) nounwind readnone
+declare void @llvm.arm.neon.vst1f.v2f32(float*, <2 x float>) nounwind readnone
+declare void @llvm.arm.neon.vst1i.v1i64(i64*, <1 x i64>) nounwind readnone
 
-declare void @llvm.arm.neon.vsti.v16i8(i8*, <16 x i8>, i32) nounwind readnone
-declare void @llvm.arm.neon.vsti.v8i16(i16*, <8 x i16>, i32) nounwind readnone
-declare void @llvm.arm.neon.vsti.v4i32(i32*, <4 x i32>, i32) nounwind readnone
-declare void @llvm.arm.neon.vstf.v4f32(float*, <4 x float>, i32) nounwind readnone
-declare void @llvm.arm.neon.vsti.v2i64(i64*, <2 x i64>, i32) nounwind readnone
+declare void @llvm.arm.neon.vst1i.v16i8(i8*, <16 x i8>) nounwind readnone
+declare void @llvm.arm.neon.vst1i.v8i16(i16*, <8 x i16>) nounwind readnone
+declare void @llvm.arm.neon.vst1i.v4i32(i32*, <4 x i32>) nounwind readnone
+declare void @llvm.arm.neon.vst1f.v4f32(float*, <4 x float>) nounwind readnone
+declare void @llvm.arm.neon.vst1i.v2i64(i64*, <2 x i64>) nounwind readnone