sdesmalen created this revision.
sdesmalen added reviewers: SjoerdMeijer, efriedma, rovka.
Herald added a subscriber: tschuett.
Herald added a reviewer: rengolin.
Herald added a project: clang.
sdesmalen added a parent revision: D77596: [SveEmitter] Add NoOverload flag and
builtin for svpfalse.
Some ACLE builtins leave out the argument to specify the predicate
pattern, which is expected to be expanded to SV_ALL pattern.
This patch adds the flag ExpandOp1SVALL to expand the second operand
to an SVE_ALL value, or append a SV_ALL if the builtin only takes
a single operand.
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D77597
Files:
clang/include/clang/Basic/TargetBuiltins.h
clang/include/clang/Basic/arm_sve.td
clang/lib/CodeGen/CGBuiltin.cpp
clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ptrue.c
Index: clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ptrue.c
===================================================================
--- /dev/null
+++ clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ptrue.c
@@ -0,0 +1,201 @@
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+
+#include <arm_sve.h>
+
+svbool_t test_svptrue_b8()
+{
+ // CHECK-LABEL: test_svptrue_b8
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_b8();
+}
+
+svbool_t test_svptrue_b16()
+{
+ // CHECK-LABEL: test_svptrue_b16
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 31)
+ // CHECK: %[[CAST:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %[[INTRINSIC]])
+ // CHECK: ret <vscale x 16 x i1> %[[CAST]]
+ return svptrue_b16();
+}
+
+svbool_t test_svptrue_b32()
+{
+ // CHECK-LABEL: test_svptrue_b32
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 31)
+ // CHECK: %[[CAST:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %[[INTRINSIC]])
+ // CHECK: ret <vscale x 16 x i1> %[[CAST]]
+ return svptrue_b32();
+}
+
+svbool_t test_svptrue_b64()
+{
+ // CHECK-LABEL: test_svptrue_b64
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 31)
+ // CHECK: %[[CAST:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %[[INTRINSIC]])
+ // CHECK: ret <vscale x 16 x i1> %[[CAST]]
+ return svptrue_b64();
+}
+
+svbool_t test_svptrue_pat_b8()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 0)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_POW2);
+}
+
+svbool_t test_svptrue_pat_b8_1()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_1
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 1)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL1);
+}
+
+svbool_t test_svptrue_pat_b8_2()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_2
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 2)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL2);
+}
+
+svbool_t test_svptrue_pat_b8_3()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_3
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 3)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL3);
+}
+
+svbool_t test_svptrue_pat_b8_4()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_4
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 4)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL4);
+}
+
+svbool_t test_svptrue_pat_b8_5()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_5
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 5)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL5);
+}
+
+svbool_t test_svptrue_pat_b8_6()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_6
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 6)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL6);
+}
+
+svbool_t test_svptrue_pat_b8_7()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_7
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 7)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL7);
+}
+
+svbool_t test_svptrue_pat_b8_8()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_8
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 8)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL8);
+}
+
+svbool_t test_svptrue_pat_b8_9()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_9
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 9)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL16);
+}
+
+svbool_t test_svptrue_pat_b8_10()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_10
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 10)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL32);
+}
+
+svbool_t test_svptrue_pat_b8_11()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_11
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 11)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL64);
+}
+
+svbool_t test_svptrue_pat_b8_12()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_12
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 12)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL128);
+}
+
+svbool_t test_svptrue_pat_b8_13()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_13
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 13)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_VL256);
+}
+
+svbool_t test_svptrue_pat_b8_14()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_14
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 29)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_MUL4);
+}
+
+svbool_t test_svptrue_pat_b8_15()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_15
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 30)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_MUL3);
+}
+
+svbool_t test_svptrue_pat_b8_16()
+{
+ // CHECK-LABEL: test_svptrue_pat_b8_16
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
+ // CHECK: ret <vscale x 16 x i1> %[[INTRINSIC]]
+ return svptrue_pat_b8(SV_ALL);
+}
+
+svbool_t test_svptrue_pat_b16()
+{
+ // CHECK-LABEL: test_svptrue_pat_b16
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 0)
+ // CHECK: %[[CAST:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %[[INTRINSIC]])
+ // CHECK: ret <vscale x 16 x i1> %[[CAST]]
+ return svptrue_pat_b16(SV_POW2);
+}
+
+svbool_t test_svptrue_pat_b32()
+{
+ // CHECK-LABEL: test_svptrue_pat_b32
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 1)
+ // CHECK: %[[CAST:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %[[INTRINSIC]])
+ // CHECK: ret <vscale x 16 x i1> %[[CAST]]
+ return svptrue_pat_b32(SV_VL1);
+}
+
+svbool_t test_svptrue_pat_b64()
+{
+ // CHECK-LABEL: test_svptrue_pat_b64
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 2)
+ // CHECK: %[[CAST:.*]] = call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %[[INTRINSIC]])
+ // CHECK: ret <vscale x 16 x i1> %[[CAST]]
+ return svptrue_pat_b64(SV_VL2);
+}
Index: clang/lib/CodeGen/CGBuiltin.cpp
===================================================================
--- clang/lib/CodeGen/CGBuiltin.cpp
+++ clang/lib/CodeGen/CGBuiltin.cpp
@@ -7634,6 +7634,14 @@
if (TypeFlags.getMergeType() == SVETypeFlags::MergeAnyExp)
InsertExplicitUndefOperand(Builder, Ty, Ops);
+ // Insert a `SVAll (31)` for Op1
+ if (TypeFlags.isExpandOp1SVALL()) {
+ if (Ops.size() <= 1)
+ Ops.push_back(Builder.getInt32(31));
+ else
+ Ops.insert(&Ops[1], Builder.getInt32(31));
+ }
+
// Predicates must match the main datatype.
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
if (auto PredTy = dyn_cast<llvm::VectorType>(Ops[i]->getType()))
Index: clang/include/clang/Basic/arm_sve.td
===================================================================
--- clang/include/clang/Basic/arm_sve.td
+++ clang/include/clang/Basic/arm_sve.td
@@ -168,6 +168,7 @@
def IsZExtReturn : FlagType<0x00080000>; // Return value is sign-extend by default
def NoAuto : FlagType<0x00100000>;
def NoOverloadTy : FlagType<0x00200000>;
+def Op1SVALL : FlagType<0x00400000>;
// These must be kept in sync with the flags in include/clang/Basic/TargetBuiltins.h
class ImmCheckType<int val> {
@@ -408,6 +409,9 @@
def SVPFALSE : SInst<"svpfalse[_b]", "P", "", MergeNone, "", [NoOverloadTy]>;
+def SVPTRUE_PAT : SInst<"svptrue_pat_{d}", "PI", "PcPsPiPl", MergeNone, "aarch64_sve_ptrue">;
+def SVPTRUE : SInst<"svptrue_{d}", "P", "PcPsPiPl", MergeNone, "aarch64_sve_ptrue", [Op1SVALL]>;
+
////////////////////////////////////////////////////////////////////////////////
// Integer arithmetic
def SVDOT_LANE_S : SInst<"svdot_lane[_{d}]", "ddqqi", "il", MergeNone, "aarch64_sve_sdot_lane", [], [ImmCheck<3, ImmCheckLaneIndexDot, 2>]>;
Index: clang/include/clang/Basic/TargetBuiltins.h
===================================================================
--- clang/include/clang/Basic/TargetBuiltins.h
+++ clang/include/clang/Basic/TargetBuiltins.h
@@ -234,6 +234,7 @@
bool isZExtReturn() const { return Flags & IsZExtReturn; }
bool isNoAuto() const { return Flags & NoAuto; }
bool isNotOverloaded() const { return Flags & NoOverloadTy; }
+ bool isExpandOp1SVALL() const { return Flags & Op1SVALL; }
uint64_t getBits() const { return Flags; }
bool isFlagSet(uint64_t Flag) const { return Flags & Flag; }
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
