blob: 3dfed335f51fd60d7c84f7a2b838578c35fd2c43 [file] [log] [blame]
#include <limits.h>
#include <riscv_vector.h>
#include <springbok.h>
#include <stdio.h>
#include <stdlib.h>
#include <bit>
#include <tuple>
#include "pw_unit_test/framework.h"
#include "test_v_helpers.h"
namespace vmax_vx_test {
namespace {
using namespace test_v_helpers;
uint8_t test_vector_1[MAXVL_BYTES];
uint8_t test_vector_2[MAXVL_BYTES];
class VmaxVxTest : public ::testing::Test {
protected:
void SetUp() override { zero_vector_registers(); }
void TearDown() override { zero_vector_registers(); }
};
// Below is a non-macro version of the test for more convenient debugging.
// Remove the "DISABLED_" prefix to enable this test for debugging.
TEST_F(VmaxVxTest, DISABLED_vmax_vx_demo) {
for (int i = 0; i < AVL_COUNT; i++) {
int32_t avl = AVLS[i];
int vlmax;
int vl;
std::tie(vlmax, vl) = vector_test_setup<uint8_t>(
VLMUL::LMUL_M1, avl, {test_vector_1, test_vector_2});
if (avl > vlmax) {
continue;
}
int8_t *ptr_vec_1 = reinterpret_cast<int8_t *>(test_vector_1);
int8_t *ptr_vec_2 = reinterpret_cast<int8_t *>(test_vector_2);
const int8_t test_val = 8;
for (int idx = 0; idx < vl; idx++) {
ptr_vec_1[idx] = idx;
}
__asm__ volatile("vle8.v v8, (%0)" : : "r"(ptr_vec_1));
__asm__ volatile("vmax.vx v16, v8, %[RS1]" ::[RS1] "r"(test_val));
for (int idx = 0; idx < vl; idx++) {
ptr_vec_1[idx] = (ptr_vec_1[idx] > test_val) ? ptr_vec_1[idx] : test_val;
}
__asm__ volatile("vse8.v v16, (%0)" : : "r"(ptr_vec_2));
assert_vec_elem_eq<int8_t>(vlmax, test_vector_1, test_vector_2);
}
}
#define DEFINE_TEST_VMAX_VX(_SEW_, _LMUL_, TEST_VAL) \
TEST_F(VmaxVxTest, vmax_vx##_SEW_##m##_LMUL_) { \
for (int i = 0; i < AVL_COUNT; i++) { \
int32_t avl = AVLS[i]; \
int vlmax; \
int vl; \
std::tie(vlmax, vl) = vector_test_setup<int##_SEW_##_t>( \
VLMUL::LMUL_M##_LMUL_, avl, {test_vector_1, test_vector_2}); \
if (avl > vlmax) { \
continue; \
} \
int##_SEW_##_t *ptr_vec_1 = \
reinterpret_cast<int##_SEW_##_t *>(test_vector_1); \
int##_SEW_##_t *ptr_vec_2 = \
reinterpret_cast<int##_SEW_##_t *>(test_vector_2); \
const int##_SEW_##_t test_val = TEST_VAL; \
for (int idx = 0; idx < vl; idx++) { \
ptr_vec_1[idx] = idx; \
} \
__asm__ volatile("vle" #_SEW_ ".v v8, (%0)" : : "r"(ptr_vec_1)); \
__asm__ volatile("vmax.vx v16, v8, %[RS1]" ::[RS1] "r"(test_val)); \
for (int idx = 0; idx < vl; idx++) { \
ptr_vec_1[idx] = \
(ptr_vec_1[idx] > test_val) ? ptr_vec_1[idx] : test_val; \
} \
__asm__ volatile("vse" #_SEW_ ".v v16, (%0)" : : "r"(ptr_vec_2)); \
assert_vec_elem_eq<int##_SEW_##_t>(vlmax, test_vector_1, test_vector_2); \
} \
}
// TODO(gkielian): modify macro to permit more than one test per sew/lmul pair
DEFINE_TEST_VMAX_VX(8, 1, INT8_MIN)
DEFINE_TEST_VMAX_VX(8, 2, INT8_MAX)
DEFINE_TEST_VMAX_VX(8, 4, -1)
DEFINE_TEST_VMAX_VX(8, 8, 2)
DEFINE_TEST_VMAX_VX(16, 1, INT16_MIN)
DEFINE_TEST_VMAX_VX(16, 2, INT16_MAX)
DEFINE_TEST_VMAX_VX(16, 4, -1)
DEFINE_TEST_VMAX_VX(16, 8, 2)
DEFINE_TEST_VMAX_VX(32, 1, INT32_MIN)
DEFINE_TEST_VMAX_VX(32, 2, INT32_MAX)
DEFINE_TEST_VMAX_VX(32, 4, -1)
DEFINE_TEST_VMAX_VX(32, 8, 2)
} // namespace
} // namespace vmax_vx_test