Add vpadd isa test.
Change-Id: I84623627b69b7656f7ac4498db2a764b51ba7fd3
diff --git a/tests/kelvin_isa/BUILD b/tests/kelvin_isa/BUILD
index 158ef33..01f88d7 100644
--- a/tests/kelvin_isa/BUILD
+++ b/tests/kelvin_isa/BUILD
@@ -32,3 +32,13 @@
":kelvin_test",
],
)
+
+kelvin_test(
+ name = "vpadd",
+ srcs = [
+ "vpadd.cc",
+ ],
+ deps = [
+ ":kelvin_test",
+ ],
+)
diff --git a/tests/kelvin_isa/vpadd.cc b/tests/kelvin_isa/vpadd.cc
new file mode 100644
index 0000000..f9282f5
--- /dev/null
+++ b/tests/kelvin_isa/vpadd.cc
@@ -0,0 +1,222 @@
+#include "tests/kelvin_isa/kelvin_test.h"
+
+#define vpadd_v(T, Vd, Vs) \
+ { \
+ if (sizeof(T) == 2) { \
+ if (std::is_signed<T>::value) \
+ vpadd_h_v(Vd, Vs); \
+ else \
+ vpadd_h_u_v(Vd, Vs); \
+ } \
+ if (sizeof(T) == 4) { \
+ if (std::is_signed<T>::value) \
+ vpadd_w_v(Vd, Vs); \
+ else \
+ vpadd_w_u_v(Vd, Vs); \
+ } \
+ }
+
+#define vpadd_v_m(T, Vd, Vs) \
+ { \
+ if (sizeof(T) == 2) { \
+ if (std::is_signed<T>::value) \
+ vpadd_h_v_m(Vd, Vs); \
+ else \
+ vpadd_h_u_v_m(Vd, Vs); \
+ } \
+ if (sizeof(T) == 4) { \
+ if (std::is_signed<T>::value) \
+ vpadd_w_v_m(Vd, Vs); \
+ else \
+ vpadd_w_u_v_m(Vd, Vs); \
+ } \
+ }
+
+#define vpadd_vv(T, Vd, Vs, Vt) \
+ { \
+ if (sizeof(T) == 2) { \
+ if (std::is_signed<T>::value) \
+ vpadd_h_vv(Vd, Vs, Vt); \
+ else \
+ vpadd_h_u_vv(Vd, Vs, Vt); \
+ } \
+ if (sizeof(T) == 4) { \
+ if (std::is_signed<T>::value) \
+ vpadd_w_vv(Vd, Vs, Vt); \
+ else \
+ vpadd_w_u_vv(Vd, Vs, Vt); \
+ } \
+ }
+
+#define vpadd_vv_m(T, Vd, Vs, Vt) \
+ { \
+ if (sizeof(T) == 2) { \
+ if (std::is_signed<T>::value) \
+ vpadd_h_vv_m(Vd, Vs, Vt); \
+ else \
+ vpadd_h_u_vv_m(Vd, Vs, Vt); \
+ } \
+ if (sizeof(T) == 4) { \
+ if (std::is_signed<T>::value) \
+ vpadd_w_vv_m(Vd, Vs, Vt); \
+ else \
+ vpadd_w_u_vv_m(Vd, Vs, Vt); \
+ } \
+ }
+
+template <typename T1, typename T2>
+static void test_vpadd_v() {
+ constexpr int n = sizeof(T1);
+ int lanes;
+ getmaxvl(T2, lanes);
+
+ T1 inp[lanes * 2] __attribute__((aligned(64)));
+ T2 ref[lanes] __attribute__((aligned(64)));
+ T2 dut[lanes] __attribute__((aligned(64)));
+
+ for (int i = 0; i < lanes * 2; ++i) {
+ inp[i] = ((i & 1 ? 0xc0 : 0x40) << (8 * (n - 1))) + i - 4;
+ }
+
+ for (int i = 0; i < lanes; ++i) {
+ ref[i] = T2(inp[2 * i + 0]) + T2(inp[2 * i + 1]);
+ }
+
+ vld_x(T1, v16, inp);
+ vpadd_v(T2, v0, v16);
+ vst_x(T2, v0, dut);
+
+ for (int i = 0; i < lanes; ++i) {
+ if (ref[i] != dut[i]) {
+ printf("**error vpadd_v[%d] %d %d\n", i, ref[i], dut[i]);
+ printf(" inputs: %d, %d\n", inp[2 * i + 0], inp[2 * i + 1]);
+ exit(-1);
+ }
+ }
+}
+
+template <typename T1, typename T2>
+static void test_vpadd_v_m() {
+ constexpr int n = sizeof(T1);
+ int lanes;
+ getmaxvl_m(T2, lanes);
+
+ T1 inp[lanes * 2] __attribute__((aligned(64)));
+ T2 ref[lanes] __attribute__((aligned(64)));
+ T2 dut[lanes] __attribute__((aligned(64)));
+
+ for (int i = 0; i < lanes * 2; ++i) {
+ inp[i] = ((i & 1 ? 0xc0 : 0x40) << (8 * (n - 1))) + i - 4;
+ }
+
+ for (int i = 0; i < lanes; ++i) {
+ ref[i] = T2(inp[2 * i + 0]) + T2(inp[2 * i + 1]);
+ }
+
+ vld_x_m(T1, v16, inp);
+ vpadd_v_m(T2, v0, v16);
+ vst_x_m(T2, v0, dut);
+
+ for (int i = 0; i < lanes; ++i) {
+ if (ref[i] != dut[i]) {
+ printf("**error vpadd_v_m[%d] %x %x\n", i, ref[i], dut[i]);
+ exit(-1);
+ }
+ }
+}
+
+template <typename T1, typename T2>
+static void test_vpadd_vv() {
+ constexpr int n = sizeof(T1);
+ int lanes;
+ getmaxvl(T2, lanes);
+
+ T1 inp[2][lanes * 2] __attribute__((aligned(64)));
+ T2 ref[2][lanes] __attribute__((aligned(64)));
+ T2 dut[2][lanes] __attribute__((aligned(64)));
+
+ for (int i = 0; i < lanes * 2; ++i) {
+ inp[0][i] = ((i & 1 ? 0xc0 : 0x40) << (8 * (n - 1))) + i - 4;
+ inp[1][i] = ((i & 1 ? 0x80 : 0x20) << (8 * (n - 1))) + i;
+ }
+
+ for (int i = 0; i < lanes; ++i) {
+ ref[0][i] = T2(inp[0][2 * i + 0]) + T2(inp[0][2 * i + 1]);
+ ref[1][i] = T2(inp[1][2 * i + 0]) + T2(inp[1][2 * i + 1]);
+ }
+
+ vld_x(T1, v16, inp[0]);
+ vld_x(T1, v20, inp[1]);
+ vpadd_vv(T2, v0, v16, v20);
+ vst_x(T2, v0, dut[0]);
+ vst_x(T2, v1, dut[1]);
+
+ for (int j = 0; j < 2; ++j) {
+ for (int i = 0; i < lanes; ++i) {
+ if (ref[j][i] != dut[j][i]) {
+ printf("**error vpadd_vv[%d,%d] %x %x\n", j, i, ref[j][i], dut[j][i]);
+ exit(-1);
+ }
+ }
+ }
+}
+
+template <typename T1, typename T2>
+static void test_vpadd_vv_m() {
+ constexpr int n = sizeof(T1);
+ int lanes;
+ getmaxvl_m(T2, lanes);
+
+ T1 inp[2][lanes * 2] __attribute__((aligned(64)));
+ T2 ref[2][lanes] __attribute__((aligned(64)));
+ T2 dut[2][lanes] __attribute__((aligned(64)));
+
+ for (int i = 0; i < lanes * 2; ++i) {
+ inp[0][i] = ((i & 1 ? 0xc0 : 0x40) << (8 * (n - 1))) + i - 4;
+ inp[1][i] = ((i & 1 ? 0x80 : 0x20) << (8 * (n - 1))) + i;
+ }
+
+ for (int i = 0; i < lanes; ++i) {
+ ref[0][i] = T2(inp[0][2 * i + 0]) + T2(inp[0][2 * i + 1]);
+ ref[1][i] = T2(inp[1][2 * i + 0]) + T2(inp[1][2 * i + 1]);
+ }
+
+ vld_x_m(T1, v16, inp[0]);
+ vld_x_m(T1, v20, inp[1]);
+ vpadd_vv_m(T2, v0, v16, v20);
+ vst_x_m(T2, v0, dut[0]);
+ vst_x_m(T2, v4, dut[1]);
+
+ for (int j = 0; j < 2; ++j) {
+ for (int i = 0; i < lanes; ++i) {
+ if (ref[j][i] != dut[j][i]) {
+ printf("**error vpadd_vv_m[%d,%d] %x %x\n", j, i, ref[j][i], dut[j][i]);
+ exit(-1);
+ }
+ }
+ }
+}
+
+int main() {
+ test_vpadd_v<int8_t, int16_t>();
+ test_vpadd_v<int16_t, int32_t>();
+ test_vpadd_v<uint8_t, uint16_t>();
+ test_vpadd_v<uint16_t, uint32_t>();
+
+ test_vpadd_v_m<int8_t, int16_t>();
+ test_vpadd_v_m<int16_t, int32_t>();
+ test_vpadd_v_m<uint8_t, uint16_t>();
+ test_vpadd_v_m<uint16_t, uint32_t>();
+
+ test_vpadd_vv<int8_t, int16_t>();
+ test_vpadd_vv<int16_t, int32_t>();
+ test_vpadd_vv<uint8_t, uint16_t>();
+ test_vpadd_vv<uint16_t, uint32_t>();
+
+ test_vpadd_vv_m<int8_t, int16_t>();
+ test_vpadd_vv_m<int16_t, int32_t>();
+ test_vpadd_vv_m<uint8_t, uint16_t>();
+ test_vpadd_vv_m<uint16_t, uint32_t>();
+
+ return 0;
+}
