github-mirrors/vortex
1
0
Fork 0
mirror of https://github.com/vortexgpgpu/vortex.git synced 2025-04-23 21:39:10 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

adding dogfood fclamp test

Browse source
This commit is contained in:
Blaise Tine 2024-03-27 06:40:13 -07:00
parent 70f2f58ac9
commit 697e2fffbb
2 changed files with 37 additions and 37 deletions
Download patch file Download diff file Expand all files Collapse all files

52
tests/regression/dogfood/kernel.cpp
View file

@ -306,7 +306,19 @@ void kernel_utof(int task_id, kernel_arg_t* __UNIFORM__ arg) {
}
}
void kernel_minmax(int task_id, kernel_arg_t* __UNIFORM__ arg) {
float clamp(float a, float b, float c) {
float result;
asm (
"fmin.s %[result], %[a], %[c]\n\t" // result = min(a, c)
"fmax.s %[result], %[result], %[b]" // result = max(result, b)
: [result] "=f" (result) // Output
: [a] "f" (a), [b] "f" (b), [c] "f" (c) // Inputs
: // No clobbered registers
);
return result;
}
void kernel_fclamp(int task_id, kernel_arg_t* __UNIFORM__ arg) {
auto count = arg->task_size;
auto src0_ptr = (float*)arg->src0_addr;
auto src1_ptr = (float*)arg->src1_addr;
@ -316,14 +328,12 @@ void kernel_minmax(int task_id, kernel_arg_t* __UNIFORM__ arg) {
for (uint32_t i = 0; i < count; ++i) {
auto a = src0_ptr[offset+i];
auto b = src1_ptr[offset+i];
auto c = fmin(a, b);
auto d = fmax(a, b);
auto e = c + d;
dst_ptr[offset+i] = e;
dst_ptr[offset+i] = clamp(1.0f, a, b);
}
}
void kernel_bar(int task_id, kernel_arg_t* __UNIFORM__ arg) {
auto num_cores = vx_num_cores();
auto num_warps = vx_num_warps();
auto num_threads = vx_num_threads();
@ -334,11 +344,14 @@ void kernel_bar(int task_id, kernel_arg_t* __UNIFORM__ arg) {
auto src0_ptr = (uint32_t*)arg->src0_addr;
auto dst_ptr = (uint32_t*)arg->dst_addr;
// per warp delay
uint32_t barrier_stall = 0;
for (int i = 0; i <= wid; ++i) {
barrier_stall += src0_ptr[0] * src0_ptr[i];
}
// update destination using the first threads in core
if (wid == 0 && tid == 0) {
int block_size = arg->num_tasks / num_cores;
int offset = cid * block_size;
for (int i = 0; i <= block_size; ++i) {
dst_ptr[i + offset] = src0_ptr[i + offset];
}
}
// memory fence
vx_fence();
@ -347,8 +360,7 @@ void kernel_bar(int task_id, kernel_arg_t* __UNIFORM__ arg) {
vx_barrier(0, num_warps);
// update destination
auto src_idx = (cid * num_warps + (num_warps - 1 - wid)) * num_threads + tid;
dst_ptr[task_id] = src0_ptr[src_idx] + barrier_stall;
dst_ptr[task_id] += 1;
}
void kernel_gbar(int task_id, kernel_arg_t* __UNIFORM__ arg) {
@ -363,13 +375,12 @@ void kernel_gbar(int task_id, kernel_arg_t* __UNIFORM__ arg) {
auto src0_ptr = (uint32_t*)arg->src0_addr;
auto dst_ptr = (uint32_t*)arg->dst_addr;
// per core delay
uint32_t barrier_stall = 0;
for (int i = 0; i <= cid; ++i) {
for (int j = 0; j <= wid; ++j) {
barrier_stall += src0_ptr[0] * src0_ptr[i + j];
// update destination using the first threads in processor
if (cid == 0 && wid == 0 && tid == 0) {
for (int i = 0, n = arg->num_tasks; i <= n; ++i) {
dst_ptr[i] = src0_ptr[i];
}
}
}
// memory fence
vx_fence();
@ -378,8 +389,7 @@ void kernel_gbar(int task_id, kernel_arg_t* __UNIFORM__ arg) {
vx_barrier(0x80000000, num_cores);
// update destination
auto src_idx = ((num_cores - 1 - cid) * num_warps + (num_warps - 1 - wid)) * num_threads + tid;
dst_ptr[task_id] = src0_ptr[src_idx] + barrier_stall;
dst_ptr[task_id] += 1;
}
static const PFN_Kernel sc_tests[] = {
@ -402,7 +412,7 @@ static const PFN_Kernel sc_tests[] = {
kernel_ftou,
kernel_itof,
kernel_utof,
kernel_minmax,
kernel_fclamp,
kernel_bar,
kernel_gbar
};

22
tests/regression/dogfood/testcases.h
View file

@ -689,9 +689,9 @@ public:
}
};
class Test_MINMAX : public ITestCase {
class Test_FCLAMP : public ITestCase {
public:
Test_MINMAX(TestSuite* suite) : ITestCase(suite, "minmax") {}
Test_FCLAMP(TestSuite* suite) : ITestCase(suite, "fclamp") {}
int setup(uint32_t n, void* src1, void* src2) override {
auto a = (float*)src1;
@ -709,9 +709,7 @@ public:
auto b = (float*)src2;
auto c = (float*)dst;
for (uint32_t i = 0; i < n; ++i) {
auto x = fmin(a[i], b[i]);
auto y = fmax(a[i], b[i]);
auto ref = x + y;
auto ref = fmin(fmax(1.0f, a[i]), b[i]);
if (!almost_equal(c[i], ref)) {
std::cout << "error at result #" << i << ": expected=" << ref << ", actual=" << c[i] << ", a=" << a[i] << ", b=" << b[i] << std::endl;
++errors;
@ -744,11 +742,7 @@ public:
auto a = (uint32_t*)src1;
auto c = (uint32_t*)dst;
for (uint32_t i = 0; i < n; ++i) {
auto tid = i % num_threads_;
auto wid = (i / num_threads_) % num_warps_;
auto cid = i / (num_warps_ * num_threads_);
auto src_idx = (cid * num_warps_ + (num_warps_ - 1 - wid)) * num_threads_ + tid;
uint32_t ref = a[src_idx];
uint32_t ref = a[i] + 1;
if (c[i] != ref) {
std::cout << "error at result #" << i << ": expected=" << std::hex << ref << ", actual=" << c[i] << std::endl;
++errors;
@ -785,11 +779,7 @@ public:
auto a = (uint32_t*)src1;
auto c = (uint32_t*)dst;
for (uint32_t i = 0; i < n; ++i) {
auto tid = i % num_threads_;
auto wid = (i / num_threads_) % num_warps_;
auto cid = i / (num_warps_ * num_threads_);
auto src_idx = ((num_cores_ - 1 - cid) * num_warps_ + (num_warps_ - 1 - wid)) * num_threads_ + tid;
uint32_t ref = a[src_idx];
uint32_t ref = a[i] + 1;
if (c[i] != ref) {
std::cout << "error at result #" << i << ": expected=" << std::hex << ref << ", actual=" << c[i] << std::endl;
++errors;
@ -826,7 +816,7 @@ TestSuite::TestSuite(vx_device_h device)
this->add_test(new Test_FTOU(this));
this->add_test(new Test_ITOF(this));
this->add_test(new Test_UTOF(this));
// this->add_test(new Test_MINMAX(this)); TODO: ISSUE #117
this->add_test(new Test_FCLAMP(this));
this->add_test(new Test_BAR(this));
this->add_test(new Test_GBAR(this));
}