From: Sanjay Patel <spatel@rotateright.com>
Date: Thu, 12 Aug 2021 16:35:39 +0000 (-0400)
Subject: [InstCombine] fold umax/umin intrinsics based on demanded bits
X-Git-Tag: upstream/15.0.7~34057
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=790c29ab86d3c2e46a228ac32e3d10e18afd4d65;p=platform%2Fupstream%2Fllvm.git

[InstCombine] fold umax/umin intrinsics based on demanded bits

This is a direct translation of the select folds added with
D53033 / D53036 and another step towards canonicalization
using the intrinsics (see D98152).
---

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
index 15b51ae..502bd15 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
@@ -829,6 +829,29 @@ Value *InstCombinerImpl::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
         KnownBitsComputed = true;
         break;
       }
+      case Intrinsic::umax: {
+        // UMax(A, C) == A if ...
+        // The lowest non-zero bit of DemandMask is higher than the highest
+        // non-zero bit of C.
+        const APInt *C;
+        unsigned CTZ = DemandedMask.countTrailingZeros();
+        if (match(II->getArgOperand(1), m_APInt(C)) &&
+            CTZ >= C->getActiveBits())
+          return II->getArgOperand(0);
+        break;
+      }
+      case Intrinsic::umin: {
+        // UMin(A, C) == A if ...
+        // The lowest non-zero bit of DemandMask is higher than the highest
+        // non-one bit of C.
+        // This comes from using DeMorgans on the above umax example.
+        const APInt *C;
+        unsigned CTZ = DemandedMask.countTrailingZeros();
+        if (match(II->getArgOperand(1), m_APInt(C)) &&
+            CTZ >= C->getBitWidth() - C->countLeadingOnes())
+          return II->getArgOperand(0);
+        break;
+      }
       default: {
         // Handle target specific intrinsics
         Optional<Value *> V = targetSimplifyDemandedUseBitsIntrinsic(
diff --git a/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll b/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
index b41b68a5..8693526 100644
--- a/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
+++ b/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
@@ -976,8 +976,7 @@ define i8 @smax_uses(i8 %x, i8 %y, i8 %z) {
 
 define i8 @umax_demand_lshr(i8 %x) {
 ; CHECK-LABEL: @umax_demand_lshr(
-; CHECK-NEXT:    [[M:%.*]] = call i8 @llvm.umax.i8(i8 [[X:%.*]], i8 15)
-; CHECK-NEXT:    [[R:%.*]] = lshr i8 [[M]], 4
+; CHECK-NEXT:    [[R:%.*]] = lshr i8 [[X:%.*]], 4
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %m = call i8 @llvm.umax.i8(i8 %x, i8 15)
@@ -987,8 +986,7 @@ define i8 @umax_demand_lshr(i8 %x) {
 
 define i8 @umax_demand_and(i8 %x) {
 ; CHECK-LABEL: @umax_demand_and(
-; CHECK-NEXT:    [[M:%.*]] = call i8 @llvm.umax.i8(i8 [[X:%.*]], i8 1)
-; CHECK-NEXT:    [[R:%.*]] = and i8 [[M]], 10
+; CHECK-NEXT:    [[R:%.*]] = and i8 [[X:%.*]], 10
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %m = call i8 @llvm.umax.i8(i8 1, i8 %x)
@@ -998,8 +996,7 @@ define i8 @umax_demand_and(i8 %x) {
 
 define i8 @umin_demand_or_31_30(i8 %x) {
 ; CHECK-LABEL: @umin_demand_or_31_30(
-; CHECK-NEXT:    [[M:%.*]] = call i8 @llvm.umin.i8(i8 [[X:%.*]], i8 -30)
-; CHECK-NEXT:    [[R:%.*]] = or i8 [[M]], 31
+; CHECK-NEXT:    [[R:%.*]] = or i8 [[X:%.*]], 31
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %m = call i8 @llvm.umin.i8(i8 -30, i8 %x)
@@ -1009,8 +1006,7 @@ define i8 @umin_demand_or_31_30(i8 %x) {
 
 define i8 @umin_demand_and_7_8(i8 %x) {
 ; CHECK-LABEL: @umin_demand_and_7_8(
-; CHECK-NEXT:    [[M:%.*]] = call i8 @llvm.umin.i8(i8 [[X:%.*]], i8 -7)
-; CHECK-NEXT:    [[R:%.*]] = and i8 [[M]], -8
+; CHECK-NEXT:    [[R:%.*]] = and i8 [[X:%.*]], -8
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %m = call i8 @llvm.umin.i8(i8 %x, i8 -7)