struct amdgpu_buffer *amdgpu_bo_open(amdgpu_device_handle pDev,
uint32_t alloc_size,
uint32_t phys_alignment,
uint32_t domains)
{
struct amdgpu_bo_alloc_request alloc_request;
struct amdgpu_bo_alloc_result buffer;
struct amdgpu_buffer *bo = NULL;
memset(&alloc_request, 0, sizeof(struct amdgpu_bo_alloc_request));
memset(&buffer, 0, sizeof(struct amdgpu_bo_alloc_result));
bo = (struct amdgpu_buffer *)calloc(1, sizeof(struct amdgpu_buffer));
if (bo == NULL) {
return NULL;
}
alloc_request.alloc_size = alloc_size;
alloc_request.phys_alignment = phys_alignment;
alloc_request.preferred_heap = domains;
if (amdgpu_bo_alloc(pDev, &alloc_request, &buffer)) {
free(bo);
return NULL;
}
bo->bo.amdgpu = buffer.buf_handle;
bo->ref_count = 1;
return bo;
}
static void alloc_resource(struct amdgpu_vce_bo *vce_bo, unsigned size, unsigned domain)
{
struct amdgpu_bo_alloc_request req = {0};
amdgpu_bo_handle buf_handle;
amdgpu_va_handle va_handle;
uint64_t va = 0;
int r;
req.alloc_size = ALIGN(size, 4096);
req.preferred_heap = domain;
r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_va_range_alloc(device_handle,
amdgpu_gpu_va_range_general,
req.alloc_size, 1, 0, &va,
&va_handle, 0);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_va_op(buf_handle, 0, req.alloc_size, va, 0,
AMDGPU_VA_OP_MAP);
CU_ASSERT_EQUAL(r, 0);
vce_bo->addr = va;
vce_bo->handle = buf_handle;
vce_bo->size = req.alloc_size;
vce_bo->va_handle = va_handle;
r = amdgpu_bo_cpu_map(vce_bo->handle, (void **)&vce_bo->ptr);
CU_ASSERT_EQUAL(r, 0);
memset(vce_bo->ptr, 0, size);
r = amdgpu_bo_cpu_unmap(vce_bo->handle);
CU_ASSERT_EQUAL(r, 0);
}
static void amdgpu_cs_uvd_destroy(void)
{
struct amdgpu_bo_alloc_request req = {0};
amdgpu_bo_handle buf_handle;
amdgpu_va_handle va_handle;
uint64_t va = 0;
void *msg;
int i, r;
req.alloc_size = 4*1024;
req.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_va_range_alloc(device_handle,
amdgpu_gpu_va_range_general,
req.alloc_size, 1, 0, &va,
&va_handle, 0);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_va_op(buf_handle, 0, req.alloc_size, va, 0,
AMDGPU_VA_OP_MAP);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_cpu_map(buf_handle, &msg);
CU_ASSERT_EQUAL(r, 0);
memcpy(msg, uvd_destroy_msg, sizeof(uvd_destroy_msg));
if (family_id >= AMDGPU_FAMILY_VI)
((uint8_t*)msg)[0x10] = 7;
r = amdgpu_bo_cpu_unmap(buf_handle);
CU_ASSERT_EQUAL(r, 0);
num_resources = 0;
resources[num_resources++] = buf_handle;
resources[num_resources++] = ib_handle;
i = 0;
uvd_cmd(va, 0x0, &i);
for (; i % 16; ++i)
ib_cpu[i] = 0x80000000;
r = submit(i, AMDGPU_HW_IP_UVD);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_va_op(buf_handle, 0, req.alloc_size, va, 0, AMDGPU_VA_OP_UNMAP);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_va_range_free(va_handle);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_free(buf_handle);
CU_ASSERT_EQUAL(r, 0);
}
int suite_bo_tests_init(void)
{
struct amdgpu_bo_alloc_request req = {0};
amdgpu_bo_handle buf_handle;
uint64_t va;
int r;
r = amdgpu_device_initialize(drm_amdgpu[0], &major_version,
&minor_version, &device_handle);
if (r) {
if ((r == -EACCES) && (errno == EACCES))
printf("\n\nError:%s. "
"Hint:Try to run this test program as root.",
strerror(errno));
return CUE_SINIT_FAILED;
}
req.alloc_size = BUFFER_SIZE;
req.phys_alignment = BUFFER_ALIGN;
req.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
if (r)
return CUE_SINIT_FAILED;
r = amdgpu_va_range_alloc(device_handle,
amdgpu_gpu_va_range_general,
BUFFER_SIZE, BUFFER_ALIGN, 0,
&va, &va_handle, 0);
if (r)
goto error_va_alloc;
r = amdgpu_bo_va_op(buf_handle, 0, BUFFER_SIZE, va, 0, AMDGPU_VA_OP_MAP);
if (r)
goto error_va_map;
buffer_handle = buf_handle;
virtual_mc_base_address = va;
return CUE_SUCCESS;
error_va_map:
amdgpu_va_range_free(va_handle);
error_va_alloc:
amdgpu_bo_free(buf_handle);
return CUE_SINIT_FAILED;
}
static struct radeon_winsys_ctx *amdgpu_ctx_create(struct radeon_winsys *ws)
{
struct amdgpu_ctx *ctx = CALLOC_STRUCT(amdgpu_ctx);
int r;
struct amdgpu_bo_alloc_request alloc_buffer = {};
amdgpu_bo_handle buf_handle;
ctx->ws = amdgpu_winsys(ws);
ctx->refcount = 1;
r = amdgpu_cs_ctx_create(ctx->ws->dev, &ctx->ctx);
if (r) {
fprintf(stderr, "amdgpu: amdgpu_cs_ctx_create failed. (%i)\n", r);
FREE(ctx);
return NULL;
}
alloc_buffer.alloc_size = 4 * 1024;
alloc_buffer.phys_alignment = 4 *1024;
alloc_buffer.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
r = amdgpu_bo_alloc(ctx->ws->dev, &alloc_buffer, &buf_handle);
if (r) {
fprintf(stderr, "amdgpu: amdgpu_bo_alloc failed. (%i)\n", r);
amdgpu_cs_ctx_free(ctx->ctx);
FREE(ctx);
return NULL;
}
r = amdgpu_bo_cpu_map(buf_handle, (void**)&ctx->user_fence_cpu_address_base);
if (r) {
fprintf(stderr, "amdgpu: amdgpu_bo_cpu_map failed. (%i)\n", r);
amdgpu_bo_free(buf_handle);
amdgpu_cs_ctx_free(ctx->ctx);
FREE(ctx);
return NULL;
}
memset(ctx->user_fence_cpu_address_base, 0, alloc_buffer.alloc_size);
ctx->user_fence_bo = buf_handle;
return (struct radeon_winsys_ctx*)ctx;
}
static void amdgpu_mem_fail_alloc(void)
{
amdgpu_bo_handle bo;
int r;
struct amdgpu_bo_alloc_request req = {0};
amdgpu_bo_handle buf_handle;
/* Test impossible mem allocation, 1TB */
req.alloc_size = 0xE8D4A51000;
req.phys_alignment = 4096;
req.preferred_heap = AMDGPU_GEM_DOMAIN_VRAM;
req.flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
CU_ASSERT_EQUAL(r, -ENOMEM);
if (!r) {
r = amdgpu_bo_free(bo);
CU_ASSERT_EQUAL(r, 0);
}
}
static struct pb_buffer *amdgpu_bomgr_create_bo(struct pb_manager *_mgr,
pb_size size,
const struct pb_desc *desc)
{
struct amdgpu_winsys *rws = get_winsys(_mgr);
struct amdgpu_bo_desc *rdesc = (struct amdgpu_bo_desc*)desc;
struct amdgpu_bo_alloc_request request = {0};
amdgpu_bo_handle buf_handle;
uint64_t va = 0;
struct amdgpu_winsys_bo *bo;
amdgpu_va_handle va_handle;
int r;
assert(rdesc->initial_domain & RADEON_DOMAIN_VRAM_GTT);
bo = CALLOC_STRUCT(amdgpu_winsys_bo);
if (!bo) {
return NULL;
}
request.alloc_size = size;
request.phys_alignment = desc->alignment;
if (rdesc->initial_domain & RADEON_DOMAIN_VRAM) {
request.preferred_heap |= AMDGPU_GEM_DOMAIN_VRAM;
if (rdesc->flags & RADEON_FLAG_CPU_ACCESS)
request.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
}
if (rdesc->initial_domain & RADEON_DOMAIN_GTT) {
request.preferred_heap |= AMDGPU_GEM_DOMAIN_GTT;
if (rdesc->flags & RADEON_FLAG_GTT_WC)
request.flags |= AMDGPU_GEM_CREATE_CPU_GTT_USWC;
}
r = amdgpu_bo_alloc(rws->dev, &request, &buf_handle);
if (r) {
fprintf(stderr, "amdgpu: Failed to allocate a buffer:\n");
fprintf(stderr, "amdgpu: size : %d bytes\n", size);
fprintf(stderr, "amdgpu: alignment : %d bytes\n", desc->alignment);
fprintf(stderr, "amdgpu: domains : %d\n", rdesc->initial_domain);
goto error_bo_alloc;
}
r = amdgpu_va_range_alloc(rws->dev, amdgpu_gpu_va_range_general,
size, desc->alignment, 0, &va, &va_handle, 0);
if (r)
goto error_va_alloc;
r = amdgpu_bo_va_op(buf_handle, 0, size, va, 0, AMDGPU_VA_OP_MAP);
if (r)
goto error_va_map;
pipe_reference_init(&bo->base.reference, 1);
bo->base.alignment = desc->alignment;
bo->base.usage = desc->usage;
bo->base.size = size;
bo->base.vtbl = &amdgpu_winsys_bo_vtbl;
bo->rws = rws;
bo->bo = buf_handle;
bo->va = va;
bo->va_handle = va_handle;
bo->initial_domain = rdesc->initial_domain;
bo->unique_id = __sync_fetch_and_add(&rws->next_bo_unique_id, 1);
if (rdesc->initial_domain & RADEON_DOMAIN_VRAM)
rws->allocated_vram += align(size, rws->gart_page_size);
else if (rdesc->initial_domain & RADEON_DOMAIN_GTT)
rws->allocated_gtt += align(size, rws->gart_page_size);
return &bo->base;
error_va_map:
amdgpu_va_range_free(va_handle);
error_va_alloc:
amdgpu_bo_free(buf_handle);
error_bo_alloc:
FREE(bo);
return NULL;
}
static void amdgpu_cs_uvd_decode(void)
{
const unsigned dpb_size = 15923584, ctx_size = 5287680, dt_size = 737280;
uint64_t msg_addr, fb_addr, bs_addr, dpb_addr, ctx_addr, dt_addr, it_addr;
struct amdgpu_bo_alloc_request req = {0};
amdgpu_bo_handle buf_handle;
amdgpu_va_handle va_handle;
uint64_t va = 0;
uint64_t sum;
uint8_t *ptr;
int i, r;
req.alloc_size = 4*1024; /* msg */
req.alloc_size += 4*1024; /* fb */
if (family_id >= AMDGPU_FAMILY_VI)
req.alloc_size += 4096; /*it_scaling_table*/
req.alloc_size += ALIGN(sizeof(uvd_bitstream), 4*1024);
req.alloc_size += ALIGN(dpb_size, 4*1024);
req.alloc_size += ALIGN(dt_size, 4*1024);
req.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_va_range_alloc(device_handle,
amdgpu_gpu_va_range_general,
req.alloc_size, 1, 0, &va,
&va_handle, 0);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_va_op(buf_handle, 0, req.alloc_size, va, 0,
AMDGPU_VA_OP_MAP);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_cpu_map(buf_handle, (void **)&ptr);
CU_ASSERT_EQUAL(r, 0);
memcpy(ptr, uvd_decode_msg, sizeof(uvd_create_msg));
if (family_id >= AMDGPU_FAMILY_VI) {
ptr[0x10] = 7;
ptr[0x98] = 0x00;
ptr[0x99] = 0x02;
/* chip polaris10/11 */
if (chip_id == chip_rev+0x50 || chip_id == chip_rev+0x5A) {
/*dpb size */
ptr[0x24] = 0x00;
ptr[0x25] = 0x94;
ptr[0x26] = 0x6B;
ptr[0x27] = 0x00;
/*ctx size */
ptr[0x2C] = 0x00;
ptr[0x2D] = 0xAF;
ptr[0x2E] = 0x50;
ptr[0x2F] = 0x00;
}
}
ptr += 4*1024;
memset(ptr, 0, 4*1024);
if (family_id >= AMDGPU_FAMILY_VI) {
ptr += 4*1024;
memcpy(ptr, uvd_it_scaling_table, sizeof(uvd_it_scaling_table));
}
ptr += 4*1024;
memcpy(ptr, uvd_bitstream, sizeof(uvd_bitstream));
ptr += ALIGN(sizeof(uvd_bitstream), 4*1024);
memset(ptr, 0, dpb_size);
ptr += ALIGN(dpb_size, 4*1024);
memset(ptr, 0, dt_size);
num_resources = 0;
resources[num_resources++] = buf_handle;
resources[num_resources++] = ib_handle;
msg_addr = va;
fb_addr = msg_addr + 4*1024;
if (family_id >= AMDGPU_FAMILY_VI) {
it_addr = fb_addr + 4*1024;
bs_addr = it_addr + 4*1024;
} else
bs_addr = fb_addr + 4*1024;
dpb_addr = ALIGN(bs_addr + sizeof(uvd_bitstream), 4*1024);
if ((family_id >= AMDGPU_FAMILY_VI) &&
(chip_id == chip_rev+0x50 || chip_id == chip_rev+0x5A)) {
ctx_addr = ALIGN(dpb_addr + 0x006B9400, 4*1024);
}
dt_addr = ALIGN(dpb_addr + dpb_size, 4*1024);
i = 0;
uvd_cmd(msg_addr, 0x0, &i);
uvd_cmd(dpb_addr, 0x1, &i);
uvd_cmd(dt_addr, 0x2, &i);
uvd_cmd(fb_addr, 0x3, &i);
uvd_cmd(bs_addr, 0x100, &i);
if (family_id >= AMDGPU_FAMILY_VI) {
uvd_cmd(it_addr, 0x204, &i);
if (chip_id == chip_rev+0x50 || chip_id == chip_rev+0x5A)
uvd_cmd(ctx_addr, 0x206, &i);
}
ib_cpu[i++] = 0x3BC6;
ib_cpu[i++] = 0x1;
for (; i % 16; ++i)
ib_cpu[i] = 0x80000000;
r = submit(i, AMDGPU_HW_IP_UVD);
CU_ASSERT_EQUAL(r, 0);
/* TODO: use a real CRC32 */
for (i = 0, sum = 0; i < dt_size; ++i)
sum += ptr[i];
CU_ASSERT_EQUAL(sum, 0x20345d8);
r = amdgpu_bo_cpu_unmap(buf_handle);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_va_op(buf_handle, 0, req.alloc_size, va, 0, AMDGPU_VA_OP_UNMAP);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_va_range_free(va_handle);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_free(buf_handle);
CU_ASSERT_EQUAL(r, 0);
}
static void amdgpu_cs_uvd_create(void)
{
struct amdgpu_bo_alloc_request req = {0};
amdgpu_bo_handle buf_handle;
uint64_t va = 0;
amdgpu_va_handle va_handle;
void *msg;
int i, r;
req.alloc_size = 4*1024;
req.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_va_range_alloc(device_handle,
amdgpu_gpu_va_range_general,
4096, 1, 0, &va,
&va_handle, 0);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_va_op(buf_handle, 0, 4096, va, 0, AMDGPU_VA_OP_MAP);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_cpu_map(buf_handle, &msg);
CU_ASSERT_EQUAL(r, 0);
memcpy(msg, uvd_create_msg, sizeof(uvd_create_msg));
if (family_id >= AMDGPU_FAMILY_VI) {
((uint8_t*)msg)[0x10] = 7;
/* chip polaris 10/11 */
if (chip_id == chip_rev+0x50 || chip_id == chip_rev+0x5A) {
/* dpb size */
((uint8_t*)msg)[0x28] = 0x00;
((uint8_t*)msg)[0x29] = 0x94;
((uint8_t*)msg)[0x2A] = 0x6B;
((uint8_t*)msg)[0x2B] = 0x00;
}
}
r = amdgpu_bo_cpu_unmap(buf_handle);
CU_ASSERT_EQUAL(r, 0);
num_resources = 0;
resources[num_resources++] = buf_handle;
resources[num_resources++] = ib_handle;
i = 0;
uvd_cmd(va, 0x0, &i);
for (; i % 16; ++i)
ib_cpu[i] = 0x80000000;
r = submit(i, AMDGPU_HW_IP_UVD);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_va_op(buf_handle, 0, 4096, va, 0, AMDGPU_VA_OP_UNMAP);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_va_range_free(va_handle);
CU_ASSERT_EQUAL(r, 0);
r = amdgpu_bo_free(buf_handle);
CU_ASSERT_EQUAL(r, 0);
}
static struct amdgpu_winsys_bo *amdgpu_create_bo(struct amdgpu_winsys *ws,
unsigned size,
unsigned alignment,
unsigned usage,
enum radeon_bo_domain initial_domain,
unsigned flags)
{
struct amdgpu_bo_alloc_request request = {0};
amdgpu_bo_handle buf_handle;
uint64_t va = 0;
struct amdgpu_winsys_bo *bo;
amdgpu_va_handle va_handle;
int r;
assert(initial_domain & RADEON_DOMAIN_VRAM_GTT);
bo = CALLOC_STRUCT(amdgpu_winsys_bo);
if (!bo) {
return NULL;
}
pb_cache_init_entry(&ws->bo_cache, &bo->cache_entry, &bo->base);
request.alloc_size = size;
request.phys_alignment = alignment;
if (initial_domain & RADEON_DOMAIN_VRAM) {
request.preferred_heap |= AMDGPU_GEM_DOMAIN_VRAM;
if (flags & RADEON_FLAG_CPU_ACCESS)
request.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
}
if (initial_domain & RADEON_DOMAIN_GTT) {
request.preferred_heap |= AMDGPU_GEM_DOMAIN_GTT;
if (flags & RADEON_FLAG_GTT_WC)
request.flags |= AMDGPU_GEM_CREATE_CPU_GTT_USWC;
}
r = amdgpu_bo_alloc(ws->dev, &request, &buf_handle);
if (r) {
fprintf(stderr, "amdgpu: Failed to allocate a buffer:\n");
fprintf(stderr, "amdgpu: size : %d bytes\n", size);
fprintf(stderr, "amdgpu: alignment : %d bytes\n", alignment);
fprintf(stderr, "amdgpu: domains : %d\n", initial_domain);
goto error_bo_alloc;
}
r = amdgpu_va_range_alloc(ws->dev, amdgpu_gpu_va_range_general,
size, alignment, 0, &va, &va_handle, 0);
if (r)
goto error_va_alloc;
r = amdgpu_bo_va_op(buf_handle, 0, size, va, 0, AMDGPU_VA_OP_MAP);
if (r)
goto error_va_map;
pipe_reference_init(&bo->base.reference, 1);
bo->base.alignment = alignment;
bo->base.usage = usage;
bo->base.size = size;
bo->base.vtbl = &amdgpu_winsys_bo_vtbl;
bo->ws = ws;
bo->bo = buf_handle;
bo->va = va;
bo->va_handle = va_handle;
bo->initial_domain = initial_domain;
bo->unique_id = __sync_fetch_and_add(&ws->next_bo_unique_id, 1);
if (initial_domain & RADEON_DOMAIN_VRAM)
ws->allocated_vram += align(size, ws->gart_page_size);
else if (initial_domain & RADEON_DOMAIN_GTT)
ws->allocated_gtt += align(size, ws->gart_page_size);
return bo;
error_va_map:
amdgpu_va_range_free(va_handle);
error_va_alloc:
amdgpu_bo_free(buf_handle);
error_bo_alloc:
FREE(bo);
return NULL;
}
