[Flang] Turn on alias analysis for locally allocated objects #139682
Merged
+57
−53
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Previously, a bug in the MemCptOpt LLVM IR pass caused issues with adding alias tags for locally allocated objects for Fortran code. However, the bug has now been fixed ( llvm#129537 ), and we can safely enable alias tags for these objects. This change should improve the accuracy of the alias analysis.
llvmbot
added
flang
|
@llvm/pr-subscribers-flang-fir-hlfir Author: Dominik Adamski (DominikAdamski) ChangesPreviously, a bug in the MemCptOpt LLVM IR pass caused issues with adding alias tags for locally allocated objects for Fortran code. However, the bug has now been fixed ( #129537 ), and we can safely enable alias tags for these objects. This change should improve the accuracy of the alias analysis. Patch is 24.67 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/139682.diff 5 Files Affected:
diff --git a/flang/lib/Optimizer/Transforms/AddAliasTags.cpp b/flang/lib/Optimizer/Transforms/AddAliasTags.cpp
index 66b4b84998801..5cfbdc33285f9 100644
--- a/flang/lib/Optimizer/Transforms/AddAliasTags.cpp
+++ b/flang/lib/Optimizer/Transforms/AddAliasTags.cpp
@@ -43,13 +43,10 @@ static llvm::cl::opt<bool>
static llvm::cl::opt<bool>
enableDirect("direct-tbaa", llvm::cl::init(true), llvm::cl::Hidden,
llvm::cl::desc("Add TBAA tags to direct variables"));
-// This is **known unsafe** (misscompare in spec2017/wrf_r). It should
-// not be enabled by default.
-// The code is kept so that these may be tried with new benchmarks to see if
-// this is worth fixing in the future.
-static llvm::cl::opt<bool> enableLocalAllocs(
- "local-alloc-tbaa", llvm::cl::init(false), llvm::cl::Hidden,
- llvm::cl::desc("Add TBAA tags to local allocations. UNSAFE."));
+static llvm::cl::opt<bool>
+ enableLocalAllocs("local-alloc-tbaa", llvm::cl::init(true),
+ llvm::cl::Hidden,
+ llvm::cl::desc("Add TBAA tags to local allocations."));
namespace {
diff --git a/flang/test/Fir/tbaa-codegen2.fir b/flang/test/Fir/tbaa-codegen2.fir
index 8f8b6a29129e7..e4bfa9087ec75 100644
--- a/flang/test/Fir/tbaa-codegen2.fir
+++ b/flang/test/Fir/tbaa-codegen2.fir
@@ -100,7 +100,7 @@ module attributes {fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", llvm.targ
// [...]
// CHECK: %[[VAL50:.*]] = getelementptr i32, ptr %{{.*}}, i64 %{{.*}}
// store to the temporary:
-// CHECK: store i32 %{{.*}}, ptr %[[VAL50]], align 4, !tbaa ![[DATA_ACCESS_TAG:.*]]
+// CHECK: store i32 %{{.*}}, ptr %[[VAL50]], align 4, !tbaa ![[TMP_DATA_ACCESS_TAG:.*]]
// [...]
// CHECK: [[BOX_ACCESS_TAG]] = !{![[BOX_ACCESS_TYPE:.*]], ![[BOX_ACCESS_TYPE]], i64 0}
@@ -111,4 +111,7 @@ module attributes {fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", llvm.targ
// CHECK: ![[A_ACCESS_TYPE]] = !{!"dummy arg data/_QFfuncEa", ![[ARG_ACCESS_TYPE:.*]], i64 0}
// CHECK: ![[ARG_ACCESS_TYPE]] = !{!"dummy arg data", ![[DATA_ACCESS_TYPE:.*]], i64 0}
// CHECK: ![[DATA_ACCESS_TYPE]] = !{!"any data access", ![[ANY_ACCESS_TYPE]], i64 0}
-// CHECK: ![[DATA_ACCESS_TAG]] = !{![[DATA_ACCESS_TYPE]], ![[DATA_ACCESS_TYPE]], i64 0}
+// CHECK: ![[TMP_DATA_ACCESS_TAG]] = !{![[TMP_DATA_ACCESS_TYPE:.*]], ![[TMP_DATA_ACCESS_TYPE]], i64 0}
+// CHECK: ![[TMP_DATA_ACCESS_TYPE]] = !{!"allocated data/", ![[TMP_ACCESS_TYPE:.*]], i64 0}
+// CHECK: ![[TMP_ACCESS_TYPE]] = !{!"allocated data", ![[TARGET_ACCESS_TAG:.*]], i64 0}
+// CHECK: ![[TARGET_ACCESS_TAG]] = !{!"target data", ![[DATA_ACCESS_TYPE]], i64 0}
diff --git a/flang/test/Transforms/tbaa-with-dummy-scope2.fir b/flang/test/Transforms/tbaa-with-dummy-scope2.fir
index c8f419fbee652..249471de458d3 100644
--- a/flang/test/Transforms/tbaa-with-dummy-scope2.fir
+++ b/flang/test/Transforms/tbaa-with-dummy-scope2.fir
@@ -43,12 +43,15 @@ func.func @_QPtest1() attributes {noinline} {
// CHECK: #[[$ATTR_0:.+]] = #llvm.tbaa_root<id = "Flang function root _QPtest1">
// CHECK: #[[$ATTR_1:.+]] = #llvm.tbaa_type_desc<id = "any access", members = {<#[[$ATTR_0]], 0>}>
// CHECK: #[[$ATTR_2:.+]] = #llvm.tbaa_type_desc<id = "any data access", members = {<#[[$ATTR_1]], 0>}>
+// CHECK: #[[$TARGETDATA:.+]] = #llvm.tbaa_type_desc<id = "target data", members = {<#[[$ATTR_2]], 0>}>
// CHECK: #[[$ATTR_3:.+]] = #llvm.tbaa_type_desc<id = "dummy arg data", members = {<#[[$ATTR_2]], 0>}>
-// CHECK: #[[TARGETDATA:.+]] = #llvm.tbaa_type_desc<id = "target data", members = {<#[[$ATTR_2]], 0>}>
+// CHECK: #[[$LOCAL_ATTR_0:.+]] = #llvm.tbaa_type_desc<id = "allocated data", members = {<#[[$TARGETDATA]], 0>}>
// CHECK: #[[$ATTR_5:.+]] = #llvm.tbaa_type_desc<id = "dummy arg data/_QFtest1FinnerEx", members = {<#[[$ATTR_3]], 0>}>
-// CHECK: #[[$ATTR_4:.+]] = #llvm.tbaa_type_desc<id = "global data", members = {<#[[TARGETDATA]], 0>}>
+// CHECK: #[[$ATTR_4:.+]] = #llvm.tbaa_type_desc<id = "global data", members = {<#[[$TARGETDATA]], 0>}>
// CHECK: #[[$ATTR_7:.+]] = #llvm.tbaa_tag<base_type = #[[$ATTR_5]], access_type = #[[$ATTR_5]], offset = 0>
+// CHECK: #[[$LOCAL_ATTR_1:.+]] = #llvm.tbaa_type_desc<id = "allocated data/_QFtest1FinnerEy", members = {<#[[$LOCAL_ATTR_0]], 0>}>
// CHECK: #[[$ATTR_6:.+]] = #llvm.tbaa_type_desc<id = "global data/_QMmEglob", members = {<#[[$ATTR_4]], 0>}>
+// CHECK: #[[$LOCAL_ATTR_2:.+]] = #llvm.tbaa_tag<base_type = #[[$LOCAL_ATTR_1]], access_type = #[[$LOCAL_ATTR_1]], offset = 0>
// CHECK: #[[$ATTR_8:.+]] = #llvm.tbaa_tag<base_type = #[[$ATTR_6]], access_type = #[[$ATTR_6]], offset = 0>
// CHECK-LABEL: func.func @_QPtest1() attributes {noinline} {
// CHECK: %[[VAL_2:.*]] = fir.alloca i32 {bindc_name = "y", uniq_name = "_QFtest1FinnerEy"}
@@ -57,8 +60,8 @@ func.func @_QPtest1() attributes {noinline} {
// CHECK: %[[VAL_5:.*]] = fir.dummy_scope : !fir.dscope
// CHECK: %[[VAL_6:.*]] = fir.declare %[[VAL_4]] dummy_scope %[[VAL_5]] {uniq_name = "_QFtest1FinnerEx"} : (!fir.ref<i32>, !fir.dscope) -> !fir.ref<i32>
// CHECK: %[[VAL_7:.*]] = fir.declare %[[VAL_2]] {uniq_name = "_QFtest1FinnerEy"} : (!fir.ref<i32>) -> !fir.ref<i32>
-// CHECK: fir.store %{{.*}} to %[[VAL_7]] : !fir.ref<i32>
-// CHECK: %[[VAL_8:.*]] = fir.load %[[VAL_7]] : !fir.ref<i32>
+// CHECK: fir.store %{{.*}} to %[[VAL_7]] {tbaa = [#[[$LOCAL_ATTR_2]]]} : !fir.ref<i32>
+// CHECK: %[[VAL_8:.*]] = fir.load %[[VAL_7]] {tbaa = [#[[$LOCAL_ATTR_2]]]} : !fir.ref<i32>
// CHECK: fir.store %[[VAL_8]] to %[[VAL_6]] {tbaa = [#[[$ATTR_7]]]} : !fir.ref<i32>
// CHECK: fir.store %{{.*}} to %[[VAL_4]] {tbaa = [#[[$ATTR_8]]]} : !fir.ref<i32>
@@ -87,12 +90,16 @@ func.func @_QPtest2() attributes {noinline} {
// CHECK: #[[$ATTR_3:.+]] = #llvm.tbaa_type_desc<id = "any access", members = {<#[[$ATTR_1]], 0>}>
// CHECK: #[[$ATTR_4:.+]] = #llvm.tbaa_type_desc<id = "any data access", members = {<#[[$ATTR_2]], 0>}>
// CHECK: #[[$ATTR_5:.+]] = #llvm.tbaa_type_desc<id = "any data access", members = {<#[[$ATTR_3]], 0>}>
+// CHECK: #[[$TARGETDATA_0:.+]] = #llvm.tbaa_type_desc<id = "target data", members = {<#[[$ATTR_4]], 0>}>
// CHECK: #[[$ATTR_6:.+]] = #llvm.tbaa_type_desc<id = "dummy arg data", members = {<#[[$ATTR_4]], 0>}>
-// CHECK: #[[TARGETDATA:.+]] = #llvm.tbaa_type_desc<id = "target data", members = {<#[[$ATTR_5]], 0>}>
+// CHECK: #[[$TARGETDATA_1:.+]] = #llvm.tbaa_type_desc<id = "target data", members = {<#[[$ATTR_5]], 0>}>
+// CHECK: #[[$LOCAL_ATTR_0:.+]] = #llvm.tbaa_type_desc<id = "allocated data", members = {<#[[$TARGETDATA_0]], 0>}>
// CHECK: #[[$ATTR_8:.+]] = #llvm.tbaa_type_desc<id = "dummy arg data/_QFtest2FinnerEx", members = {<#[[$ATTR_6]], 0>}>
-// CHECK: #[[$ATTR_7:.+]] = #llvm.tbaa_type_desc<id = "global data", members = {<#[[TARGETDATA]], 0>}>
+// CHECK: #[[$ATTR_7:.+]] = #llvm.tbaa_type_desc<id = "global data", members = {<#[[$TARGETDATA_1]], 0>}>
// CHECK: #[[$ATTR_10:.+]] = #llvm.tbaa_tag<base_type = #[[$ATTR_8]], access_type = #[[$ATTR_8]], offset = 0>
+// CHECK: #[[$LOCAL_ATTR_1:.+]] = #llvm.tbaa_type_desc<id = "allocated data/_QFtest2FinnerEy", members = {<#[[$LOCAL_ATTR_0]], 0>}>
// CHECK: #[[$ATTR_9:.+]] = #llvm.tbaa_type_desc<id = "global data/_QMmEglob", members = {<#[[$ATTR_7]], 0>}>
+// CHECK: #[[$LOCAL_ATTR_2:.+]] = #llvm.tbaa_tag<base_type = #[[$LOCAL_ATTR_1]], access_type = #[[$LOCAL_ATTR_1]], offset = 0>
// CHECK: #[[$ATTR_11:.+]] = #llvm.tbaa_tag<base_type = #[[$ATTR_9]], access_type = #[[$ATTR_9]], offset = 0>
// CHECK-LABEL: func.func @_QPtest2() attributes {noinline} {
// CHECK: %[[VAL_2:.*]] = fir.alloca i32 {bindc_name = "y", uniq_name = "_QFtest2FinnerEy"}
@@ -102,7 +109,7 @@ func.func @_QPtest2() attributes {noinline} {
// CHECK: %[[VAL_6:.*]] = fir.dummy_scope : !fir.dscope
// CHECK: %[[VAL_7:.*]] = fir.declare %[[VAL_5]] dummy_scope %[[VAL_6]] {uniq_name = "_QFtest2FinnerEx"} : (!fir.ref<i32>, !fir.dscope) -> !fir.ref<i32>
// CHECK: %[[VAL_8:.*]] = fir.declare %[[VAL_2]] {uniq_name = "_QFtest2FinnerEy"} : (!fir.ref<i32>) -> !fir.ref<i32>
-// CHECK: fir.store %{{.*}} to %[[VAL_8]] : !fir.ref<i32>
-// CHECK: %[[VAL_9:.*]] = fir.load %[[VAL_8]] : !fir.ref<i32>
+// CHECK: fir.store %{{.*}} to %[[VAL_8]] {tbaa = [#[[$LOCAL_ATTR_2]]]} : !fir.ref<i32>
+// CHECK: %[[VAL_9:.*]] = fir.load %[[VAL_8]] {tbaa = [#[[$LOCAL_ATTR_2]]]} : !fir.ref<i32>
// CHECK: fir.store %[[VAL_9]] to %[[VAL_7]] {tbaa = [#[[$ATTR_10]]]} : !fir.ref<i32>
// CHECK: fir.store %{{.*}} to %[[VAL_5]] {tbaa = [#[[$ATTR_11]]]} : !fir.ref<i32>
diff --git a/flang/test/Transforms/tbaa2.fir b/flang/test/Transforms/tbaa2.fir
index 4678a1cd4a686..1429d0b420766 100644
--- a/flang/test/Transforms/tbaa2.fir
+++ b/flang/test/Transforms/tbaa2.fir
@@ -50,6 +50,7 @@
// CHECK: #[[TARGETDATA:.+]] = #llvm.tbaa_type_desc<id = "target data", members = {<#[[ANY_DATA]], 0>}>
// CHECK: #[[ANY_ARG:.+]] = #llvm.tbaa_type_desc<id = "dummy arg data", members = {<#[[ANY_DATA]], 0>}>
// CHECK: #[[ANY_GLBL:.+]] = #llvm.tbaa_type_desc<id = "global data", members = {<#[[TARGETDATA]], 0>}>
+// CHECK: #[[ANY_LOCAL:.+]] = #llvm.tbaa_type_desc<id = "allocated data", members = {<#[[TARGETDATA]], 0>}>
// CHECK: #[[ARG_LOW:.+]] = #llvm.tbaa_type_desc<id = "dummy arg data/_QMmodFcalleeElow", members = {<#[[ANY_ARG]], 0>}>
// CHECK: #[[ANY_DIRECT:.+]] = #llvm.tbaa_type_desc<id = "direct data", members = {<#[[TARGETDATA]], 0>}>
// CHECK: #[[ARG_Z:.+]] = #llvm.tbaa_type_desc<id = "dummy arg data/_QMmodFcalleeEz", members = {<#[[ANY_ARG]], 0>}>
@@ -61,21 +62,31 @@
// CHECK: #[[GLBL_ZSTART:.+]] = #llvm.tbaa_type_desc<id = "global data/_QMmodEzstart", members = {<#[[ANY_GLBL]], 0>}>
// CHECK: #[[GLBL_ZSTOP:.+]] = #llvm.tbaa_type_desc<id = "global data/_QMmodEzstop", members = {<#[[ANY_GLBL]], 0>}>
+// CHECK: #[[LOCAL1_ALLOC:.+]] = #llvm.tbaa_type_desc<id = "allocated data/_QMmodFcalleeEk", members = {<#[[ANY_LOCAL]], 0>}>
// CHECK: #[[GLBL_YSTART:.+]] = #llvm.tbaa_type_desc<id = "global data/_QMmodEystart", members = {<#[[ANY_GLBL]], 0>}>
// CHECK: #[[GLBL_YSTOP:.+]] = #llvm.tbaa_type_desc<id = "global data/_QMmodEystop", members = {<#[[ANY_GLBL]], 0>}>
+// CHECK: #[[LOCAL2_ALLOC:.+]] = #llvm.tbaa_type_desc<id = "allocated data/_QMmodFcalleeEj", members = {<#[[ANY_LOCAL]], 0>}>
// CHECK: #[[GLBL_XSTART:.+]] = #llvm.tbaa_type_desc<id = "global data/_QMmodExstart", members = {<#[[ANY_GLBL]], 0>}>
+// CHECK: #[[LOCAL3_ALLOC:.+]] = #llvm.tbaa_type_desc<id = "allocated data/_QMmodFcalleeEi", members = {<#[[ANY_LOCAL]], 0>}>
+// CHECK: #[[LOCAL4_ALLOC:.+]] = #llvm.tbaa_type_desc<id = "allocated data/_QMmodFcalleeEdxold", members = {<#[[ANY_LOCAL]], 0>}>
// CHECK: #[[DIRECT_A:.+]] = #llvm.tbaa_type_desc<id = "direct data/_QMmodEa", members = {<#[[ANY_DIRECT]], 0>}>
// CHECK: #[[DIRECT_B:.+]] = #llvm.tbaa_type_desc<id = "direct data/_QMmodEb", members = {<#[[ANY_DIRECT]], 0>}>
// CHECK: #[[GLBL_DYINV:.+]] = #llvm.tbaa_type_desc<id = "global data/_QMmodEdyinv", members = {<#[[ANY_GLBL]], 0>}>
+// CHECK: #[[LOCAL5_ALLOC:.+]] = #llvm.tbaa_type_desc<id = "allocated data/_QMmodFcalleeEdzinv", members = {<#[[ANY_LOCAL]], 0>}>
// CHECK: #[[GLBL_ZSTART_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[GLBL_ZSTART]], access_type = #[[GLBL_ZSTART]], offset = 0>
// CHECK: #[[GLBL_ZSTOP_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[GLBL_ZSTOP]], access_type = #[[GLBL_ZSTOP]], offset = 0>
+// CHECK: #[[LOCAL1_ALLOC_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[LOCAL1_ALLOC]], access_type = #[[LOCAL1_ALLOC]], offset = 0>
// CHECK: #[[GLBL_YSTART_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[GLBL_YSTART]], access_type = #[[GLBL_YSTART]], offset = 0>
// CHECK: #[[GLBL_YSTOP_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[GLBL_YSTOP]], access_type = #[[GLBL_YSTOP]], offset = 0>
+// CHECK: #[[LOCAL2_ALLOC_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[LOCAL2_ALLOC]], access_type = #[[LOCAL2_ALLOC]], offset = 0>
// CHECK: #[[GLBL_XSTART_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[GLBL_XSTART]], access_type = #[[GLBL_XSTART]], offset = 0>
+// CHECK: #[[LOCAL3_ALLOC_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[LOCAL3_ALLOC]], access_type = #[[LOCAL3_ALLOC]], offset = 0>
+// CHECK: #[[LOCAL4_ALLOC_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[LOCAL4_ALLOC]], access_type = #[[LOCAL4_ALLOC]], offset = 0>
// CHECK: #[[DIRECT_A_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[DIRECT_A]], access_type = #[[DIRECT_A]], offset = 0>
// CHECK: #[[DIRECT_B_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[DIRECT_B]], access_type = #[[DIRECT_B]], offset = 0>
// CHECK: #[[GLBL_DYINV_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[GLBL_DYINV]], access_type = #[[GLBL_DYINV]], offset = 0>
+// CHECK: #[[LOCAL5_ALLOC_TAG:.+]] = #llvm.tbaa_tag<base_type = #[[LOCAL5_ALLOC]], access_type = #[[LOCAL5_ALLOC]], offset = 0>
func.func @_QMmodPcallee(%arg0: !fir.box<!fir.array<?x?x?xf32>> {fir.bindc_name = "z"}, %arg1: !fir.box<!fir.array<?x?x?xf32>> {fir.bindc_name = "y"}, %arg2: !fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?xf32>>>> {fir.bindc_name = "low"}) {
%c2 = arith.constant 2 : index
@@ -277,7 +288,7 @@
// CHECK: %[[VAL_44:.*]] = fir.convert %[[VAL_43]] : (i32) -> index
// CHECK: %[[VAL_45:.*]] = fir.convert %[[VAL_42]] : (index) -> i32
// CHECK: %[[VAL_46:.*]]:2 = fir.do_loop %[[VAL_47:.*]] = %[[VAL_42]] to %[[VAL_44]] step %[[VAL_5]] iter_args(%[[VAL_48:.*]] = %[[VAL_45]]) -> (index, i32) {
-// CHECK: fir.store %[[VAL_48]] to %[[VAL_34]] : !fir.ref<i32>
+// CHECK: fir.store %[[VAL_48]] to %[[VAL_34]] {tbaa = [#[[LOCAL1_ALLOC_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_49:.*]] = fir.load %[[VAL_18]] {tbaa = [#[[GLBL_YSTART_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_50:.*]] = arith.addi %[[VAL_49]], %[[VAL_6]] : i32
// CHECK: %[[VAL_51:.*]] = fir.convert %[[VAL_50]] : (i32) -> index
@@ -285,24 +296,20 @@
// CHECK: %[[VAL_53:.*]] = fir.convert %[[VAL_52]] : (i32) -> index
// CHECK: %[[VAL_54:.*]] = fir.convert %[[VAL_51]] : (index) -> i32
// CHECK: %[[VAL_55:.*]]:2 = fir.do_loop %[[VAL_56:.*]] = %[[VAL_51]] to %[[VAL_53]] step %[[VAL_5]] iter_args(%[[VAL_57:.*]] = %[[VAL_54]]) -> (index, i32) {
-// CHECK: fir.store %[[VAL_57]] to %[[VAL_32]] : !fir.ref<i32>
+// CHECK: fir.store %[[VAL_57]] to %[[VAL_32]] {tbaa = [#[[LOCAL2_ALLOC_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_58:.*]] = fir.load %[[VAL_16]] {tbaa = [#[[GLBL_XSTART_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_59:.*]] = arith.addi %[[VAL_58]], %[[VAL_6]] : i32
// CHECK: %[[VAL_60:.*]] = fir.convert %[[VAL_59]] : (i32) -> index
// CHECK: %[[VAL_61:.*]] = fir.convert %[[VAL_60]] : (index) -> i32
// CHECK: %[[VAL_62:.*]]:2 = fir.do_loop %[[VAL_63:.*]] = %[[VAL_60]] to %[[VAL_4]] step %[[VAL_5]] iter_args(%[[VAL_64:.*]] = %[[VAL_61]]) -> (index, i32) {
-// TODO: local allocation assumed to always alias
-// CHECK: fir.store %[[VAL_64]] to %[[VAL_30]] : !fir.ref<i32>
+// CHECK: fir.store %[[VAL_64]] to %[[VAL_30]] {tbaa = [#[[LOCAL3_ALLOC_TAG]]]} : !fir.ref<i32>
// load from box tagged in CodeGen
// CHECK: %[[VAL_65:.*]] = fir.load %[[VAL_35]] : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?xf32>>>>
-// TODO: local allocation assumed to always alias
-// CHECK: %[[VAL_66:.*]] = fir.load %[[VAL_30]] : !fir.ref<i32>
+// CHECK: %[[VAL_66:.*]] = fir.load %[[VAL_30]] {tbaa = [#[[LOCAL3_ALLOC_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_67:.*]] = fir.convert %[[VAL_66]] : (i32) -> i64
-// TODO: local allocation assumed to always alias
-// CHECK: %[[VAL_68:.*]] = fir.load %[[VAL_32]] : !fir.ref<i32>
+// CHECK: %[[VAL_68:.*]] = fir.load %[[VAL_32]] {tbaa = [#[[LOCAL2_ALLOC_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_69:.*]] = fir.convert %[[VAL_68]] : (i32) -> i64
-// TODO: local allocation assumed to always alias
-// CHECK: %[[VAL_70:.*]] = fir.load %[[VAL_34]] : !fir.ref<i32>
+// CHECK: %[[VAL_70:.*]] = fir.load %[[VAL_34]] {tbaa = [#[[LOCAL1_ALLOC_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_71:.*]] = fir.convert %[[VAL_70]] : (i32) -> i64
// CHECK: %[[VAL_72:.*]] = fir.box_addr %[[VAL_65]] : (!fir.box<!fir.heap<!fir.array<?x?x?xf32>>>) -> !fir.heap<!fir.array<?x?x?xf32>>
// CHECK: %[[VAL_73:.*]]:3 = fir.box_dims %[[VAL_65]], %[[VAL_4]] : (!fir.box<!fir.heap<!fir.array<?x?x?xf32>>>, index) -> (index, index, index)
@@ -311,11 +318,10 @@
// CHECK: %[[VAL_76:.*]] = fir.shape_shift %[[VAL_73]]#0, %[[VAL_73]]#1, %[[VAL_74]]#0, %[[VAL_74]]#1, %[[VAL_75]]#0, %[[VAL_75]]#1 : (index, index, index, index, index, index) -> !fir.shapeshift<3>
// CHECK: %[[VAL_77:.*]] = fir.array_coor %[[VAL_72]](%[[VAL_76]]) %[[VAL_67]], %[[VAL_69]], %[[VAL_71]] : (!fir.heap<!fir.array<?x?x?xf32>>, !fir.shapeshift<3>, i64, i64, i64) -> !fir.ref<f32>
// CHECK: %[[VAL_78:.*]] = fir.load %[[VAL_77]] {tbaa = [#[[ARG_LOW_TAG]]]} : !fir.ref<f32>
-// CHECK: fir.store %[[VAL_78]] to %[[VAL_26]] : !fir.ref<f32>
+// CHECK: fir.store %[[VAL_78]] to %[[VAL_26]] {tbaa = [#[[LOCAL4_ALLOC_TAG]]]} : !fir.ref<f32>
// load from box tagged in CodeGen
// CHECK: %[[VAL_79:.*]] = fir.load %[[VAL_8]] : !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>
-// TODO: local allocation assumed to always alias
-// CHECK: %[[VAL_80:.*]] = fir.load %[[VAL_32]] : !fir.ref<i32>
+// CHECK: %[[VAL_80:.*]] = fir.load %[[VAL_32]] {tbaa = [#[[LOCAL2_ALLOC_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_81:.*]] = fir.convert %[[VAL_80]] : (i32) -> i64
// CHECK: %[[VAL_82:.*]] = fir.box_addr %[[VAL_79]] : (!fir.box<!fir.heap<!fir.array<?xf32>>>) -> !fir.heap<!fir.array<?xf32>>
// CHECK: %[[VAL_83:.*]]:3 = fir.box_dims %[[VAL_79]], %[[VAL_4]] : (!fir.box<!fir.heap<!fir.array<?xf32>>>, index) -> (index, index, index)
@@ -324,11 +330,9 @@
// CHECK: %[[VAL_86:.*]] = fir.load %[[VAL_85]] {tbaa = [#[[DIRECT_A_TAG]]]} : !fir.ref<f32>
// load from box
// CHECK: %[[VAL_87:.*]] = fir.load %[[VAL_35]] : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?xf32>>>>
-// load from local allocation
-// CHECK: %[[VAL_88:.*]] = fir.load %[[VAL_30]] : !fir.ref<i32>
+// CHECK: %[[VAL_88:.*]] = fir.load %[[VAL_30]] {tbaa = [#[[LOCAL3_ALLOC_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_89:.*]] = fir.convert %[[VAL_88]] : (i32) -> i64
-// load from local allocation
-// CHECK: %[[VAL_90:.*]] = fir.load %[[VAL_34]] : !fir.ref<i32>
+// CHECK: %[[VAL_90:.*]] = fir.load %[[VAL_34]] {tbaa = [#[[LOCAL1_ALLOC_TAG]]]} : !fir.ref<i32>
// CHECK: %[[VAL_91:.*]] = fir.convert %[[VAL_90]] : (i32) -> i64
// CHECK: %[[VAL_92:.*]] = fir.box_addr %[[VAL_87]] : (!fir.box<!fir.heap<!fir.array<?x?x?xf32>>>) -> !fir.heap<!fir.array<?x?x?xf32>>
// CHECK: %[[VAL_93:.*]]:3 = fir.box_dims %[[VAL_87]], %[[VAL_4]] : (!fir.box<!fir.heap<!fir.array<?x?x?xf32>>>, index) -> (index, index, index)
@@ -363,8 +367,7 @@
// CHECK: %[[VAL_121:.*]] = fir.load %[[VAL_120]] {tbaa = [#[[ARG_Y_TAG]]]} : !fir.ref<f32>
// CHECK: %[[VAL_122:.*]] = arith.subf %[[VAL_119]], %[[VAL_121]] fastmath<contract> : f32
// CHECK: %[[VAL_123:.*]] = fir.no_reassoc %[[VAL_122]] : f32
-// load from local allocation
-// CHECK: %[[VAL_124:.*]] = fir.load %[[VAL_28]] : !fir.ref<f32>
+// CHECK: %[[VAL_124:.*]] = fir.load %[[VAL_28]] {tbaa = [#[[LOCAL5_ALLOC_TAG]]]} : !fir.ref<f32>
// CHECK: %[[VAL_125:.*]] = arith.mulf %[[VAL_123]], %[[VAL_124]] fastmath<contract> : f32
// CHECK: %[[VAL_126:.*]] = arith.addf %[[VAL_115]], %[[VAL_125]] fastmath<contract> : f32
// CHECK: %[[VAL_127:.*]] = fir.no_reassoc %[[VAL_126]] : f32
@@ -373,30 +376,24 @@
// CHECK: fir.store %[[VAL_129]] to %[[VAL_97]] {tbaa = [#[[ARG_LOW_TAG]]]} : !fir.ref...
[truncated]
|
|
FWIW I tried it out and this improves the performance of the atmosphere kernel from https://github.com/E3SM-Project/codesign-kernels by well over 30%. |
tblah
approved these changes
May 13, 2025
vzakhari
approved these changes
May 13, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Previously, a bug in the MemCptOpt LLVM IR pass caused issues with adding alias tags for locally allocated objects for Fortran code.
However, the bug has now been fixed ( #129537 ), and we can safely enable alias tags for these objects. This change should improve the accuracy of the alias analysis.