[SveEmitter] Add builtins for SVE2 Large integer arithmetic

This patch adds builtins for: - svadclb - svadclt - svsbclb - svsbclt
llvm · May 7, 2020 · 60615cf · 60615cf
1 parent 36aab0c
commit 60615cf
Show file tree

Hide file tree

Showing 5 changed files with 235 additions and 0 deletions.
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
@@ -1381,6 +1381,21 @@ def SVNBSL_N  : SInst<"svnbsl[_n_{d}]",  "ddda", "csilUcUsUiUl", MergeNone, "aar
 def SVXAR_N   : SInst<"svxar[_n_{d}]",   "dddi", "csilUcUsUiUl", MergeNone, "aarch64_sve_xar", [], [ImmCheck<2, ImmCheckShiftRight, 1>]>;
 }
 
+////////////////////////////////////////////////////////////////////////////////
+// SVE2 - Large integer arithmetic
+
+let ArchGuard = "defined(__ARM_FEATURE_SVE2)" in {
+def SVADCLB : SInst<"svadclb[_{d}]", "dddd", "UiUl", MergeNone, "aarch64_sve_adclb">;
+def SVADCLT : SInst<"svadclt[_{d}]", "dddd", "UiUl", MergeNone, "aarch64_sve_adclt">;
+def SVSBCLB : SInst<"svsbclb[_{d}]", "dddd", "UiUl", MergeNone, "aarch64_sve_sbclb">;
+def SVSBCLT : SInst<"svsbclt[_{d}]", "dddd", "UiUl", MergeNone, "aarch64_sve_sbclt">;
+
+def SVADCLB_N : SInst<"svadclb[_n_{d}]", "ddda", "UiUl", MergeNone, "aarch64_sve_adclb">;
+def SVADCLT_N : SInst<"svadclt[_n_{d}]", "ddda", "UiUl", MergeNone, "aarch64_sve_adclt">;
+def SVSBCLB_N : SInst<"svsbclb[_n_{d}]", "ddda", "UiUl", MergeNone, "aarch64_sve_sbclb">;
+def SVSBCLT_N : SInst<"svsbclt[_n_{d}]", "ddda", "UiUl", MergeNone, "aarch64_sve_sbclt">;
+}
+
 ////////////////////////////////////////////////////////////////////////////////
 // SVE2 - Non-temporal gather/scatter
 let ArchGuard = "defined(__ARM_FEATURE_SVE2)" in {

diff --git a/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_adclb.c b/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_adclb.c
@@ -0,0 +1,55 @@
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -D__ARM_FEATURE_SVE2 -triple aarch64-none-linux-gnu -target-feature +sve2 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -D__ARM_FEATURE_SVE2 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -fsyntax-only -verify -verify-ignore-unexpected=error %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -fsyntax-only -verify=overload -verify-ignore-unexpected=error %s
+
+#include <arm_sve.h>
+
+#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
+
+svuint32_t test_svadclb_u32(svuint32_t op1, svuint32_t op2, svuint32_t op3)
+{
+  // CHECK-LABEL: test_svadclb_u32
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.adclb.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 4 x i32> %op2, <vscale x 4 x i32> %op3)
+  // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svadclb'}}
+  // expected-warning@+1 {{implicit declaration of function 'svadclb_u32'}}
+  return SVE_ACLE_FUNC(svadclb,_u32,,)(op1, op2, op3);
+}
+
+svuint64_t test_svadclb_u64(svuint64_t op1, svuint64_t op2, svuint64_t op3)
+{
+  // CHECK-LABEL: test_svadclb_u64
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.adclb.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 2 x i64> %op2, <vscale x 2 x i64> %op3)
+  // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svadclb'}}
+  // expected-warning@+1 {{implicit declaration of function 'svadclb_u64'}}
+  return SVE_ACLE_FUNC(svadclb,_u64,,)(op1, op2, op3);
+}
+
+svuint32_t test_svadclb_n_u32(svuint32_t op1, svuint32_t op2, uint32_t op3)
+{
+  // CHECK-LABEL: test_svadclb_n_u32
+  // CHECK: %[[DUP:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.dup.x.nxv4i32(i32 %op3)
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.adclb.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 4 x i32> %op2, <vscale x 4 x i32> %[[DUP]])
+  // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svadclb'}}
+  // expected-warning@+1 {{implicit declaration of function 'svadclb_n_u32'}}
+  return SVE_ACLE_FUNC(svadclb,_n_u32,,)(op1, op2, op3);
+}
+
+svuint64_t test_svadclb_n_u64(svuint64_t op1, svuint64_t op2, uint64_t op3)
+{
+  // CHECK-LABEL: test_svadclb_n_u64
+  // CHECK: %[[DUP:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.dup.x.nxv2i64(i64 %op3)
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.adclb.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 2 x i64> %op2, <vscale x 2 x i64> %[[DUP]])
+  // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svadclb'}}
+  // expected-warning@+1 {{implicit declaration of function 'svadclb_n_u64'}}
+  return SVE_ACLE_FUNC(svadclb,_n_u64,,)(op1, op2, op3);
+}
diff --git a/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_adclt.c b/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_adclt.c
@@ -0,0 +1,55 @@
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -D__ARM_FEATURE_SVE2 -triple aarch64-none-linux-gnu -target-feature +sve2 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -D__ARM_FEATURE_SVE2 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -fsyntax-only -verify -verify-ignore-unexpected=error %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -fsyntax-only -verify=overload -verify-ignore-unexpected=error %s
+
+#include <arm_sve.h>
+
+#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
+
+svuint32_t test_svadclt_u32(svuint32_t op1, svuint32_t op2, svuint32_t op3)
+{
+  // CHECK-LABEL: test_svadclt_u32
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.adclt.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 4 x i32> %op2, <vscale x 4 x i32> %op3)
+  // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svadclt'}}
+  // expected-warning@+1 {{implicit declaration of function 'svadclt_u32'}}
+  return SVE_ACLE_FUNC(svadclt,_u32,,)(op1, op2, op3);
+}
+
+svuint64_t test_svadclt_u64(svuint64_t op1, svuint64_t op2, svuint64_t op3)
+{
+  // CHECK-LABEL: test_svadclt_u64
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.adclt.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 2 x i64> %op2, <vscale x 2 x i64> %op3)
+  // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svadclt'}}
+  // expected-warning@+1 {{implicit declaration of function 'svadclt_u64'}}
+  return SVE_ACLE_FUNC(svadclt,_u64,,)(op1, op2, op3);
+}
+
+svuint32_t test_svadclt_n_u32(svuint32_t op1, svuint32_t op2, uint32_t op3)
+{
+  // CHECK-LABEL: test_svadclt_n_u32
+  // CHECK: %[[DUP:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.dup.x.nxv4i32(i32 %op3)
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.adclt.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 4 x i32> %op2, <vscale x 4 x i32> %[[DUP]])
+  // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svadclt'}}
+  // expected-warning@+1 {{implicit declaration of function 'svadclt_n_u32'}}
+  return SVE_ACLE_FUNC(svadclt,_n_u32,,)(op1, op2, op3);
+}
+
+svuint64_t test_svadclt_n_u64(svuint64_t op1, svuint64_t op2, uint64_t op3)
+{
+  // CHECK-LABEL: test_svadclt_n_u64
+  // CHECK: %[[DUP:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.dup.x.nxv2i64(i64 %op3)
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.adclt.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 2 x i64> %op2, <vscale x 2 x i64> %[[DUP]])
+  // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svadclt'}}
+  // expected-warning@+1 {{implicit declaration of function 'svadclt_n_u64'}}
+  return SVE_ACLE_FUNC(svadclt,_n_u64,,)(op1, op2, op3);
+}
diff --git a/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sbclb.c b/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sbclb.c
@@ -0,0 +1,55 @@
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -D__ARM_FEATURE_SVE2 -triple aarch64-none-linux-gnu -target-feature +sve2 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -D__ARM_FEATURE_SVE2 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -fsyntax-only -verify -verify-ignore-unexpected=error %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -fsyntax-only -verify=overload -verify-ignore-unexpected=error %s
+
+#include <arm_sve.h>
+
+#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
+
+svuint32_t test_svsbclb_u32(svuint32_t op1, svuint32_t op2, svuint32_t op3)
+{
+  // CHECK-LABEL: test_svsbclb_u32
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.sbclb.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 4 x i32> %op2, <vscale x 4 x i32> %op3)
+  // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svsbclb'}}
+  // expected-warning@+1 {{implicit declaration of function 'svsbclb_u32'}}
+  return SVE_ACLE_FUNC(svsbclb,_u32,,)(op1, op2, op3);
+}
+
+svuint64_t test_svsbclb_u64(svuint64_t op1, svuint64_t op2, svuint64_t op3)
+{
+  // CHECK-LABEL: test_svsbclb_u64
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.sbclb.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 2 x i64> %op2, <vscale x 2 x i64> %op3)
+  // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svsbclb'}}
+  // expected-warning@+1 {{implicit declaration of function 'svsbclb_u64'}}
+  return SVE_ACLE_FUNC(svsbclb,_u64,,)(op1, op2, op3);
+}
+
+svuint32_t test_svsbclb_n_u32(svuint32_t op1, svuint32_t op2, uint32_t op3)
+{
+  // CHECK-LABEL: test_svsbclb_n_u32
+  // CHECK: %[[DUP:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.dup.x.nxv4i32(i32 %op3)
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.sbclb.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 4 x i32> %op2, <vscale x 4 x i32> %[[DUP]])
+  // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svsbclb'}}
+  // expected-warning@+1 {{implicit declaration of function 'svsbclb_n_u32'}}
+  return SVE_ACLE_FUNC(svsbclb,_n_u32,,)(op1, op2, op3);
+}
+
+svuint64_t test_svsbclb_n_u64(svuint64_t op1, svuint64_t op2, uint64_t op3)
+{
+  // CHECK-LABEL: test_svsbclb_n_u64
+  // CHECK: %[[DUP:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.dup.x.nxv2i64(i64 %op3)
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.sbclb.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 2 x i64> %op2, <vscale x 2 x i64> %[[DUP]])
+  // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svsbclb'}}
+  // expected-warning@+1 {{implicit declaration of function 'svsbclb_n_u64'}}
+  return SVE_ACLE_FUNC(svsbclb,_n_u64,,)(op1, op2, op3);
+}
diff --git a/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sbclt.c b/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sbclt.c
@@ -0,0 +1,55 @@
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -D__ARM_FEATURE_SVE2 -triple aarch64-none-linux-gnu -target-feature +sve2 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -D__ARM_FEATURE_SVE2 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -fsyntax-only -verify -verify-ignore-unexpected=error %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -fsyntax-only -verify=overload -verify-ignore-unexpected=error %s
+
+#include <arm_sve.h>
+
+#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
+
+svuint32_t test_svsbclt_u32(svuint32_t op1, svuint32_t op2, svuint32_t op3)
+{
+  // CHECK-LABEL: test_svsbclt_u32
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.sbclt.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 4 x i32> %op2, <vscale x 4 x i32> %op3)
+  // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svsbclt'}}
+  // expected-warning@+1 {{implicit declaration of function 'svsbclt_u32'}}
+  return SVE_ACLE_FUNC(svsbclt,_u32,,)(op1, op2, op3);
+}
+
+svuint64_t test_svsbclt_u64(svuint64_t op1, svuint64_t op2, svuint64_t op3)
+{
+  // CHECK-LABEL: test_svsbclt_u64
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.sbclt.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 2 x i64> %op2, <vscale x 2 x i64> %op3)
+  // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svsbclt'}}
+  // expected-warning@+1 {{implicit declaration of function 'svsbclt_u64'}}
+  return SVE_ACLE_FUNC(svsbclt,_u64,,)(op1, op2, op3);
+}
+
+svuint32_t test_svsbclt_n_u32(svuint32_t op1, svuint32_t op2, uint32_t op3)
+{
+  // CHECK-LABEL: test_svsbclt_n_u32
+  // CHECK: %[[DUP:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.dup.x.nxv4i32(i32 %op3)
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.sbclt.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 4 x i32> %op2, <vscale x 4 x i32> %[[DUP]])
+  // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svsbclt'}}
+  // expected-warning@+1 {{implicit declaration of function 'svsbclt_n_u32'}}
+  return SVE_ACLE_FUNC(svsbclt,_n_u32,,)(op1, op2, op3);
+}
+
+svuint64_t test_svsbclt_n_u64(svuint64_t op1, svuint64_t op2, uint64_t op3)
+{
+  // CHECK-LABEL: test_svsbclt_n_u64
+  // CHECK: %[[DUP:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.dup.x.nxv2i64(i64 %op3)
+  // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.sbclt.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 2 x i64> %op2, <vscale x 2 x i64> %[[DUP]])
+  // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+  // overload-warning@+2 {{implicit declaration of function 'svsbclt'}}
+  // expected-warning@+1 {{implicit declaration of function 'svsbclt_n_u64'}}
+  return SVE_ACLE_FUNC(svsbclt,_n_u64,,)(op1, op2, op3);
+}