radv: introduce RADV_EXPERIMENTAL envvar for experimental features

Abusing RADV_PERFTEST for experimental features doesn't make real
sense, and I think we should stop doing that.

The existing RADV_PERFTEST options like RADV_PERFTEST=transfer_queue
still exists but they are marked as deprecated, they will be removed
in future Mesa releases.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40646>

.gitlab/issue_templates/Bug Report - AMD Radeon Vulkan.md

@@ -72,4 +72,4 @@ Consider recording a [GFXReconstruct](https://github.com/LunarG/gfxreconstruct/b
 
 Does the issue reproduce with the LLVM backend (`RADV_DEBUG=llvm`) or on the AMDGPU-PRO drivers?
 
-Does your environment set any of the variables `ACO_DEBUG`, `RADV_DEBUG`, and `RADV_PERFTEST`?
+Does your environment set any of the variables `ACO_DEBUG`, `RADV_DEBUG`, `RADV_PERFTEST` and `RADV_EXPERIMENTAL`?

docs/envvars.rst

@@ -1557,21 +1557,14 @@ RADV driver environment variables
 
    a comma-separated list of named flags, which do various things:
 
-   ``bfloat16``
-      enable bfloat16 cooperative matrix support on GFX11-11.5
    ``cswave32``
       enable wave32 for compute shaders (GFX10+)
    ``dccmsaa``
       enable DCC for MSAA images
    ``dmashaders``
       upload shaders to invisible VRAM (might be useful for non-resizable BAR systems)
-   ``emulate_rt``
-      forces ray-tracing to be emulated in software on GFX10_3+ and enables
-      rt extensions with older hardware.
    ``gewave32``
       enable wave32 for vertex/tess/geometry shaders (GFX10+)
-   ``hic``
-      enable experimental implementation of VK_EXT_host_image_copy (GFX10+)
    ``localbos``
       enable local BOs
    ``lowlatencydec``
@@ -1594,6 +1587,22 @@ RADV driver environment variables
       enable wave64 for ray tracing shaders (GFX10+)
    ``sam``
       enable optimizations to move more driver internal objects to VRAM.
+
+   Note that bfloat16, emulate_rt, hic, sparse, transfer_queue, video_decode
+   and video_encode are deprecated and RADV_EXPERIMENTAL should be
+   used instead.
+
+.. envvar:: RADV_EXPERIMENTAL
+
+   a comma-separated list of named flags, which do various things:
+
+   ``bfloat16``
+      enable bfloat16 cooperative matrix support on GFX11-11.5
+   ``emulate_rt``
+      forces ray-tracing to be emulated in software on GFX10_3+ and enables
+      rt extensions with older hardware.
+   ``hic``
+      enable experimental implementation of VK_EXT_host_image_copy (GFX10+)
    ``sparse``
       enable experimental sparse binding and sparse residency on GPUs where we don't support it by default (pre Polaris)
    ``transfer_queue``

src/amd/ci/gitlab-ci.yml

@@ -201,7 +201,7 @@ radv-polaris10-vkcts-full:
     CI_TRON_TIMEOUT__BOOT_CYCLE__RETRIES: 4
     CI_TRON_TIMEOUT__BOOT_CYCLE__MINUTES: 70
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 90
-    RADV_PERFTEST: video_decode,video_encode
+    RADV_EXPERIMENTAL: video_decode,video_encode
     radv_enable_float16_gfx8: true
 
 radv-vega10-vkcts-full:
@@ -212,7 +212,7 @@ radv-vega10-vkcts-full:
   timeout: 1h 15m
   variables:
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 70
-    RADV_PERFTEST: transfer_queue,video_decode,video_encode
+    RADV_EXPERIMENTAL: transfer_queue,video_decode,video_encode
 
 radv-renoir-vkcts-full:
   extends:
@@ -222,7 +222,7 @@ radv-renoir-vkcts-full:
   timeout: 2h 15m
   variables:
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 130
-    RADV_PERFTEST: transfer_queue
+    RADV_EXPERIMENTAL: transfer_queue
     FDO_CI_CONCURRENT: 12  # HACK: Reduce the memory pressure
 
 radv-navi10-vkcts-full:
@@ -232,7 +232,7 @@ radv-navi10-vkcts-full:
     - .navi10-test-valve-kws
     - .radv-valve-manual-rules
   variables:
-    RADV_PERFTEST: transfer_queue,hic
+    RADV_EXPERIMENTAL: transfer_queue,hic
 
 radv-navi21-vkcts:
   timeout: 35m
@@ -243,7 +243,7 @@ radv-navi21-vkcts:
     - .radv-valve-rules
   variables:
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 30
-    RADV_PERFTEST: transfer_queue,hic
+    RADV_EXPERIMENTAL: transfer_queue,hic
 
 radv-navi21-vkcts-full:
   timeout: 50m
@@ -253,7 +253,7 @@ radv-navi21-vkcts-full:
     - .radv-valve-manual-rules
   variables:
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 45
-    RADV_PERFTEST: transfer_queue,hic
+    RADV_EXPERIMENTAL: transfer_queue,hic
 
 radv-vangogh-vkcts-full:
   parallel: 2
@@ -263,7 +263,7 @@ radv-vangogh-vkcts-full:
     - .radv-valve-manual-rules
   timeout: 90m
   variables:
-    RADV_PERFTEST: transfer_queue,hic
+    RADV_EXPERIMENTAL: transfer_queue,hic
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 89
 
 radeonsi-vangogh-glcts:
@@ -297,7 +297,7 @@ radv-raphael-vkcts-full:
   timeout: 45m
   variables:
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 40
-    RADV_PERFTEST: transfer_queue,hic
+    RADV_EXPERIMENTAL: transfer_queue,hic
 
 radv-navi31-vkcts:
   timeout: 25m
@@ -308,7 +308,7 @@ radv-navi31-vkcts:
     - .radv-valve-rules
   variables:
     RADV_DEBUG: nomeshshader # Disable mesh shaders until task shaders stop hanging
-    RADV_PERFTEST: transfer_queue,hic
+    RADV_EXPERIMENTAL: transfer_queue,hic
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 20
 
 radv-navi31-vkcts-full:
@@ -318,7 +318,7 @@ radv-navi31-vkcts-full:
     - .navi31-test-valve
     - .radv-valve-manual-rules
   variables:
-    RADV_PERFTEST: transfer_queue,hic
+    RADV_EXPERIMENTAL: transfer_queue,hic
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 25
 
 radv-gfx1201-vkcts:
@@ -331,7 +331,7 @@ radv-gfx1201-vkcts:
   variables:
     # TODO: Verify that task shaders don't hang.
     RADV_DEBUG: nomeshshader
-    RADV_PERFTEST: transfer_queue,hic
+    RADV_EXPERIMENTAL: transfer_queue,hic
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 20
 
 radv-gfx1201-vkcts-full:
@@ -341,7 +341,7 @@ radv-gfx1201-vkcts-full:
     - .gfx1201-test-valve
     - .radv-valve-manual-rules
   variables:
-    RADV_PERFTEST: transfer_queue,hic
+    RADV_EXPERIMENTAL: transfer_queue,hic
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 30
 
 radv-gfx1201-fluster:

src/amd/ci/radv-polaris10-fails.txt

@@ -1,4 +1,4 @@
-# RADV_PERFTEST=video_decode,video_encode expected failures.
+# RADV_EXPERIMENTAL=video_decode,video_encode expected failures.
 dEQP-VK.video.synchronization.decode_h264.basic.event.device_set_reset,Fail
 dEQP-VK.video.synchronization.decode_h265.basic.event.device_set_reset,Fail
 dEQP-VK.video.synchronization2.decode_h264.basic.event.device_set_reset,Fail

src/amd/ci/radv-renoir-fails.txt

@@ -1,4 +1,4 @@
-# RADV_PERFTEST=transfer_queue+vk_require_etc2=true regressions since 3d803d7a2e0e7d46c04381896168c344d59d88e4
+# RADV_EXPERIMENTAL=transfer_queue+vk_require_etc2=true regressions since 3d803d7a2e0e7d46c04381896168c344d59d88e4
 dEQP-VK.api.copy_and_blit.copy_commands2.image_to_buffer_transfer_queue.2d_images.mip_copies_eac_r11_snorm_block_64x192,Fail
 dEQP-VK.api.copy_and_blit.copy_commands2.image_to_buffer_transfer_queue.2d_images.mip_copies_eac_r11_snorm_block_64x192_2_layers,Fail
 dEQP-VK.api.copy_and_blit.copy_commands2.image_to_buffer_transfer_queue.2d_images.mip_copies_eac_r11_snorm_block_64x192_5_layers,Fail

src/amd/ci/radv-vega10-fails.txt

@@ -1,4 +1,4 @@
-# RADV_PERFTEST=video_decode,video_encode expected failures.
+# RADV_EXPERIMENTAL=video_decode,video_encode expected failures.
 dEQP-VK.video.synchronization.decode_h264.basic.event.device_set_reset,Fail
 dEQP-VK.video.synchronization.decode_h265.basic.event.device_set_reset,Fail
 dEQP-VK.video.synchronization2.decode_h264.basic.event.device_set_reset,Fail

src/amd/vulkan/radv_debug.h

@@ -78,6 +78,9 @@ enum {
                              RADV_DEBUG_DUMP_NIR | RADV_DEBUG_DUMP_ASM | RADV_DEBUG_DUMP_BACKEND_IR,
 };
 
+/* emulate_rt, video_decode, transfer_queue, video_encode, hic, sparse and bfloat16 are deprecated,
+ * use RADV_EXPERIMENTAL instead.
+ */
 enum {
    RADV_PERFTEST_LOCAL_BOS = 1u << 0,
    RADV_PERFTEST_DCC_MSAA = 1u << 1,
@@ -103,6 +106,16 @@ enum {
    RADV_PERFTEST_LOWLATENCYENC = 1u << 21,
 };
 
+enum {
+   RADV_EXPERIMENTAL_EMULATE_RT = 1u << 0,
+   RADV_EXPERIMENTAL_VIDEO_DECODE = 1u << 1,
+   RADV_EXPERIMENTAL_TRANSFER_QUEUE = 1u << 2,
+   RADV_EXPERIMENTAL_VIDEO_ENCODE = 1u << 3,
+   RADV_EXPERIMENTAL_HIC = 1u << 4,
+   RADV_EXPERIMENTAL_SPARSE = 1u << 5,
+   RADV_EXPERIMENTAL_BFLOAT16 = 1u << 6,
+};
+
 enum {
    RADV_TRAP_EXCP_MEM_VIOL = 1u << 0,
    RADV_TRAP_EXCP_FLOAT_DIV_BY_ZERO = 1u << 1,

src/amd/vulkan/radv_instance.c

@@ -129,6 +129,17 @@ static const struct debug_control radv_perftest_options[] = {
    {NULL, 0},
 };
 
+static const struct debug_control radv_experimental_options[] = {
+   {"emulate_rt", RADV_EXPERIMENTAL_EMULATE_RT},
+   {"video_decode", RADV_EXPERIMENTAL_VIDEO_DECODE},
+   {"transfer_queue", RADV_EXPERIMENTAL_TRANSFER_QUEUE},
+   {"video_encode", RADV_EXPERIMENTAL_VIDEO_ENCODE},
+   {"hic", RADV_EXPERIMENTAL_HIC},
+   {"sparse", RADV_EXPERIMENTAL_SPARSE},
+   {"bfloat16", RADV_EXPERIMENTAL_BFLOAT16},
+   {NULL, 0},
+};
+
 static const struct debug_control radv_trap_excp_options[] = {
    {"mem_viol", RADV_TRAP_EXCP_MEM_VIOL},
    {"float_div_by_zero", RADV_TRAP_EXCP_FLOAT_DIV_BY_ZERO},
@@ -377,6 +388,27 @@ radv_parse_pstate(const char *str)
    }
 }
 
+static void
+radv_convert_perftest_to_experimental(struct radv_instance *instance)
+{
+#define CONVERT(name, flag)                                                                                            \
+   if (instance->perftest_flags & RADV_PERFTEST_##flag) {                                                              \
+      fprintf(stderr, "radv: RADV_PERFTEST=" #name " is deprecated and will be removed in future Mesa releases. "      \
+                      "Please use RADV_EXPERIMENTAL=" #name " instead.\n");                                            \
+      instance->experimental_flags |= RADV_EXPERIMENTAL_##flag;                                                        \
+   }
+
+   CONVERT(emulate_rt, EMULATE_RT);
+   CONVERT(video_decode, VIDEO_DECODE);
+   CONVERT(video_encode, VIDEO_ENCODE);
+   CONVERT(transfer_queue, TRANSFER_QUEUE);
+   CONVERT(hic, HIC);
+   CONVERT(sparse, SPARSE);
+   CONVERT(bfloat16, BFLOAT16);
+
+#undef CONVERT
+}
+
 VKAPI_ATTR VkResult VKAPI_CALL
 radv_CreateInstance(const VkInstanceCreateInfo *pCreateInfo, const VkAllocationCallbacks *pAllocator,
                     VkInstance *pInstance)
@@ -414,6 +446,7 @@ radv_CreateInstance(const VkInstanceCreateInfo *pCreateInfo, const VkAllocationC
 
    instance->debug_flags = parse_debug_string(os_get_option("RADV_DEBUG"), radv_debug_options);
    instance->perftest_flags = parse_debug_string(os_get_option("RADV_PERFTEST"), radv_perftest_options);
+   instance->experimental_flags = parse_debug_string(os_get_option("RADV_EXPERIMENTAL"), radv_experimental_options);
    instance->trap_excp_flags = parse_debug_string(os_get_option("RADV_TRAP_HANDLER_EXCP"), radv_trap_excp_options);
    instance->profile_pstate = radv_parse_pstate(debug_get_option("RADV_PROFILE_PSTATE", "peak"));
 
@@ -452,6 +485,7 @@ radv_CreateInstance(const VkInstanceCreateInfo *pCreateInfo, const VkAllocationC
 
    VG(VALGRIND_CREATE_MEMPOOL(instance, 0, false));
 
+   radv_convert_perftest_to_experimental(instance);
    radv_init_dri_options(instance);
 
    *pInstance = radv_instance_to_handle(instance);

src/amd/vulkan/radv_instance.h

@@ -99,6 +99,7 @@ struct radv_instance {
 
    uint64_t debug_flags;
    uint64_t perftest_flags;
+   uint64_t experimental_flags;
    uint64_t trap_excp_flags;
    enum radeon_ctx_pstate profile_pstate;
 

src/amd/vulkan/radv_physical_device.c

@@ -89,8 +89,7 @@ radv_sparse_enabled(const struct radv_physical_device *pdev)
 {
    const struct radv_instance *instance = radv_physical_device_instance(pdev);
 
-   return pdev->info.has_sparse ||
-          (instance->perftest_flags & RADV_PERFTEST_SPARSE);
+   return pdev->info.has_sparse || (instance->experimental_flags & RADV_EXPERIMENTAL_SPARSE);
 }
 
 bool
@@ -100,7 +99,7 @@ radv_transfer_queue_enabled(const struct radv_physical_device *pdev)
 
    /* Check if the GPU has SDMA support and transfer queues are allowed. */
    if (pdev->info.sdma_ip_version == SDMA_UNKNOWN || !pdev->info.ip[AMD_IP_SDMA].num_queues ||
-       !(instance->perftest_flags & RADV_PERFTEST_TRANSFER_QUEUE))
+       !(instance->experimental_flags & RADV_EXPERIMENTAL_TRANSFER_QUEUE))
       return false;
 
    if (!pdev->info.has_gang_submit || !radv_compute_queue_enabled(pdev))
@@ -187,7 +186,7 @@ radv_bfloat16_enabled(const struct radv_physical_device *pdev)
       return false;
 
    /* GFX11-11.5 has precision issues. */
-   return (instance->perftest_flags & RADV_PERFTEST_BFLOAT16) || pdev->info.gfx_level >= GFX12;
+   return (instance->experimental_flags & RADV_EXPERIMENTAL_BFLOAT16) || pdev->info.gfx_level >= GFX12;
 }
 
 static bool
@@ -206,7 +205,7 @@ bool
 radv_host_image_copy_enabled(const struct radv_physical_device *pdev)
 {
    const struct radv_instance *instance = radv_physical_device_instance(pdev);
-   return pdev->info.gfx_level >= GFX10 && (instance->perftest_flags & RADV_PERFTEST_HIC);
+   return pdev->info.gfx_level >= GFX10 && (instance->experimental_flags & RADV_EXPERIMENTAL_HIC);
 }
 
 bool
@@ -225,7 +224,7 @@ bool
 radv_emulate_rt(const struct radv_physical_device *pdev)
 {
    const struct radv_instance *instance = radv_physical_device_instance(pdev);
-   if (instance->perftest_flags & RADV_PERFTEST_EMULATE_RT)
+   if (instance->experimental_flags & RADV_EXPERIMENTAL_EMULATE_RT)
       return true;
 
    /* Do not force emulated RT on GPUs that have native support. */

src/amd/vulkan/radv_video.c

@@ -240,7 +240,7 @@ radv_probe_video_decode(struct radv_physical_device *pdev)
    /* WRITE_MEMORY is needed for SetEvent and is required to pass CTS */
    pdev->video_decode_enabled = radv_video_write_memory_supported(pdev);
 
-   if (instance->perftest_flags & RADV_PERFTEST_VIDEO_DECODE) {
+   if (instance->experimental_flags & RADV_EXPERIMENTAL_VIDEO_DECODE) {
       pdev->video_decode_enabled = true;
    }
 }

src/amd/vulkan/radv_video_enc.c

@@ -79,7 +79,7 @@ radv_probe_video_encode(struct radv_physical_device *pdev)
       return;
    }
 
-   pdev->video_encode_enabled = !!(instance->perftest_flags & RADV_PERFTEST_VIDEO_ENCODE);
+   pdev->video_encode_enabled = !!(instance->experimental_flags & RADV_EXPERIMENTAL_VIDEO_ENCODE);
 }
 
 void

src/gallium/drivers/zink/ci/gitlab-ci.yml

@@ -266,7 +266,7 @@ zink-radv-navi10-valve:
     HWCI_START_WESTON: 1
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 30
     GPU_VERSION: zink-radv-navi10
-    RADV_PERFTEST: hic
+    RADV_EXPERIMENTAL: hic
 
 zink-radv-vangogh-valve:
   timeout: 35m
@@ -283,7 +283,7 @@ zink-radv-vangogh-valve:
     CI_TRON_TIMEOUT__BOOT_CYCLE__MINUTES: 30
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 30
     FDO_CI_CONCURRENT: 6
-    RADV_PERFTEST: hic
+    RADV_EXPERIMENTAL: hic
 
 zink-radv-vangogh-traces:
   extends:
@@ -307,7 +307,7 @@ zink-radv-navi31-valve:
     HWCI_START_WESTON: 1
     GPU_VERSION: zink-radv-navi31
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 10
-    RADV_PERFTEST: hic
+    RADV_EXPERIMENTAL: hic
 
 zink-radv-cezanne:
   extends:
@@ -333,7 +333,7 @@ zink-radv-gfx1201-valve:
     HWCI_START_WESTON: 1
     GPU_VERSION: zink-radv-gfx1201
     CI_TRON_TIMEOUT__OVERALL__MINUTES: 10
-    RADV_PERFTEST: hic
+    RADV_EXPERIMENTAL: hic
 
 zink-radv-gfx1201-traces:
   extends: