[llvm] d2b7c09 - [Hexagon] Simplify intrinsic (vandvrt (vandqrt q b) m) -> q if possible
Krzysztof Parzyszek via llvm-commits
llvm-commits at lists.llvm.org
Fri Feb 21 11:56:31 PST 2020
Author: Krzysztof Parzyszek
Date: 2020-02-21T13:56:04-06:00
New Revision: d2b7c09e79a12cb61fc424429b348b2c04364d07
URL: https://github.com/llvm/llvm-project/commit/d2b7c09e79a12cb61fc424429b348b2c04364d07
DIFF: https://github.com/llvm/llvm-project/commit/d2b7c09e79a12cb61fc424429b348b2c04364d07.diff
LOG: [Hexagon] Simplify intrinsic (vandvrt (vandqrt q b) m) -> q if possible
When each byte in b&m is non-zero, this conversion Q->V->Q is a no-op.
Added:
llvm/test/Transforms/InstCombine/Hexagon/lit.local.cfg
llvm/test/Transforms/InstCombine/Hexagon/simplify-hvx-qvq.ll
Modified:
llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index 849ac5237c3a..b542b363f6f2 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -44,6 +44,7 @@
#include "llvm/IR/IntrinsicsX86.h"
#include "llvm/IR/IntrinsicsARM.h"
#include "llvm/IR/IntrinsicsAArch64.h"
+#include "llvm/IR/IntrinsicsHexagon.h"
#include "llvm/IR/IntrinsicsNVPTX.h"
#include "llvm/IR/IntrinsicsAMDGPU.h"
#include "llvm/IR/IntrinsicsPowerPC.h"
@@ -3987,6 +3988,24 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
break;
}
+ case Intrinsic::hexagon_V6_vandvrt:
+ case Intrinsic::hexagon_V6_vandvrt_128B: {
+ // Simplify Q -> V -> Q conversion.
+ if (auto Op0 = dyn_cast<IntrinsicInst>(II->getArgOperand(0))) {
+ Intrinsic::ID ID0 = Op0->getIntrinsicID();
+ if (ID0 != Intrinsic::hexagon_V6_vandqrt &&
+ ID0 != Intrinsic::hexagon_V6_vandqrt_128B)
+ break;
+ Value *Bytes = Op0->getArgOperand(1), *Mask = II->getArgOperand(1);
+ uint64_t Bytes1 = computeKnownBits(Bytes, 0, Op0).One.getZExtValue();
+ uint64_t Mask1 = computeKnownBits(Mask, 0, II).One.getZExtValue();
+ // Check if every byte has common bits in Bytes and Mask.
+ uint64_t C = Bytes1 & Mask1;
+ if ((C & 0xFF) && (C & 0xFF00) && (C & 0xFF0000) && (C & 0xFF000000))
+ return replaceInstUsesWith(*II, Op0->getArgOperand(0));
+ }
+ break;
+ }
case Intrinsic::stackrestore: {
// If the save is right next to the restore, remove the restore. This can
// happen when variable allocas are DCE'd.
diff --git a/llvm/test/Transforms/InstCombine/Hexagon/lit.local.cfg b/llvm/test/Transforms/InstCombine/Hexagon/lit.local.cfg
new file mode 100644
index 000000000000..cc6a7edf05f3
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/Hexagon/lit.local.cfg
@@ -0,0 +1,2 @@
+if not 'Hexagon' in config.root.targets:
+ config.unsupported = True
diff --git a/llvm/test/Transforms/InstCombine/Hexagon/simplify-hvx-qvq.ll b/llvm/test/Transforms/InstCombine/Hexagon/simplify-hvx-qvq.ll
new file mode 100644
index 000000000000..9501f4d0b191
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/Hexagon/simplify-hvx-qvq.ll
@@ -0,0 +1,82 @@
+; RUN: opt -instcombine -S < %s | FileCheck %s
+
+; Simplify the Q -> V -> Q sequence, i.e. (vandvrt (vandqrt q b) m) -> q
+; when every byte in (b & m) is non-zero.
+
+target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"
+target triple = "hexagon"
+
+
+; CHECK-LABEL: define {{.*}} @f0(<32 x i32> %a0, <32 x i32> %a1, <32 x i32> %a2)
+; CHECK: %[[V0:v[0-9]+]] = call <128 x i1> @llvm.hexagon.V6.vgtw.128B(<32 x i32> %a0, <32 x i32> %a1)
+; CHECK: %[[V1:v[0-9]+]] = call <128 x i1> @llvm.hexagon.V6.veqb.or.128B(<128 x i1> %[[V0]], <32 x i32> %a1, <32 x i32> %a2)
+; CHECK: %[[V2:v[0-9]+]] = call <128 x i1> @llvm.hexagon.V6.pred.and.128B(<128 x i1> %[[V0]], <128 x i1> %[[V1]])
+; CHECK: %[[V3:v[0-9]+]] = call <128 x i1> @llvm.hexagon.V6.pred.xor.128B(<128 x i1> %[[V1]], <128 x i1> %[[V2]])
+; CHECK: %[[V4:v[0-9]+]] = call <32 x i32> @llvm.hexagon.V6.vmux.128B(<128 x i1> %[[V3]], <32 x i32> %a1, <32 x i32> %a2)
+; CHECK: ret <32 x i32> %[[V4]]
+define inreg <32 x i32> @f0(<32 x i32> %a0, <32 x i32> %a1, <32 x i32> %a2) #0 {
+b0:
+ %v0 = call <128 x i1> @llvm.hexagon.V6.vgtw.128B(<32 x i32> %a0, <32 x i32> %a1)
+ %v1 = call <32 x i32> @llvm.hexagon.V6.vandqrt.128B(<128 x i1> %v0, i32 -1)
+ %v2 = call <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32> %v1, i32 -1)
+ %v3 = call <128 x i1> @llvm.hexagon.V6.veqb.or.128B(<128 x i1> %v2, <32 x i32> %a1, <32 x i32> %a2)
+ %v4 = call <32 x i32> @llvm.hexagon.V6.vandqrt.128B(<128 x i1> %v3, i32 -1)
+ %v5 = call <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32> %v1, i32 -1)
+ %v6 = call <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32> %v4, i32 -1)
+ %v7 = call <128 x i1> @llvm.hexagon.V6.pred.and.128B(<128 x i1> %v5, <128 x i1> %v6)
+ %v8 = call <32 x i32> @llvm.hexagon.V6.vandqrt.128B(<128 x i1> %v7, i32 -1)
+ %v9 = call <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32> %v4, i32 -1)
+ %v10 = call <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32> %v8, i32 -1)
+ %v11 = call <128 x i1> @llvm.hexagon.V6.pred.xor.128B(<128 x i1> %v9, <128 x i1> %v10)
+ %v12 = call <32 x i32> @llvm.hexagon.V6.vandqrt.128B(<128 x i1> %v11, i32 -1)
+ %v13 = call <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32> %v12, i32 -1)
+ %v14 = call <32 x i32> @llvm.hexagon.V6.vmux.128B(<128 x i1> %v13, <32 x i32> %a1, <32 x i32> %a2)
+ ret <32 x i32> %v14
+}
+
+; Bytes = 0x08040201, Mask = 0x0C060309, Common = 0x08040201: all bytes in
+; the common bits are non-zero, expect simplification.
+; CHECK-LABEL: define {{.*}} @f1(<32 x i32> %a0, <32 x i32> %a1, <32 x i32> %a2)
+; CHECK: %[[V0:v[0-9]+]] = call <128 x i1> @llvm.hexagon.V6.vgtw.128B(<32 x i32> %a0, <32 x i32> %a1)
+; CHECK: %[[V1:v[0-9]+]] = call <32 x i32> @llvm.hexagon.V6.vmux.128B(<128 x i1> %[[V0]], <32 x i32> %a1, <32 x i32> %a2)
+; CHECK: ret <32 x i32> %[[V1]]
+define inreg <32 x i32> @f1(<32 x i32> %a0, <32 x i32> %a1, <32 x i32> %a2) #0 {
+b0:
+ %v0 = call <128 x i1> @llvm.hexagon.V6.vgtw.128B(<32 x i32> %a0, <32 x i32> %a1)
+ %v1 = call <32 x i32> @llvm.hexagon.V6.vandqrt.128B(<128 x i1> %v0, i32 134480385)
+ %v2 = call <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32> %v1, i32 201720585)
+ %v3 = call <32 x i32> @llvm.hexagon.V6.vmux.128B(<128 x i1> %v2, <32 x i32> %a1, <32 x i32> %a2)
+ ret <32 x i32> %v3
+}
+
+; Bytes = 0x08040201, Mask = 0x0C060309, Common = 0x08040200: there is a
+; zero byte in the common bits, so vandqrt->vandvrt cannot be simplified.
+; CHECK-LABEL: define {{.*}} @f2(<32 x i32> %a0, <32 x i32> %a1, <32 x i32> %a2)
+; CHECK: %[[V0:v[0-9]+]] = call <128 x i1> @llvm.hexagon.V6.vgtw.128B(<32 x i32> %a0, <32 x i32> %a1)
+; CHECK: %[[V1:v[0-9]+]] = call <32 x i32> @llvm.hexagon.V6.vandqrt.128B(<128 x i1> %[[V0]], i32 134480385)
+; CHECK: %[[V2:v[0-9]+]] = call <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32> %[[V1]], i32 201720584)
+; CHECK: %[[V3:v[0-9]+]] = call <32 x i32> @llvm.hexagon.V6.vmux.128B(<128 x i1> %[[V2]], <32 x i32> %a1, <32 x i32> %a2)
+; CHECK: ret <32 x i32> %[[V3]]
+define inreg <32 x i32> @f2(<32 x i32> %a0, <32 x i32> %a1, <32 x i32> %a2) #0 {
+b0:
+ %v0 = call <128 x i1> @llvm.hexagon.V6.vgtw.128B(<32 x i32> %a0, <32 x i32> %a1)
+ %v1 = call <32 x i32> @llvm.hexagon.V6.vandqrt.128B(<128 x i1> %v0, i32 134480385)
+ %v2 = call <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32> %v1, i32 201720584)
+ %v3 = call <32 x i32> @llvm.hexagon.V6.vmux.128B(<128 x i1> %v2, <32 x i32> %a1, <32 x i32> %a2)
+ ret <32 x i32> %v3
+}
+
+declare <128 x i1> @llvm.hexagon.V6.vgtw.128B(<32 x i32>, <32 x i32>) #1
+declare <32 x i32> @llvm.hexagon.V6.vandqrt.128B(<128 x i1>, i32) #1
+declare <128 x i1> @llvm.hexagon.V6.veqb.or.128B(<128 x i1>, <32 x i32>, <32 x i32>) #1
+declare <128 x i1> @llvm.hexagon.V6.vandvrt.128B(<32 x i32>, i32) #1
+declare <128 x i1> @llvm.hexagon.V6.pred.and.128B(<128 x i1>, <128 x i1>) #1
+declare <128 x i1> @llvm.hexagon.V6.pred.xor.128B(<128 x i1>, <128 x i1>) #1
+declare <32 x i32> @llvm.hexagon.V6.vmux.128B(<128 x i1>, <32 x i32>, <32 x i32>) #1
+
+attributes #0 = { noinline nounwind "target-cpu"="hexagonv65" "target-features"="+hvx-length128b,+hvxv65,+v65,-long-calls" }
+attributes #1 = { nounwind readnone }
+
+!llvm.module.flags = !{!0}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
More information about the llvm-commits
mailing list