Adding more arithmetic operators to MMX. This is an almost exact copy of

the addition. Please let me know if you have suggestions.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@35055 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 94 insertions, 15 deletions

diff --git a/include/llvm/IntrinsicsX86.td b/include/llvm/IntrinsicsX86.td
index 76db55ab29..48aa1a7120 100644
--- a/include/llvm/IntrinsicsX86.td
+++ b/include/llvm/IntrinsicsX86.td
@@ -547,6 +547,7 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
 
 // Integer arithmetic ops.
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
+  // Addition
   def int_x86_mmx_padds_b : GCCBuiltin<"__builtin_ia32_paddsb">,
               Intrinsic<[llvm_v8i8_ty, llvm_v8i8_ty,
                          llvm_v8i8_ty], [IntrNoMem]>;
@@ -560,4 +561,19 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_mmx_paddus_w : GCCBuiltin<"__builtin_ia32_paddusw">,
               Intrinsic<[llvm_v4i16_ty, llvm_v4i16_ty,
                          llvm_v4i16_ty], [IntrNoMem]>;
+
+  // Subtraction
+  def int_x86_mmx_psubs_b : GCCBuiltin<"__builtin_ia32_psubsb">,
+              Intrinsic<[llvm_v8i8_ty, llvm_v8i8_ty,
+                         llvm_v8i8_ty], [IntrNoMem]>;
+  def int_x86_mmx_psubs_w : GCCBuiltin<"__builtin_ia32_psubsw">,
+              Intrinsic<[llvm_v4i16_ty, llvm_v4i16_ty,
+                         llvm_v4i16_ty], [IntrNoMem]>;
+
+  def int_x86_mmx_psubus_b : GCCBuiltin<"__builtin_ia32_psubusb">,
+              Intrinsic<[llvm_v8i8_ty, llvm_v8i8_ty,
+                         llvm_v8i8_ty], [IntrNoMem]>;
+  def int_x86_mmx_psubus_w : GCCBuiltin<"__builtin_ia32_psubusw">,
+              Intrinsic<[llvm_v4i16_ty, llvm_v4i16_ty,
+                         llvm_v4i16_ty], [IntrNoMem]>;
 }
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 2ab5e08619..4fecb70cd2 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -331,6 +331,10 @@ X86TargetLowering::X86TargetLowering(TargetMachine &TM)
     setOperationAction(ISD::ADD,                MVT::v4i16, Legal);
     setOperationAction(ISD::ADD,                MVT::v2i32, Legal);
 
+    setOperationAction(ISD::SUB,                MVT::v8i8,  Legal);
+    setOperationAction(ISD::SUB,                MVT::v4i16, Legal);
+    setOperationAction(ISD::SUB,                MVT::v2i32, Legal);
+
     setOperationAction(ISD::LOAD,               MVT::v8i8,  Promote);
     AddPromotedToType (ISD::LOAD,               MVT::v8i8,  MVT::v2i32);
     setOperationAction(ISD::LOAD,               MVT::v4i16, Promote);
diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td
index 5096c203ba..73f2f11a86 100644
--- a/lib/Target/X86/X86InstrMMX.td
+++ b/lib/Target/X86/X86InstrMMX.td
@@ -101,6 +101,16 @@ defm MMX_PADDSW  : MMXI_binop_rm_int<0xED, "paddsw" , int_x86_mmx_padds_w, 1>;
 defm MMX_PADDUSB : MMXI_binop_rm_int<0xDC, "paddusb", int_x86_mmx_paddus_b, 1>;
 defm MMX_PADDUSW : MMXI_binop_rm_int<0xDD, "paddusw", int_x86_mmx_paddus_w, 1>;
 
+defm MMX_PSUBB : MMXI_binop_rm<0xF8, "psubb", sub, v8i8>;
+defm MMX_PSUBW : MMXI_binop_rm<0xF9, "psubw", sub, v4i16>;
+defm MMX_PSUBD : MMXI_binop_rm<0xFA, "psubd", sub, v2i32>;
+
+defm MMX_PSUBSB  : MMXI_binop_rm_int<0xE8, "psubsb" , int_x86_mmx_psubs_b>;
+defm MMX_PSUBSW  : MMXI_binop_rm_int<0xE9, "psubsw" , int_x86_mmx_psubs_w>;
+
+defm MMX_PSUBUSB : MMXI_binop_rm_int<0xD8, "psubusb", int_x86_mmx_psubus_b>;
+defm MMX_PSUBUSW : MMXI_binop_rm_int<0xD9, "psubusw", int_x86_mmx_psubus_w>;
+
 // Move Instructions
 def MOVD64rr : MMXI<0x6E, MRMSrcReg, (ops VR64:$dst, GR32:$src),
                     "movd {$src, $dst|$dst, $src}", []>;
diff --git a/test/CodeGen/X86/mmx-arith.ll b/test/CodeGen/X86/mmx-arith.ll
index 3911be60b3..7c71cfc9f1 100644
--- a/test/CodeGen/X86/mmx-arith.ll
+++ b/test/CodeGen/X86/mmx-arith.ll
@@ -2,34 +2,83 @@
 
 ;; A basic sanity check to make sure that MMX arithmetic actually compiles.
 
-define void @foo(<2 x i32>* %A, <2 x i32>* %B) {
+define void @foo(<8 x i8>* %A, <8 x i8>* %B) {
 entry:
-	%tmp1 = load <2 x i32>* %A		; <<2 x i32>> [#uses=1]
-	%tmp3 = load <2 x i32>* %B		; <<2 x i32>> [#uses=1]
-	%tmp4 = add <2 x i32> %tmp1, %tmp3		; <<2 x i32>> [#uses=1]
-	store <2 x i32> %tmp4, <2 x i32>* %A
+	%tmp5 = load <8 x i8>* %A		; <<8 x i8>> [#uses=1]
+	%tmp7 = load <8 x i8>* %B		; <<8 x i8>> [#uses=1]
+	%tmp8 = add <8 x i8> %tmp5, %tmp7		; <<8 x i8>> [#uses=2]
+	store <8 x i8> %tmp8, <8 x i8>* %A
+	%tmp14 = load <8 x i8>* %B		; <<8 x i8>> [#uses=1]
+	%tmp25 = tail call <8 x i8> @llvm.x86.mmx.padds.b( <8 x i8> %tmp14, <8 x i8> %tmp8 )		; <<8 x i8>> [#uses=2]
+	store <8 x i8> %tmp25, <8 x i8>* %B
+	%tmp36 = load <8 x i8>* %A		; <<8 x i8>> [#uses=1]
+	%tmp49 = tail call <8 x i8> @llvm.x86.mmx.paddus.b( <8 x i8> %tmp36, <8 x i8> %tmp25 )		; <<8 x i8>> [#uses=2]
+	store <8 x i8> %tmp49, <8 x i8>* %B
+	%tmp58 = load <8 x i8>* %A		; <<8 x i8>> [#uses=1]
+	%tmp61 = sub <8 x i8> %tmp58, %tmp49		; <<8 x i8>> [#uses=2]
+	store <8 x i8> %tmp61, <8 x i8>* %B
+	%tmp64 = load <8 x i8>* %A		; <<8 x i8>> [#uses=1]
+	%tmp80 = tail call <8 x i8> @llvm.x86.mmx.psubs.b( <8 x i8> %tmp61, <8 x i8> %tmp64 )		; <<8 x i8>> [#uses=2]
+	store <8 x i8> %tmp80, <8 x i8>* %A
+	%tmp89 = load <8 x i8>* %B		; <<8 x i8>> [#uses=1]
+	%tmp105 = tail call <8 x i8> @llvm.x86.mmx.psubus.b( <8 x i8> %tmp80, <8 x i8> %tmp89 )		; <<8 x i8>> [#uses=1]
+	store <8 x i8> %tmp105, <8 x i8>* %A
 	tail call void @llvm.x86.mmx.emms( )
 	ret void
 }
 
-define void @bar(<4 x i16>* %A, <4 x i16>* %B) {
+define void @baz(<2 x i32>* %A, <2 x i32>* %B) {
 entry:
-	%tmp1 = load <4 x i16>* %A		; <<4 x i16>> [#uses=1]
-	%tmp3 = load <4 x i16>* %B		; <<4 x i16>> [#uses=1]
-	%tmp4 = add <4 x i16> %tmp1, %tmp3		; <<4 x i16>> [#uses=1]
-	store <4 x i16> %tmp4, <4 x i16>* %A
+	%tmp1 = load <2 x i32>* %A		; <<2 x i32>> [#uses=1]
+	%tmp3 = load <2 x i32>* %B		; <<2 x i32>> [#uses=1]
+	%tmp4 = add <2 x i32> %tmp1, %tmp3		; <<2 x i32>> [#uses=2]
+	store <2 x i32> %tmp4, <2 x i32>* %A
+	%tmp9 = load <2 x i32>* %B		; <<2 x i32>> [#uses=1]
+	%tmp10 = sub <2 x i32> %tmp4, %tmp9		; <<2 x i32>> [#uses=1]
+	store <2 x i32> %tmp10, <2 x i32>* %B
 	tail call void @llvm.x86.mmx.emms( )
 	ret void
 }
 
-define void @baz(<8 x i8>* %A, <8 x i8>* %B) {
+define void @bar(<4 x i16>* %A, <4 x i16>* %B) {
 entry:
-	%tmp1 = load <8 x i8>* %A		; <<8 x i8>> [#uses=1]
-	%tmp3 = load <8 x i8>* %B		; <<8 x i8>> [#uses=1]
-	%tmp4 = add <8 x i8> %tmp1, %tmp3		; <<8 x i8>> [#uses=1]
-	store <8 x i8> %tmp4, <8 x i8>* %A
+	%tmp5 = load <4 x i16>* %A		; <<4 x i16>> [#uses=1]
+	%tmp7 = load <4 x i16>* %B		; <<4 x i16>> [#uses=1]
+	%tmp8 = add <4 x i16> %tmp5, %tmp7		; <<4 x i16>> [#uses=2]
+	store <4 x i16> %tmp8, <4 x i16>* %A
+	%tmp14 = load <4 x i16>* %B		; <<4 x i16>> [#uses=1]
+	%tmp25 = tail call <4 x i16> @llvm.x86.mmx.padds.w( <4 x i16> %tmp14, <4 x i16> %tmp8 )		; <<4 x i16>> [#uses=2]
+	store <4 x i16> %tmp25, <4 x i16>* %B
+	%tmp36 = load <4 x i16>* %A		; <<4 x i16>> [#uses=1]
+	%tmp49 = tail call <4 x i16> @llvm.x86.mmx.paddus.w( <4 x i16> %tmp36, <4 x i16> %tmp25 )		; <<4 x i16>> [#uses=2]
+	store <4 x i16> %tmp49, <4 x i16>* %B
+	%tmp58 = load <4 x i16>* %A		; <<4 x i16>> [#uses=1]
+	%tmp61 = sub <4 x i16> %tmp58, %tmp49		; <<4 x i16>> [#uses=2]
+	store <4 x i16> %tmp61, <4 x i16>* %B
+	%tmp64 = load <4 x i16>* %A		; <<4 x i16>> [#uses=1]
+	%tmp80 = tail call <4 x i16> @llvm.x86.mmx.psubs.w( <4 x i16> %tmp61, <4 x i16> %tmp64 )		; <<4 x i16>> [#uses=2]
+	store <4 x i16> %tmp80, <4 x i16>* %A
+	%tmp89 = load <4 x i16>* %B		; <<4 x i16>> [#uses=1]
+	%tmp105 = tail call <4 x i16> @llvm.x86.mmx.psubus.w( <4 x i16> %tmp80, <4 x i16> %tmp89 )		; <<4 x i16>> [#uses=1]
+	store <4 x i16> %tmp105, <4 x i16>* %A
 	tail call void @llvm.x86.mmx.emms( )
 	ret void
 }
 
+declare <4 x i16> @llvm.x86.mmx.padds.w(<4 x i16>, <4 x i16>)
+
+declare <4 x i16> @llvm.x86.mmx.paddus.w(<4 x i16>, <4 x i16>)
+
+declare <4 x i16> @llvm.x86.mmx.psubs.w(<4 x i16>, <4 x i16>)
+
+declare <4 x i16> @llvm.x86.mmx.psubus.w(<4 x i16>, <4 x i16>)
+
+declare <8 x i8> @llvm.x86.mmx.padds.b(<8 x i8>, <8 x i8>)
+
+declare <8 x i8> @llvm.x86.mmx.paddus.b(<8 x i8>, <8 x i8>)
+
+declare <8 x i8> @llvm.x86.mmx.psubs.b(<8 x i8>, <8 x i8>)
+
+declare <8 x i8> @llvm.x86.mmx.psubus.b(<8 x i8>, <8 x i8>)
+
 declare void @llvm.x86.mmx.emms()