[clang] e76256e - [SveEmitter] Add builtins for SVE2 Complex integer dot product
Sander de Smalen via cfe-commits
cfe-commits at lists.llvm.org
Thu May 7 08:12:53 PDT 2020
Author: Sander de Smalen
Date: 2020-05-07T16:09:31+01:00
New Revision: e76256e7c1b27087288e8fceb3b6c4aec8359389
URL: https://github.com/llvm/llvm-project/commit/e76256e7c1b27087288e8fceb3b6c4aec8359389
DIFF: https://github.com/llvm/llvm-project/commit/e76256e7c1b27087288e8fceb3b6c4aec8359389.diff
LOG: [SveEmitter] Add builtins for SVE2 Complex integer dot product
This patch adds builtins for:
- svcdot, svcdot_lane
Added:
clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_cdot.c
clang/test/CodeGen/aarch64-sve2-intrinsics/negative/acle_sve2_cdot.c
Modified:
clang/include/clang/Basic/arm_sve.td
Removed:
################################################################################
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index b12bd92d867b..880ca4e19f3d 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -1708,6 +1708,15 @@ def SVPMULLT_PAIR : SInst<"svpmullt_pair[_{d}]", "ddd", "UcUi", Mer
def SVPMULLT_PAIR_N : SInst<"svpmullt_pair[_n_{d}]", "dda", "UcUi", MergeNone, "aarch64_sve_pmullt_pair">;
}
+////////////////////////////////////////////////////////////////////////////////
+// SVE2 - Complex integer dot product
+
+let ArchGuard = "defined(__ARM_FEATURE_SVE2)" in {
+def SVCDOT : SInst<"svcdot[_{d}]", "ddqqi", "il", MergeNone, "aarch64_sve_cdot", [], [ImmCheck<3, ImmCheckComplexRotAll90>]>;
+def SVCDOT_LANE : SInst<"svcdot_lane[_{d}]", "ddqqii", "il", MergeNone, "aarch64_sve_cdot_lane", [], [ImmCheck<4, ImmCheckComplexRotAll90>,
+ ImmCheck<3, ImmCheckLaneIndexDot, 2>]>;
+}
+
////////////////////////////////////////////////////////////////////////////////
// SVE2 - Contiguous conflict detection
let ArchGuard = "defined(__ARM_FEATURE_SVE2)" in {
diff --git a/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_cdot.c b/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_cdot.c
new file mode 100644
index 000000000000..c8094c894892
--- /dev/null
+++ b/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_cdot.c
@@ -0,0 +1,123 @@
+// 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
+
+svint32_t test_svcdot_s32(svint32_t op1, svint8_t op2, svint8_t op3)
+{
+ // CHECK-LABEL: test_svcdot_s32
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.cdot.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 16 x i8> %op2, <vscale x 16 x i8> %op3, i32 0)
+ // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_s32'}}
+ return SVE_ACLE_FUNC(svcdot,_s32,,)(op1, op2, op3, 0);
+}
+
+svint32_t test_svcdot_s32_1(svint32_t op1, svint8_t op2, svint8_t op3)
+{
+ // CHECK-LABEL: test_svcdot_s32_1
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.cdot.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 16 x i8> %op2, <vscale x 16 x i8> %op3, i32 90)
+ // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_s32'}}
+ return SVE_ACLE_FUNC(svcdot,_s32,,)(op1, op2, op3, 90);
+}
+
+svint32_t test_svcdot_s32_2(svint32_t op1, svint8_t op2, svint8_t op3)
+{
+ // CHECK-LABEL: test_svcdot_s32_2
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.cdot.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 16 x i8> %op2, <vscale x 16 x i8> %op3, i32 180)
+ // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_s32'}}
+ return SVE_ACLE_FUNC(svcdot,_s32,,)(op1, op2, op3, 180);
+}
+
+svint32_t test_svcdot_s32_3(svint32_t op1, svint8_t op2, svint8_t op3)
+{
+ // CHECK-LABEL: test_svcdot_s32_3
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.cdot.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 16 x i8> %op2, <vscale x 16 x i8> %op3, i32 270)
+ // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_s32'}}
+ return SVE_ACLE_FUNC(svcdot,_s32,,)(op1, op2, op3, 270);
+}
+
+svint64_t test_svcdot_s64(svint64_t op1, svint16_t op2, svint16_t op3)
+{
+ // CHECK-LABEL: test_svcdot_s64
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.cdot.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 8 x i16> %op2, <vscale x 8 x i16> %op3, i32 0)
+ // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_s64'}}
+ return SVE_ACLE_FUNC(svcdot,_s64,,)(op1, op2, op3, 0);
+}
+
+svint64_t test_svcdot_s64_1(svint64_t op1, svint16_t op2, svint16_t op3)
+{
+ // CHECK-LABEL: test_svcdot_s64_1
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.cdot.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 8 x i16> %op2, <vscale x 8 x i16> %op3, i32 90)
+ // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_s64'}}
+ return SVE_ACLE_FUNC(svcdot,_s64,,)(op1, op2, op3, 90);
+}
+
+svint64_t test_svcdot_s64_2(svint64_t op1, svint16_t op2, svint16_t op3)
+{
+ // CHECK-LABEL: test_svcdot_s64_2
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.cdot.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 8 x i16> %op2, <vscale x 8 x i16> %op3, i32 180)
+ // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_s64'}}
+ return SVE_ACLE_FUNC(svcdot,_s64,,)(op1, op2, op3, 180);
+}
+
+svint64_t test_svcdot_s64_3(svint64_t op1, svint16_t op2, svint16_t op3)
+{
+ // CHECK-LABEL: test_svcdot_s64_3
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.cdot.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 8 x i16> %op2, <vscale x 8 x i16> %op3, i32 270)
+ // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_s64'}}
+ return SVE_ACLE_FUNC(svcdot,_s64,,)(op1, op2, op3, 270);
+}
+
+svint32_t test_svcdot_lane_s32(svint32_t op1, svint8_t op2, svint8_t op3)
+{
+ // CHECK-LABEL: test_svcdot_lane_s32
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.cdot.lane.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 16 x i8> %op2, <vscale x 16 x i8> %op3, i32 0, i32 0)
+ // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot_lane'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_lane_s32'}}
+ return SVE_ACLE_FUNC(svcdot_lane,_s32,,)(op1, op2, op3, 0, 0);
+}
+
+svint32_t test_svcdot_lane_s32_1(svint32_t op1, svint8_t op2, svint8_t op3)
+{
+ // CHECK-LABEL: test_svcdot_lane_s32_1
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.cdot.lane.nxv4i32(<vscale x 4 x i32> %op1, <vscale x 16 x i8> %op2, <vscale x 16 x i8> %op3, i32 2, i32 90)
+ // CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot_lane'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_lane_s32'}}
+ return SVE_ACLE_FUNC(svcdot_lane,_s32,,)(op1, op2, op3, 2, 90);
+}
+
+svint64_t test_svcdot_lane_s64(svint64_t op1, svint16_t op2, svint16_t op3)
+{
+ // CHECK-LABEL: test_svcdot_lane_s64
+ // CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.cdot.lane.nxv2i64(<vscale x 2 x i64> %op1, <vscale x 8 x i16> %op2, <vscale x 8 x i16> %op3, i32 0, i32 180)
+ // CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
+ // overload-warning at +2 {{implicit declaration of function 'svcdot_lane'}}
+ // expected-warning at +1 {{implicit declaration of function 'svcdot_lane_s64'}}
+ return SVE_ACLE_FUNC(svcdot_lane,_s64,,)(op1, op2, op3, 0, 180);
+}
diff --git a/clang/test/CodeGen/aarch64-sve2-intrinsics/negative/acle_sve2_cdot.c b/clang/test/CodeGen/aarch64-sve2-intrinsics/negative/acle_sve2_cdot.c
new file mode 100644
index 000000000000..cb177b943cd5
--- /dev/null
+++ b/clang/test/CodeGen/aarch64-sve2-intrinsics/negative/acle_sve2_cdot.c
@@ -0,0 +1,47 @@
+// RUN: %clang_cc1 -D__ARM_FEATURE_SVE -D__ARM_FEATURE_SVE2 -triple aarch64-none-linux-gnu -target-feature +sve2 -fallow-half-arguments-and-returns -fsyntax-only -verify -verify-ignore-unexpected=warning %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 -fsyntax-only -verify -verify-ignore-unexpected=warning %s
+
+#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
+
+#include <arm_sve.h>
+
+svint32_t test_svcdot_s32(svint32_t op1, svint8_t op2, svint8_t op3)
+{
+ // expected-error at +1 {{argument should be the value 0, 90, 180 or 270}}
+ return SVE_ACLE_FUNC(svcdot,_s32,,)(op1, op2, op3, 19);
+}
+
+svint64_t test_svcdot_s64(svint64_t op1, svint16_t op2, svint16_t op3)
+{
+ // expected-error at +1 {{argument should be the value 0, 90, 180 or 270}}
+ return SVE_ACLE_FUNC(svcdot,_s64,,)(op1, op2, op3, 19);
+}
+
+svint32_t test_svcdot_lane_s32(svint32_t op1, svint8_t op2, svint8_t op3)
+{
+ // expected-error-re at +1 {{argument value {{[0-9]+}} is outside the valid range [0, 3]}}
+ return SVE_ACLE_FUNC(svcdot_lane,_s32,,)(op1, op2, op3, -1, 0);
+}
+
+svint32_t test_svcdot_lane_s32_1(svint32_t op1, svint8_t op2, svint8_t op3)
+{
+ // expected-error at +1 {{argument should be the value 0, 90, 180 or 270}}
+ return SVE_ACLE_FUNC(svcdot_lane,_s32,,)(op1, op2, op3, 0, 19);
+}
+
+svint64_t test_svcdot_lane_s64(svint64_t op1, svint16_t op2, svint16_t op3)
+{
+ // expected-error-re at +1 {{argument value {{[0-9]+}} is outside the valid range [0, 1]}}
+ return SVE_ACLE_FUNC(svcdot_lane,_s64,,)(op1, op2, op3, -1, 0);
+}
+
+svint64_t test_svcdot_lane_s64_1(svint64_t op1, svint16_t op2, svint16_t op3)
+{
+ // expected-error at +1 {{argument should be the value 0, 90, 180 or 270}}
+ return SVE_ACLE_FUNC(svcdot_lane,_s64,,)(op1, op2, op3, 0, 19);
+}
More information about the cfe-commits
mailing list