[PATCH] D77591: [SveEmitter] Explicitly merge with zero/undef

[Sander de Smalen via Phabricator via cfe-commits]

This revision was automatically updated to reflect the committed changes.
Closed by commit rG9986b3de26d3: [SveEmitter] Explicitly merge with zero/undef 
(authored by sdesmalen).

Changed prior to commit:
  https://reviews.llvm.org/D77591?vs=255501&id=258760#toc

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D77591/new/

https://reviews.llvm.org/D77591

Files:
  clang/include/clang/Basic/arm_sve.td
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_abs.c
  clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_neg.c

Index: clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_neg.c
===
--- /dev/null
+++ clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_neg.c
@@ -0,0 +1,197 @@
+// 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
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+
+#include 
+
+#ifdef SVE_OVERLOADED_FORMS
+// A simple used,unused... macro, long enough to represent any SVE builtin.
+#define SVE_ACLE_FUNC(A1,A2_UNUSED,A3,A4_UNUSED) A1##A3
+#else
+#define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
+#endif
+
+svint8_t test_svneg_s8_z(svbool_t pg, svint8_t op)
+{
+  // CHECK-LABEL: test_svneg_s8_z
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv16i8( zeroinitializer,  %pg,  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s8,_z,)(pg, op);
+}
+
+svint16_t test_svneg_s16_z(svbool_t pg, svint16_t op)
+{
+  // CHECK-LABEL: test_svneg_s16_z
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv8i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv8i16( zeroinitializer,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s16,_z,)(pg, op);
+}
+
+svint32_t test_svneg_s32_z(svbool_t pg, svint32_t op)
+{
+  // CHECK-LABEL: test_svneg_s32_z
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv4i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv4i32( zeroinitializer,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s32,_z,)(pg, op);
+}
+
+svint64_t test_svneg_s64_z(svbool_t pg, svint64_t op)
+{
+  // CHECK-LABEL: test_svneg_s64_z
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv2i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv2i64( zeroinitializer,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s64,_z,)(pg, op);
+}
+
+svint8_t test_svneg_s8_m(svint8_t inactive, svbool_t pg, svint8_t op)
+{
+  // CHECK-LABEL: test_svneg_s8_m
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv16i8( %inactive,  %pg,  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s8,_m,)(inactive, pg, op);
+}
+
+svint16_t test_svneg_s16_m(svint16_t inactive, svbool_t pg, svint16_t op)
+{
+  // CHECK-LABEL: test_svneg_s16_m
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv8i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv8i16( %inactive,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s16,_m,)(inactive, pg, op);
+}
+
+svint32_t test_svneg_s32_m(svint32_t inactive, svbool_t pg, svint32_t op)
+{
+  // CHECK-LABEL: test_svneg_s32_m
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv4i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv4i32( %inactive,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s32,_m,)(inactive, pg, op);
+}
+
+svint64_t test_svneg_s64_m(svint64_t inactive, svbool_t pg, svint64_t op)
+{
+  // CHECK-LABEL: test_svneg_s64_m
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv2i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv2i64( %inactive,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s64,_m,)(inactive, pg, op);
+}
+
+svint8_t test_svneg_s8_x(svbool_t pg, svint8_t op)
+{
+  // CHECK-LABEL: test_svneg_s8_x
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv16i8( undef,  %pg,  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s8,_x,)(pg, op);
+}
+
+svint16_t test_svneg_s16_x(svbool_t pg, svint16_t op)
+{
+  // CHECK-LABEL: test_svneg_s16_x
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv8i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv8i16( undef,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s16,_x,)(pg, op);
+}
+
+svint32_t test_svneg_s32_x(svbool_t pg, svint32_t op)
+{
+  // CHECK-L

[PATCH] D77591: [SveEmitter] Explicitly merge with zero/undef

[Eli Friedman via Phabricator via cfe-commits]

efriedma accepted this revision.
efriedma added a comment.
This revision is now accepted and ready to land.

LGTM


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D77591/new/

https://reviews.llvm.org/D77591



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D77591: [SveEmitter] Explicitly merge with zero/undef

[Sander de Smalen via Phabricator via cfe-commits]

sdesmalen created this revision.
sdesmalen added reviewers: SjoerdMeijer, efriedma, rovka.
Herald added a subscriber: tschuett.
Herald added a project: clang.
sdesmalen added a child revision: D77593: [SveEmitter] Implement zeroing of 
false lanes.

Builtins that have the merge type MergeAnyExp or MergeZeroExp,
merge into a 'undef' or 'zero' vector respectively, which enables the
_x and _z behaviour for unary operations.

  

This patch also adds builtins for svabs and svneg.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D77591

Files:
  clang/include/clang/Basic/arm_sve.td
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_abs.c
  clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_neg.c

Index: clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_neg.c
===
--- /dev/null
+++ clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_neg.c
@@ -0,0 +1,197 @@
+// 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
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+
+#include 
+
+#ifdef SVE_OVERLOADED_FORMS
+// A simple used,unused... macro, long enough to represent any SVE builtin.
+#define SVE_ACLE_FUNC(A1,A2_UNUSED,A3,A4_UNUSED) A1##A3
+#else
+#define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
+#endif
+
+svint8_t test_svneg_s8_z(svbool_t pg, svint8_t op)
+{
+  // CHECK-LABEL: test_svneg_s8_z
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv16i8( zeroinitializer,  %pg,  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s8,_z,)(pg, op);
+}
+
+svint16_t test_svneg_s16_z(svbool_t pg, svint16_t op)
+{
+  // CHECK-LABEL: test_svneg_s16_z
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv8i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv8i16( zeroinitializer,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s16,_z,)(pg, op);
+}
+
+svint32_t test_svneg_s32_z(svbool_t pg, svint32_t op)
+{
+  // CHECK-LABEL: test_svneg_s32_z
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv4i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv4i32( zeroinitializer,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s32,_z,)(pg, op);
+}
+
+svint64_t test_svneg_s64_z(svbool_t pg, svint64_t op)
+{
+  // CHECK-LABEL: test_svneg_s64_z
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv2i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv2i64( zeroinitializer,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s64,_z,)(pg, op);
+}
+
+svint8_t test_svneg_s8_m(svint8_t inactive, svbool_t pg, svint8_t op)
+{
+  // CHECK-LABEL: test_svneg_s8_m
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv16i8( %inactive,  %pg,  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s8,_m,)(inactive, pg, op);
+}
+
+svint16_t test_svneg_s16_m(svint16_t inactive, svbool_t pg, svint16_t op)
+{
+  // CHECK-LABEL: test_svneg_s16_m
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv8i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv8i16( %inactive,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s16,_m,)(inactive, pg, op);
+}
+
+svint32_t test_svneg_s32_m(svint32_t inactive, svbool_t pg, svint32_t op)
+{
+  // CHECK-LABEL: test_svneg_s32_m
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv4i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv4i32( %inactive,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s32,_m,)(inactive, pg, op);
+}
+
+svint64_t test_svneg_s64_m(svint64_t inactive, svbool_t pg, svint64_t op)
+{
+  // CHECK-LABEL: test_svneg_s64_m
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv2i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv2i64( %inactive,  %[[PG]],  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s64,_m,)(inactive, pg, op);
+}
+
+svint8_t test_svneg_s8_x(svbool_t pg, svint8_t op)
+{
+  // CHECK-LABEL: test_svneg_s8_x
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv16i8( undef,  %pg,  %op)
+  // CHECK: ret  %[[INTRINSIC]]
+  return SVE_ACLE_FUNC(svneg,_s8,_x,)(pg, op);
+}
+
+svint16_t test_svneg_s16_x(svbool_t pg, svint16_t op)
+{
+  // CHECK-LABEL: test_svneg_s16_x
+  // CHECK: %[[PG:.*]] = call  @llvm.aarch64.sve.convert.from.svbool.nxv8i1( %pg)
+  // CHECK: %[[INTRINSIC:.*]] = call  @llvm.aarch64.sve.neg.nxv8i16( undef,  %[[PG]],  %op)
+