Add a couple dag combines to transform mulhi/mullo into a wider multiply

when the wider type is legal.  This allows us to compile:

define zeroext i16 @test1(i16 zeroext %x) nounwind {
entry:
	%div = udiv i16 %x, 33
	ret i16 %div
}

into:

test1:                                  # @test1
	movzwl	4(%esp), %eax
	imull	$63551, %eax, %eax      # imm = 0xF83F
	shrl	$21, %eax
	ret

instead of:

test1:                                  # @test1
        movw    $-1985, %ax             # imm = 0xFFFFFFFFFFFFF83F
        mulw    4(%esp)
        andl    $65504, %edx            # imm = 0xFFE0
        movl    %edx, %eax
        shrl    $5, %eax
        ret

Implementing rdar://8760399 and example #4 from:
http://blog.regehr.org/archives/320

We should implement the same thing for [su]mul_hilo, but I don't
have immediate plans to do this.




git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@121696 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 72 insertions, 12 deletions

diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 0d3b0df..740b29b 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -1916,6 +1916,7 @@ SDValue DAGCombiner::visitMULHS(SDNode *N) {
   SDValue N1 = N->getOperand(1);
   ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
   EVT VT = N->getValueType(0);
+  DebugLoc DL = N->getDebugLoc();
 
   // fold (mulhs x, 0) -> 0
   if (N1C && N1C->isNullValue())
@@ -1929,6 +1930,22 @@ SDValue DAGCombiner::visitMULHS(SDNode *N) {
   if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
     return DAG.getConstant(0, VT);
 
+  // If the type twice as wide is legal, transform the mulhs to a wider multiply
+  // plus a shift.
+  if (VT.isSimple() && !VT.isVector()) {
+    MVT Simple = VT.getSimpleVT();
+    unsigned SimpleSize = Simple.getSizeInBits();
+    EVT NewVT = EVT::getIntegerVT(*DAG.getContext(), SimpleSize*2);
+    if (TLI.isOperationLegal(ISD::MUL, NewVT)) {
+      N0 = DAG.getNode(ISD::SIGN_EXTEND, DL, NewVT, N0);
+      N1 = DAG.getNode(ISD::SIGN_EXTEND, DL, NewVT, N1);
+      N1 = DAG.getNode(ISD::MUL, DL, NewVT, N0, N1);
+      N1 = DAG.getNode(ISD::SRA, DL, NewVT, N1,
+                       DAG.getConstant(SimpleSize, getShiftAmountTy()));
+      return DAG.getNode(ISD::TRUNCATE, DL, VT, N1);
+    }
+  }
+  
   return SDValue();
 }
 
@@ -1937,6 +1954,7 @@ SDValue DAGCombiner::visitMULHU(SDNode *N) {
   SDValue N1 = N->getOperand(1);
   ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
   EVT VT = N->getValueType(0);
+  DebugLoc DL = N->getDebugLoc();
 
   // fold (mulhu x, 0) -> 0
   if (N1C && N1C->isNullValue())
@@ -1948,6 +1966,22 @@ SDValue DAGCombiner::visitMULHU(SDNode *N) {
   if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
     return DAG.getConstant(0, VT);
 
+  // If the type twice as wide is legal, transform the mulhu to a wider multiply
+  // plus a shift.
+  if (VT.isSimple() && !VT.isVector()) {
+    MVT Simple = VT.getSimpleVT();
+    unsigned SimpleSize = Simple.getSizeInBits();
+    EVT NewVT = EVT::getIntegerVT(*DAG.getContext(), SimpleSize*2);
+    if (TLI.isOperationLegal(ISD::MUL, NewVT)) {
+      N0 = DAG.getNode(ISD::ZERO_EXTEND, DL, NewVT, N0);
+      N1 = DAG.getNode(ISD::ZERO_EXTEND, DL, NewVT, N1);
+      N1 = DAG.getNode(ISD::MUL, DL, NewVT, N0, N1);
+      N1 = DAG.getNode(ISD::SRL, DL, NewVT, N1,
+                       DAG.getConstant(SimpleSize, getShiftAmountTy()));
+      return DAG.getNode(ISD::TRUNCATE, DL, VT, N1);
+    }
+  }
+  
   return SDValue();
 }
 
@@ -2011,6 +2045,7 @@ SDValue DAGCombiner::visitSMUL_LOHI(SDNode *N) {
   SDValue Res = SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHS);
   if (Res.getNode()) return Res;
 
+  // TODO: Transform smul_lohi to mul if the wider mul is legal!
   return SDValue();
 }
 
@@ -2018,6 +2053,7 @@ SDValue DAGCombiner::visitUMUL_LOHI(SDNode *N) {
   SDValue Res = SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHU);
   if (Res.getNode()) return Res;
 
+  // TODO: Transform umul_lohi to mul if the wider mul is legal!
   return SDValue();
 }
 
diff --git a/test/CodeGen/X86/2008-11-29-DivideConstant16bit.ll b/test/CodeGen/X86/2008-11-29-DivideConstant16bit.ll
index 2e114ab..6b423a1 100644
--- a/test/CodeGen/X86/2008-11-29-DivideConstant16bit.ll
+++ b/test/CodeGen/X86/2008-11-29-DivideConstant16bit.ll
@@ -1,9 +1,42 @@
-; RUN: llc < %s -mtriple=i686-pc-linux-gnu | grep -- -1985 | count 1
+; RUN: llc < %s -mtriple=i686-pc-linux-gnu -asm-verbose=0 | FileCheck %s
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32"
 target triple = "i686-pc-linux-gnu"
 
-define zeroext i16 @a(i16 zeroext %x) nounwind {
+define zeroext i16 @test1(i16 zeroext %x) nounwind {
 entry:
-	%div = udiv i16 %x, 33		; <i32> [#uses=1]
+	%div = udiv i16 %x, 33
+	ret i16 %div
+; CHECK: test1:
+; CHECK: imull	$63551, %eax, %eax
+; CHECK-NEXT: shrl	$21, %eax
+; CHECK-NEXT: ret
+}
+
+define zeroext i16 @test2(i8 signext %x, i16 zeroext %c) nounwind readnone ssp noredzone {
+entry:
+  %div = udiv i16 %c, 3
+  ret i16 %div
+
+; CHECK: test2:
+; CHECK: imull	$43691, %eax, %eax
+; CHECK-NEXT: shrl	$17, %eax
+; CHECK-NEXT: ret
+}
+
+define zeroext i8 @test3(i8 zeroext %x, i8 zeroext %c) nounwind readnone ssp noredzone {
+entry:
+  %div = udiv i8 %c, 3
+  ret i8 %div
+
+; CHECK: test3:
+; CHECK: imull	$171, %eax, %eax
+; CHECK-NEXT: shrb	%ah
+; CHECK-NEXT: movzbl	%ah, %eax
+; CHECK-NEXT: ret
+}
+
+define signext i16 @test4(i16 signext %x) nounwind {
+entry:
+	%div = sdiv i16 %x, 33		; <i32> [#uses=1]
 	ret i16 %div
 }
diff --git a/test/CodeGen/X86/2008-11-29-DivideConstant16bitSigned.ll b/test/CodeGen/X86/2008-11-29-DivideConstant16bitSigned.ll
deleted file mode 100644
index 7c811af..0000000
--- a/test/CodeGen/X86/2008-11-29-DivideConstant16bitSigned.ll
+++ /dev/null
@@ -1,9 +0,0 @@
-; RUN: llc < %s -mtriple=i686-pc-linux-gnu | grep -- -1985
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32"
-target triple = "i686-pc-linux-gnu"
-
-define signext i16 @a(i16 signext %x) nounwind {
-entry:
-	%div = sdiv i16 %x, 33		; <i32> [#uses=1]
-	ret i16 %div
-}