amd, radv: create null device without winsys

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37525>

4 files changed, 243 insertions, 46 deletions

diff --git a/src/amd/common/ac_null_device.c b/src/amd/common/ac_null_device.c
new file mode 100644
index 00000000000..088203731c3
--- /dev/null
+++ b/src/amd/common/ac_null_device.c
@@ -0,0 +1,117 @@
+/*
+ * Copyright © 2020 Valve Corporation
+ *
+ * SPDX-License-Identifier: MIT
+ */
+#include "ac_null_device.h"
+#include "ac_gpu_info.h"
+#include "util/u_string.h"
+
+bool
+ac_null_device_create(struct radeon_info *gpu_info, const char *family)
+{
+   unsigned i;
+
+   gpu_info->gfx_level = CLASS_UNKNOWN;
+   gpu_info->family = CHIP_UNKNOWN;
+
+   for (i = CHIP_TAHITI; i < CHIP_LAST; i++) {
+      if (!strcasecmp(family, ac_get_family_name(i))) {
+         /* Override family and gfx_level. */
+         gpu_info->family = i;
+         gpu_info->name = ac_get_family_name(i);
+
+         if (gpu_info->family >= CHIP_GFX1200)
+            gpu_info->gfx_level = GFX12;
+         else if (gpu_info->family >= CHIP_NAVI31)
+            gpu_info->gfx_level = GFX11;
+         else if (i >= CHIP_NAVI21)
+            gpu_info->gfx_level = GFX10_3;
+         else if (i >= CHIP_NAVI10)
+            gpu_info->gfx_level = GFX10;
+         else if (i >= CHIP_VEGA10)
+            gpu_info->gfx_level = GFX9;
+         else if (i >= CHIP_TONGA)
+            gpu_info->gfx_level = GFX8;
+         else if (i >= CHIP_BONAIRE)
+            gpu_info->gfx_level = GFX7;
+         else
+            gpu_info->gfx_level = GFX6;
+      }
+   }
+
+   if (gpu_info->family == CHIP_UNKNOWN)
+      return false;
+
+   gpu_info->pci_id = pci_ids[gpu_info->family].pci_id;
+   gpu_info->max_se = pci_ids[gpu_info->family].has_dedicated_vram ? 4 : 1;
+   gpu_info->num_se = gpu_info->max_se;
+   if (gpu_info->gfx_level >= GFX10_3)
+      gpu_info->max_waves_per_simd = 16;
+   else if (gpu_info->gfx_level >= GFX10)
+      gpu_info->max_waves_per_simd = 20;
+   else if (gpu_info->family >= CHIP_POLARIS10 && gpu_info->family <= CHIP_VEGAM)
+      gpu_info->max_waves_per_simd = 8;
+   else
+      gpu_info->max_waves_per_simd = 10;
+
+   if (gpu_info->gfx_level >= GFX10)
+      gpu_info->num_physical_sgprs_per_simd = 128 * gpu_info->max_waves_per_simd;
+   else if (gpu_info->gfx_level >= GFX8)
+      gpu_info->num_physical_sgprs_per_simd = 800;
+   else
+      gpu_info->num_physical_sgprs_per_simd = 512;
+
+   gpu_info->has_timeline_syncobj = true;
+   gpu_info->has_vm_always_valid = true;
+   gpu_info->has_3d_cube_border_color_mipmap = true;
+   gpu_info->has_image_opcodes = true;
+   gpu_info->has_attr_ring = gpu_info->gfx_level >= GFX11;
+   gpu_info->has_attr_ring_wait_bug = gpu_info->gfx_level == GFX11 || gpu_info->gfx_level == GFX11_5;
+   gpu_info->has_ngg_fully_culled_bug = gpu_info->gfx_level == GFX10;
+   gpu_info->has_ngg_passthru_no_msg = gpu_info->family >= CHIP_NAVI23;
+
+   if (gpu_info->family == CHIP_NAVI31 || gpu_info->family == CHIP_NAVI32 || gpu_info->gfx_level >= GFX12)
+      gpu_info->num_physical_wave64_vgprs_per_simd = 768;
+   else if (gpu_info->gfx_level >= GFX10)
+      gpu_info->num_physical_wave64_vgprs_per_simd = 512;
+   else
+      gpu_info->num_physical_wave64_vgprs_per_simd = 256;
+   gpu_info->num_simd_per_compute_unit = gpu_info->gfx_level >= GFX10 ? 2 : 4;
+   gpu_info->lds_size_per_workgroup = gpu_info->gfx_level >= GFX7 ? 64 * 1024 : 32 * 1024;
+   gpu_info->max_render_backends = pci_ids[gpu_info->family].num_render_backends;
+
+   gpu_info->has_dedicated_vram = pci_ids[gpu_info->family].has_dedicated_vram;
+   gpu_info->has_packed_math_16bit = gpu_info->gfx_level >= GFX9;
+
+   gpu_info->has_image_load_dcc_bug = gpu_info->family == CHIP_NAVI23 || gpu_info->family == CHIP_VANGOGH;
+
+   gpu_info->has_distributed_tess =
+      gpu_info->gfx_level >= GFX10 || (gpu_info->gfx_level >= GFX8 && gpu_info->max_se >= 2);
+
+   gpu_info->has_accelerated_dot_product =
+      gpu_info->family == CHIP_VEGA20 ||
+      (gpu_info->family >= CHIP_MI100 && gpu_info->family != CHIP_NAVI10 && gpu_info->family != CHIP_GFX1013);
+
+   gpu_info->has_image_bvh_intersect_ray = gpu_info->gfx_level >= GFX10_3 || gpu_info->family == CHIP_GFX1013;
+
+   gpu_info->address32_hi = gpu_info->gfx_level >= GFX9 ? 0xffff8000u : 0x0;
+
+   gpu_info->has_rbplus = gpu_info->family == CHIP_STONEY || gpu_info->gfx_level >= GFX9;
+   gpu_info->rbplus_allowed =
+      gpu_info->has_rbplus &&
+      (gpu_info->family == CHIP_STONEY || gpu_info->family == CHIP_VEGA12 || gpu_info->family == CHIP_RAVEN ||
+       gpu_info->family == CHIP_RAVEN2 || gpu_info->family == CHIP_RENOIR || gpu_info->gfx_level >= GFX10_3);
+
+   gpu_info->has_gang_submit = true;
+   gpu_info->mesh_fast_launch_2 = gpu_info->gfx_level >= GFX11;
+   gpu_info->hs_offchip_workgroup_dw_size = gpu_info->family == CHIP_HAWAII ? 4096 : 8192;
+   gpu_info->has_ls_vgpr_init_bug = gpu_info->family == CHIP_VEGA10 || gpu_info->family == CHIP_RAVEN;
+   gpu_info->has_graphics = true;
+   gpu_info->ip[AMD_IP_GFX].num_queues = 1;
+
+   gpu_info->gart_page_size = 4096;
+   gpu_info->family_overridden = true;
+
+   return true;
+}
diff --git a/src/amd/common/ac_null_device.h b/src/amd/common/ac_null_device.h
new file mode 100644
index 00000000000..f7842d728a5
--- /dev/null
+++ b/src/amd/common/ac_null_device.h
@@ -0,0 +1,59 @@
+/*
+ * Copyright © 2025 Valve Corporation
+ *
+ * SPDX-License-Identifier: MIT
+ */
+
+#ifndef RADV_NULL_DEVICE_H
+#define RADV_NULL_DEVICE_H
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "amd_family.h"
+
+/* Hardcode some GPU info that are needed for the driver or for some tools. */
+static const struct {
+   uint32_t pci_id;
+   uint32_t num_render_backends;
+   bool has_dedicated_vram;
+} pci_ids[] = {
+   /* clang-format off */
+   [CHIP_TAHITI] = {0x6780, 8, true},
+   [CHIP_PITCAIRN] = {0x6800, 8, true},
+   [CHIP_VERDE] = {0x6820, 4, true},
+   [CHIP_OLAND] = {0x6060, 2, true},
+   [CHIP_HAINAN] = {0x6660, 2, true},
+   [CHIP_BONAIRE] = {0x6640, 4, true},
+   [CHIP_KAVERI] = {0x1304, 2, false},
+   [CHIP_KABINI] = {0x9830, 2, false},
+   [CHIP_HAWAII] = {0x67A0, 16, true},
+   [CHIP_TONGA] = {0x6920, 8, true},
+   [CHIP_ICELAND] = {0x6900, 2, true},
+   [CHIP_CARRIZO] = {0x9870, 2, false},
+   [CHIP_FIJI] = {0x7300, 16, true},
+   [CHIP_STONEY] = {0x98E4, 2, false},
+   [CHIP_POLARIS10] = {0x67C0, 8, true},
+   [CHIP_POLARIS11] = {0x67E0, 4, true},
+   [CHIP_POLARIS12] = {0x6980, 4, true},
+   [CHIP_VEGAM] = {0x694C, 4, true},
+   [CHIP_VEGA10] = {0x6860, 16, true},
+   [CHIP_VEGA12] = {0x69A0, 8, true},
+   [CHIP_VEGA20] = {0x66A0, 16, true},
+   [CHIP_RAVEN] = {0x15DD, 2, false},
+   [CHIP_RENOIR] = {0x1636, 2, false},
+   [CHIP_MI100] = {0x738C, 2, true},
+   [CHIP_NAVI10] = {0x7310, 16, true},
+   [CHIP_NAVI12] = {0x7360, 8, true},
+   [CHIP_NAVI14] = {0x7340, 8, true},
+   [CHIP_NAVI21] = {0x73A0, 16, true},
+   [CHIP_VANGOGH] = {0x163F, 8, false},
+   [CHIP_NAVI22] = {0x73C0, 8, true},
+   [CHIP_NAVI23] = {0x73E0, 8, true},
+   [CHIP_NAVI31] = {0x744C, 24, true},
+   [CHIP_GFX1201] = {0x7550, 16, true},
+   /* clang-format on */
+};
+
+bool ac_null_device_create(struct radeon_info *gpu_info, const char *family);
+
+#endif /* RADV_NULL_DEVICE_H */
diff --git a/src/amd/common/meson.build b/src/amd/common/meson.build
index de5e8ba7af7..3d6cee5df02 100644
--- a/src/amd/common/meson.build
+++ b/src/amd/common/meson.build
@@ -76,6 +76,8 @@ amd_common_files = files(
   'ac_formats.c',
   'ac_formats.h',
   'ac_linux_drm.h',
+  'ac_null_device.c',
+  'ac_null_device.h',
   'ac_shadowed_regs.c',
   'ac_shadowed_regs.h',
   'ac_spm.c',
diff --git a/src/amd/vulkan/radv_physical_device.c b/src/amd/vulkan/radv_physical_device.c
index cb5416e2e3c..bed03974903 100644
--- a/src/amd/vulkan/radv_physical_device.c
+++ b/src/amd/vulkan/radv_physical_device.c
@@ -9,6 +9,9 @@
  */
 
 #include <fcntl.h>
+#include "util/os_misc.h"
+#include "vulkan/vulkan_core.h"
+#include "vk_sync_dummy.h"
 
 #ifdef MAJOR_IN_SYSMACROS
 #include <sys/sysmacros.h>
@@ -17,6 +20,7 @@
 #include "vk_log.h"
 #include "vk_shader_module.h"
 
+#include "common/ac_null_device.h"
 #include "util/disk_cache.h"
 #include "util/hex.h"
 #include "util/u_debug.h"
@@ -37,7 +41,6 @@ typedef void *drmDevicePtr;
 #include "util/os_drm.h"
 #include "winsys/amdgpu/radv_amdgpu_winsys_public.h"
 #endif
-#include "winsys/null/radv_null_winsys_public.h"
 #include "git_sha1.h"
 
 #if AMD_LLVM_AVAILABLE
@@ -2206,6 +2209,23 @@ radv_is_gpu_supported(const struct radeon_info *info)
    return true;
 }
 
+static const struct vk_sync_type *const dummy_types[2] = {
+   &vk_sync_dummy_type,
+   NULL,
+};
+
+static VkResult
+radv_create_null_device(struct radv_instance *instance, struct radv_physical_device *pdev)
+{
+   const char *family = os_get_option("RADV_FORCE_FAMILY");
+
+   if (!ac_null_device_create(&pdev->info, family))
+      return vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED, "Unknown family: %s\n", family);
+
+   pdev->vk.supported_sync_types = dummy_types;
+   return VK_SUCCESS;
+}
+
 static VkResult
 radv_physical_device_try_create(struct radv_instance *instance, drmDevicePtr drm_device,
                                 struct radv_physical_device **pdev_out)
@@ -2280,67 +2300,64 @@ radv_physical_device_try_create(struct radv_instance *instance, drmDevicePtr drm
    }
 
 #ifdef _WIN32
-   pdev->ws = radv_null_winsys_create();
-   if (!pdev->ws)
-      result = VK_ERROR_OUT_OF_HOST_MEMORY;
+   result = radv_create_null_device(instance, pdev);
+   if (result != VK_SUCCESS)
+      goto fail_base;
 #else
    if (drm_device) {
       bool reserve_vmid = instance->vk.trace_mode & RADV_TRACE_MODE_RGP;
 
       result = radv_amdgpu_winsys_create(fd, instance->debug_flags, instance->perftest_flags, reserve_vmid, is_virtio,
                                          &pdev->ws);
-   } else {
-      pdev->ws = radv_null_winsys_create();
-      if (!pdev->ws)
-         result = VK_ERROR_OUT_OF_HOST_MEMORY;
-   }
-#endif
 
-   if (result != VK_SUCCESS) {
-      result = vk_errorf(instance, result, "failed to initialize winsys");
-      goto fail_base;
-   }
-
-   pdev->vk.supported_sync_types = pdev->ws->get_sync_types(pdev->ws);
+      if (result != VK_SUCCESS) {
+         result = vk_errorf(instance, result, "failed to initialize winsys");
+         goto fail_base;
+      }
 
-#ifndef _WIN32
-   if (drm_device && instance->vk.enabled_extensions.KHR_display) {
-      master_fd = open(drm_device->nodes[DRM_NODE_PRIMARY], O_RDWR | O_CLOEXEC);
-      if (master_fd >= 0) {
-         uint32_t accel_working = 0;
-         struct drm_amdgpu_info request = {.return_pointer = (uintptr_t)&accel_working,
-                                           .return_size = sizeof(accel_working),
-                                           .query = AMDGPU_INFO_ACCEL_WORKING};
-
-         if (drm_ioctl_write(master_fd, DRM_AMDGPU_INFO, &request, sizeof(struct drm_amdgpu_info)) < 0 ||
-             !accel_working) {
-            close(master_fd);
-            master_fd = -1;
+      pdev->vk.supported_sync_types = pdev->ws->get_sync_types(pdev->ws);
+      pdev->ws->query_info(pdev->ws, &pdev->info);
+
+      if (instance->vk.enabled_extensions.KHR_display) {
+         master_fd = open(drm_device->nodes[DRM_NODE_PRIMARY], O_RDWR | O_CLOEXEC);
+         if (master_fd >= 0) {
+            uint32_t accel_working = 0;
+            struct drm_amdgpu_info request = {.return_pointer = (uintptr_t)&accel_working,
+                                              .return_size = sizeof(accel_working),
+                                              .query = AMDGPU_INFO_ACCEL_WORKING};
+
+            if (drm_ioctl_write(master_fd, DRM_AMDGPU_INFO, &request, sizeof(struct drm_amdgpu_info)) < 0 ||
+                !accel_working) {
+               close(master_fd);
+               master_fd = -1;
+            }
          }
       }
-   }
-#endif
-
-   pdev->master_fd = master_fd;
-   pdev->local_fd = fd;
-   pdev->ws->query_info(pdev->ws, &pdev->info);
-   pdev->info.family_overridden = drm_device == NULL;
 
-   /* Allow all devices on a virtual winsys, otherwise do a basic support check. */
-   if (!radv_is_gpu_supported(&pdev->info) && drm_device) {
-      if (instance->debug_flags & RADV_DEBUG_STARTUP)
-         fprintf(stderr, "radv: info: device '%s' is not supported by RADV.\n", pdev->info.name);
-      result = VK_ERROR_INCOMPATIBLE_DRIVER;
-      goto fail_wsi;
-   }
+      /* Allow all devices on a virtual winsys, otherwise do a basic support check. */
+      if (!radv_is_gpu_supported(&pdev->info)) {
+         if (instance->debug_flags & RADV_DEBUG_STARTUP)
+            fprintf(stderr, "radv: info: device '%s' is not supported by RADV.\n", pdev->info.name);
+         result = VK_ERROR_INCOMPATIBLE_DRIVER;
+         goto fail_wsi;
+      }
 
-   if (drm_device) {
       pdev->addrlib = ac_addrlib_create(&pdev->info, &pdev->info.max_alignment);
       if (!pdev->addrlib) {
          result = VK_ERROR_INITIALIZATION_FAILED;
          goto fail_wsi;
       }
+
+   } else {
+      /* Create NULL device if no DRM device was provided. */
+      result = radv_create_null_device(instance, pdev);
+      if (result != VK_SUCCESS)
+         goto fail_base;
    }
+#endif
+
+   pdev->master_fd = master_fd;
+   pdev->local_fd = fd;
 
    pdev->use_llvm = instance->debug_flags & RADV_DEBUG_LLVM;
 #if !AMD_LLVM_AVAILABLE
@@ -2542,7 +2559,8 @@ fail_perfcounters:
 fail_wsi:
    if (pdev->addrlib)
       ac_addrlib_destroy(pdev->addrlib);
-   pdev->ws->destroy(pdev->ws);
+   if (pdev->ws)
+      pdev->ws->destroy(pdev->ws);
 fail_base:
    vk_physical_device_finish(&pdev->vk);
 fail_alloc:
@@ -2605,7 +2623,8 @@ radv_physical_device_destroy(struct vk_physical_device *vk_device)
    ac_destroy_perfcounters(&pdev->ac_perfcounters);
    if (pdev->addrlib)
       ac_addrlib_destroy(pdev->addrlib);
-   pdev->ws->destroy(pdev->ws);
+   if (pdev->ws)
+      pdev->ws->destroy(pdev->ws);
    disk_cache_destroy(pdev->vk.disk_cache);
    disk_cache_destroy(pdev->disk_cache_meta);
    if (pdev->local_fd != -1)