Add missed extract_element pattern
authorAnton Korobeynikov <asl@math.spbu.ru>
Fri, 28 Aug 2009 23:41:26 +0000 (23:41 +0000)
committerAnton Korobeynikov <asl@math.spbu.ru>
Fri, 28 Aug 2009 23:41:26 +0000 (23:41 +0000)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@80408 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/ARM/ARMInstrNEON.td
test/CodeGen/ARM/2009-08-29-ExtractEltf32.ll [new file with mode: 0644]

index d473fa78e771630d390727b71fab857da2fe512c..6775591d9f0d14521d42f7a4600404af7ffa5647 100644 (file)
@@ -284,7 +284,7 @@ def DSubReg_f64_reg : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
 }]>;
 
-// Extract S sub-registers of Q registers.
+// Extract S sub-registers of Q/D registers.
 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
 def SSubReg_f32_reg : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
@@ -1672,6 +1672,8 @@ def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
           (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
                              (DSubReg_i32_reg imm:$lane))),
                      (SubReg_i32_lane imm:$lane))>;
+def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
+          (EXTRACT_SUBREG DPR:$src1, (SSubReg_f32_reg imm:$src2))>;
 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
           (EXTRACT_SUBREG QPR:$src1, (SSubReg_f32_reg imm:$src2))>;
 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
diff --git a/test/CodeGen/ARM/2009-08-29-ExtractEltf32.ll b/test/CodeGen/ARM/2009-08-29-ExtractEltf32.ll
new file mode 100644 (file)
index 0000000..4a044a5
--- /dev/null
@@ -0,0 +1,25 @@
+; RUN: llvm-as < %s | llc -mattr=+neon
+target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:64:64-v128:128:128-a0:0:32"
+target triple = "thumbv7-elf"
+
+define arm_apcscc void @foo() nounwind {
+entry:
+  %0 = tail call <2 x float> @llvm.arm.neon.vpadd.v2f32(<2 x float> undef, <2 x float> undef) nounwind ; <<2 x float>> [#uses=1]
+  %tmp28 = extractelement <2 x float> %0, i32 0   ; <float> [#uses=1]
+  %1 = fcmp une float %tmp28, 4.900000e+01        ; <i1> [#uses=1]
+  br i1 %1, label %bb, label %bb7
+
+bb:                                               ; preds = %entry
+  unreachable
+
+bb7:                                              ; preds = %entry
+  br i1 undef, label %bb8, label %bb9
+
+bb8:                                              ; preds = %bb7
+  unreachable
+
+bb9:                                              ; preds = %bb7
+  ret void
+}
+
+declare <2 x float> @llvm.arm.neon.vpadd.v2f32(<2 x float>, <2 x float>) nounwind readnone