Merge task/amdfix: 3D BC stand-ins and recorder-ended parts for AMD RDNA2 (#5)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Valiantsin DzerakhandClaude Opus 5.5 committed 2026-10-31 00:39:36 +03:00
commit 8ba0d5ce15
9 files changed
+217 -12

No files matched your search

+1 -1
View File
@@ -82,7 +82,7 @@ Every guest thread is a host thread. TotK assumes Horizon's scheduler: three app
- Vulkan 1.3 back end in place of MacNativeGames' Metal one. At `QueueSubmitCommands` the recorded NVN commands are replayed into Vulkan command buffers (`vk_exec*.c`).
- Memory pools. Apple's unified memory let MacNativeGames use the game's pools zero-copy. The back end imports the pools as host memory (`VK_EXT_external_memory_host`). On a discrete GPU, pools bound as buffers are mirrored in device-local memory and the pages the CPU wrote (write watch, above) are uploaded before a replay; idle pool regions are made read-only so the write-watch queries skip them (`vk_memory.c`). GPU-rendered textures are written back when the CPU reads them: on a pool invalidate, on a CPU copy, and for small targets once the CPU waits for a fence (`vk_texture.c`).
- Replay thread. `TOTK_REPLAY_ASYNC=1` moves the replay off the submitting thread onto a queue at most one frame behind the game; reads of GPU results wait for the newest writer instead of draining the queue (`vk_backend.c`). It is off by default because no timing has shown a gain yet.
- Recorder thread. The Vulkan calls the replay makes into its command buffer are made by a thread of their own (`vk_rec.c`, after Savage Wind Recomp and LAS Recomp): each `vkCmd*` copies its arguments into a packet of a ring and the recorder makes the calls in order; calls into other command buffers are made at once, and the submit waits until the ring is empty. In the town on an RX 9070 the replay takes 17% less time on the submitting thread. `TOTK_VK_RECORDER=0` makes every call on the submitting thread.
- Recorder thread. The Vulkan calls the replay makes into its command buffer are made by a thread of their own (`vk_rec.c`, after Savage Wind Recomp and LAS Recomp): each `vkCmd*` copies its arguments into a packet of a ring and the recorder makes the calls in order; calls into other command buffers are made at once; the recorder also ends the command buffer, and the submit waits until it has. In the town on an RX 9070 the replay takes 17% less time on the submitting thread. `TOTK_VK_RECORDER=0` makes every call on the submitting thread.
- Uniform snapshot. The game rewrites per-object uniform blocks on other threads soon after it submits the frame
that reads them, while the replay still records it. At the start of a submission's replay every uniform range it
binds from CPU-written pools is copied, and the draws bind the copy (`vk_usnap.c`, after MacNativeGames'
+2 -2
View File
@@ -35,7 +35,7 @@ static const char *const known[] = {
"BIND_LSTICK_RIGHT", "BIND_LSTICK_UP", "BIND_MINUS",
"BIND_PLUS", "BIND_R", "BIND_RCLICK", "BIND_RIGHT", "BIND_RSTICK_DOWN", "BIND_RSTICK_LEFT", "BIND_RSTICK_RIGHT", "BIND_RSTICK_UP",
"BIND_SKIP", "BIND_STATS", "BIND_UP", "BIND_X", "BIND_Y",
"BIND_ZL", "BIND_ZR", "CLOTH_REALTIME", "CORE_DIAG", "CRIT", "CRIT_OUT", "DESC_TEMPLATE", "DUMP", "DUMP_BUF", "DUMP_DIR", "DUMP_LOG",
"BIND_ZL", "BIND_ZR", "BLOCK_VIEWS_3D", "CLOTH_REALTIME", "CORE_DIAG", "CRIT", "CRIT_OUT", "DESC_TEMPLATE", "DUMP", "DUMP_BUF", "DUMP_DIR", "DUMP_LOG",
"EXEFS", "FEEDBACK", "FIX_C0", "FIX_ZERO", "FLIGHT", "FLIGHT_AT", "FLIGHT_DIR", "FLIGHT_FRAMES", "FLIGHT_SPIKE_MS",
"FLIGHT_THUMBS", "FPS60", "FPS_LIMIT", "FREEZE_IDLE", "FREE_CORES", "FULLSCREEN", "GPL", "GPL_DRAW_BUILD", "GPU",
"GPU_INDEX", "GPU_NAME", "HIGHFPS_FIX", "INPUT", "INPUT_RECORD", "INPUT_SCRIPT", "LANG", "LIB_PREBUILD",
@@ -43,7 +43,7 @@ static const char *const known[] = {
"MENU_TEST_DIR", "MENU_TEST_SET", "MINIMAL_LAYOUT", "MIRROR_GAP", "MOTION", "MOUSE_SENS", "MOVIE_HOST", "MOVIE_PACE",
"ONE_PER_CORE", "OPERATION_MODE", "OUTPUT_RES", "PAD_LAYOUT", "PATCH_OUTPUT", "PATCH_SCENE", "PERF_OVERLAY", "PERF_SLOW_MS", "PIPELINE_CACHE", "PIPE_MEMO", "PIPE_PREBUILD", "PIPE_SKIP", "PIPE_SYNC", "PIPE_WAIT",
"PREWARM_XLATE", "PROBE_MOVIE", "PROBE_MOVIE_SHOT", "PROFILE", "PROFILE_CHAINS", "PROFILE_DIR", "PROFILE_FRAMES",
"PROFILE_MAIN", "PROFILE_RENDER", "PUSH_SKIP", "PUSH_STAGE", "RDOC", "RDOC_DIR", "RDOC_DLL", "REALTIME", "REEXEC",
"PROFILE_MAIN", "PROFILE_RENDER", "PUSH_SKIP", "PUSH_STAGE", "RDOC", "RDOC_DIR", "RDOC_DLL", "REALTIME", "REC_END", "REEXEC",
"RENDER_SCALE", "REPLAY_ASYNC",
"REPLAY_CHECK", "RESOLVE_CHECK", "ROMFS", "ROMFS_INDEX", "ROMFS_MMAP", "ROMFS_MODS", "ROMFS_PATCHES",
"ROMFS_PROOF_DIR", "ROMFS_VERIFY", "RUMBLE", "RUMBLE_CURVE", "RUMBLE_STRENGTH", "RUMBLE_TRIGGERS", "SHADER_CACHE",
+156
View File
@@ -23,6 +23,7 @@
#include "host.h"
#include "nvn_api.h"
#include "nvn_tex.h"
#include "platform/platform.h"
#include "tex.h"
#include "vk.h"
#include "vk_exec.h"
@@ -141,11 +142,163 @@ static int view_format(const tex_t *t, const view_t *v, int blocks, VkFormat *ou
return 1;
}
// ---- 3D BC render targets
// A 3D BC texture has no block views (vk_texture.c create_tex, TOTK_BLOCK_VIEWS_3D): on RDNA2 cards such an image
// made BLOCK_TEXEL_VIEW_COMPATIBLE and 2D_ARRAY_COMPATIBLE took GPU writes past the end of its memory (issue #5). Draws
// into it through an uncompressed view of its blocks (agl's BC4 encoder fills 32x32x32 and 64x64x64 volumes and their
// levels at the logo) go to a stand-in: a 2D array of the blocks of one level in the view's format, one layer per
// depth plane. It starts with the level's content, so a pass that writes part of a plane keeps the rest, and each pass
// end copies the planes it drew into the level (a copy between a compressed and a size-compatible uncompressed format).
typedef struct { VkImage bc, img; VmaAllocation alloc; VkFormat fmt; uint32_t level, w, h, d; } BcStandIn;
#define BC_STANDINS 64 // one per level drawn: 6 volumes of up to 7 levels at the logo
static BcStandIn bcs[BC_STANDINS];
static int nbcs;
static PlatMutex bcs_lock = PLAT_MUTEX_INIT; // the replay adds and reads; destroy_tex (any thread) removes
static void image_barrier(VkCommandBuffer cb, VkImage img, VkImageLayout from, uint32_t layers) {
VkImageMemoryBarrier2 ib = {.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .srcStageMask = VK_PIPELINE_STAGE_2_NONE,
.dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
.dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT,
.oldLayout = from, .newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = img, .subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, layers}};
VkDependencyInfo di = {.sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO, .imageMemoryBarrierCount = 1, .pImageMemoryBarriers = &ib};
vkCmdPipelineBarrier2(cb, &di);
}
static void memory_barrier(VkCommandBuffer cb, VkPipelineStageFlags2 src, VkAccessFlags2 sa, VkPipelineStageFlags2 dst,
VkAccessFlags2 da) {
VkMemoryBarrier2 mb = {.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .srcStageMask = src, .srcAccessMask = sa,
.dstStageMask = dst, .dstAccessMask = da};
VkDependencyInfo di = {.sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO, .memoryBarrierCount = 1, .pMemoryBarriers = &mb};
vkCmdPipelineBarrier2(cb, &di);
}
// The stand-in of level level of a 3D BC image for draws through view format fmt, made on first use; 0 when none can
// be made (logged). Replay thread.
static int bc_standin(const VkTexImage *img, uint32_t level, VkFormat fmt, BcStandIn *out) {
static int warned;
plat_mutex_lock(&bcs_lock);
int full = nbcs == BC_STANDINS;
for (int i = 0; i < nbcs; i++)
if (bcs[i].bc == img->image && bcs[i].level == level && bcs[i].fmt == fmt) {
*out = bcs[i];
plat_mutex_unlock(&bcs_lock);
return 1;
}
plat_mutex_unlock(&bcs_lock);
if (full) {
if (!warned++) rt_log("vk: more than %d 3D BC render targets: the others are left out", BC_STANDINS);
return 0;
}
uint32_t lw = level_extent(img->width, level), lh = level_extent(img->height, level);
BcStandIn s = {img->image, VK_NULL_HANDLE, NULL, fmt, level, (lw + 3) / 4, (lh + 3) / 4,
level_extent(img->depth, level)};
VkImageCreateInfo ci = {.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, .imageType = VK_IMAGE_TYPE_2D, .format = fmt,
.extent = {s.w, s.h, 1}, .mipLevels = 1, .arrayLayers = s.d, .samples = VK_SAMPLE_COUNT_1_BIT,
.tiling = VK_IMAGE_TILING_OPTIMAL,
.usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
VK_IMAGE_USAGE_TRANSFER_DST_BIT,
.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED};
VmaAllocationCreateInfo ai = {.usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE};
if (vmaCreateImage(vk_ctx()->vma, &ci, &ai, &s.img, &s.alloc, NULL) != VK_SUCCESS) {
if (!warned++) rt_log("vk: stand-in for a 3D BC render target could not be created: left out");
return 0;
}
// In the upload batch, which goes to the queue before this part: GENERAL, then the texture's blocks.
VkCommandBuffer cb = vk_upload_begin();
image_barrier(cb, s.img, VK_IMAGE_LAYOUT_UNDEFINED, s.d);
memory_barrier(cb, VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, VK_ACCESS_2_MEMORY_WRITE_BIT, VK_PIPELINE_STAGE_2_COPY_BIT,
VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT);
VkImageCopy rg = {.srcSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, level, 0, 1},
.dstSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, s.d},
.extent = {lw, lh, s.d}}; // texels of the compressed source
vkCmdCopyImage(cb, img->image, VK_IMAGE_LAYOUT_GENERAL, s.img, VK_IMAGE_LAYOUT_GENERAL, 1, &rg);
memory_barrier(cb, VK_PIPELINE_STAGE_2_COPY_BIT, VK_ACCESS_2_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT);
vk_upload_end();
plat_mutex_lock(&bcs_lock);
bcs[nbcs++] = s;
plat_mutex_unlock(&bcs_lock);
rt_log("vk: 3D BC render target %ux%ux%u level %u drawn through a %ux%u x %u stand-in (format %d)", img->width,
img->height, img->depth, level, s.w, s.h, s.d, (int)fmt);
*out = s;
return 1;
}
// vk_exec_draw.c vk_exec_image_gone: the stand-in of a BC image that goes, freed once the GPU is done with it.
void vk_exec_bc_gone(VkImage image) {
plat_mutex_lock(&bcs_lock);
for (int i = 0; i < nbcs; i++)
if (bcs[i].bc == image) {
vk_defer_image(bcs[i].img, bcs[i].alloc);
bcs[i--] = bcs[--nbcs];
}
plat_mutex_unlock(&bcs_lock);
}
// At a pass end: the planes the pass drew into a stand-in go to its texture.
static void bc_copy_back(VkExec *ex, const VkExecTarget *t) {
plat_mutex_lock(&bcs_lock);
for (int i = 0; i < nbcs; i++) {
if (bcs[i].img != t->image) continue;
memory_barrier(ex->cb, VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, VK_ACCESS_2_MEMORY_WRITE_BIT, VK_PIPELINE_STAGE_2_COPY_BIT,
VK_ACCESS_2_TRANSFER_READ_BIT | VK_ACCESS_2_TRANSFER_WRITE_BIT);
VkImageCopy rg = {.srcSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, 0, t->base_layer, t->layers},
.dstSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, bcs[i].level, 0, 1},
.dstOffset = {0, 0, (int32_t)t->base_layer},
.extent = {bcs[i].w, bcs[i].h, t->layers}}; // texels of the uncompressed source
vkCmdCopyImage(ex->cb, t->image, VK_IMAGE_LAYOUT_GENERAL, bcs[i].bc, VK_IMAGE_LAYOUT_GENERAL, 1, &rg);
memory_barrier(ex->cb, VK_PIPELINE_STAGE_2_COPY_BIT, VK_ACCESS_2_TRANSFER_WRITE_BIT,
VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT, VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT);
ex->recorded++;
break;
}
plat_mutex_unlock(&bcs_lock);
}
// A draw into a 3D BC image without block views through an uncompressed view of its blocks: the target is the
// image's stand-in (bc_standin). 1 when o is filled (target or left out), 0 for any other target.
static int bc_target(const tex_t *t, const view_t *v, int has_v, const VkTexImage *img, VkExecTarget *o) {
int bw, bh;
if (img->type != VK_IMAGE_TYPE_3D || img->block_views || !nvn_fmt_bpb(t->d.format, &bw, &bh) || bw != 4 || bh != 4)
return 0;
VkFormat fmt;
BcStandIn s;
uint32_t level = has_v && v->set_levels ? v->base_level : 0;
if (level >= img->levels) {
warn_once(W_LEVEL, "render target level %u of a %u-level texture: left out", level, img->levels);
return 1;
}
if (!has_v || !v->set_format || !view_format(t, v, 1, &fmt) || !bc_standin(img, level, fmt, &s)) {
warn_once(W_BLOCK_VIEW, "render target format %#x (%ux%u) through view format %#x: no block view, left out",
t->d.format, t->d.width, t->d.height, has_v && v->set_format ? v->format : 0);
return 1;
}
uint32_t base = 0, layers = s.d;
if (v->set_layers) {
base = v->min_layer;
layers = v->nlayers ? v->nlayers : (base < s.d ? s.d - base : 0);
}
if (base >= s.d || !layers) return 1;
if (base + layers > s.d) layers = s.d - base;
o->image = s.img; // not img->image: view_of makes this command buffer's view of the stand-in
o->img = img;
o->format = fmt;
o->view_type = layers > 1 ? VK_IMAGE_VIEW_TYPE_2D_ARRAY : VK_IMAGE_VIEW_TYPE_2D;
o->aspect = VK_IMAGE_ASPECT_COLOR_BIT;
o->level = 0, o->base_layer = base, o->layers = layers;
o->width = s.w, o->height = s.h, o->samples = 1;
o->nvn_format = v->format;
return 1;
}
static void resolve_target(const tex_t *t, const view_t *v, int has_v, int depth_slot, VkExecTarget *o) {
memset(o, 0, sizeof *o);
o->tex = t;
if (!t) return;
const VkTexImage *img = vk_texture_get(t);
if (img && !depth_slot && bc_target(t, v, has_v, img, o)) return; // a 3D BC image: its stand-in
VkImageUsageFlags need = depth_slot ? VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT : VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
int is3d = img && img->type == VK_IMAGE_TYPE_3D && !depth_slot; // slices drawn through 2D views (2D_ARRAY_COMPATIBLE)
if (!img || !(img->usage & need) || (img->type != VK_IMAGE_TYPE_2D && !is3d)) {
@@ -469,6 +622,9 @@ void vk_exec_pass_end(VkExec *ex) {
vkCmdEndRendering(ex->cb);
ex->recorded++;
ex->in_pass = 0;
for (int i = 0; i < ex->pass_ncolor && i < 8; i++) // stand-ins of 3D BC targets go to their textures
if (ex->color[i].image && ex->color[i].img && ex->color[i].img->image != ex->color[i].image)
bc_copy_back(ex, &ex->color[i]);
if (vk_dump_armed) vk_dump_pass_end(ex);
uint64_t stamp = __atomic_load_n(&gfx_counters.frames, __ATOMIC_RELAXED) + 1;
for (int i = 0; i < 8; i++)
+2
View File
@@ -124,6 +124,8 @@ void vk_exec_inline_end(int reopen);
// buffers): the replay continues in a new one (vk_sync_split).
void vk_exec_draw_new_cb(void);
VkBuffer vk_exec_inline_buffer(VkDeviceSize size);
// vk_exec.c: the stand-in a 3D BC render target is drawn through goes with its image (vk_exec_image_gone).
void vk_exec_bc_gone(VkImage image);
// Scissor 0 clipped to the render area of the open pass; extent 0: nothing to draw.
VkRect2D vk_exec_scissor(const VkExec *ex);
void vk_exec_noatt_dims(const VkExec *ex, uint32_t *w, uint32_t *h); // pass size when no attachment is set
+1
View File
@@ -564,6 +564,7 @@ VkImageView vk_exec_target_view(const VkTexImage *img, VkFormat format, VkImageV
}
void vk_exec_image_gone(VkImage image) {
vk_exec_bc_gone(image);
if (!views_by_image.klen) return;
for (ViewNode *n = nvn_hm_del(&views_by_image, &image), *next; n; n = next) {
next = n->next;
+35 -2
View File
@@ -38,7 +38,7 @@ enum {
R_SetDepthWriteEnable, R_SetFrontFace, R_SetPrimitiveRestartEnable, R_SetPrimitiveTopology,
R_SetRasterizerDiscardEnable, R_SetScissorWithCount, R_SetStencilCompareMask, R_SetStencilOp, R_SetStencilReference,
R_SetStencilTestEnable, R_SetStencilWriteMask, R_SetViewportWithCount, R_WriteTimestamp, R_WriteTimestamp2,
R_PushTpl, R_Push, R_VertexInput, R_Bool, R_PolygonMode, R_BlendEnable, R_BlendEquation, R_WriteMask,
R_PushTpl, R_Push, R_VertexInput, R_Bool, R_PolygonMode, R_BlendEnable, R_BlendEquation, R_WriteMask, R_EndCb,
};
static uint64_t now_ns(void) {
@@ -131,7 +131,8 @@ void rec_begin(VkCommandBuffer cb) {
if (mode < 0) {
// Default off on AMD GPUs: with the recorder their drivers lost the device within the first frames or at menu
// opens, 1 boot in 3 on an RX 9070 XT, and TOTK_VK_RECORDER=0 ended it for three users (issue #5); under RADV
// the kernel saw an illegal opcode at the end of a command buffer (issue #40). The cause is not found.
// the kernel saw an illegal opcode at the end of a command buffer (issue #40). Letting the recorder end its own
// parts (rec_end_cb) fixed the white scene on an RX 6700 XT; the default stays off until more AMD cards run it.
const char *set = cfg_get("TOTK_VK_RECORDER");
mode = rt_knob("VK_RECORDER", vk_caps.amd ? "0" : "1");
if (!mode && vk_caps.amd && !(set && *set))
@@ -162,6 +163,32 @@ void rec_end(void) {
rec_live = 0;
}
// TOTK_REC_END (default 1): the thread that recorded the part also ends it, after the breadcrumb crumb (kind 0: none)
// that vk_sync.c writes at the end of a submission. On an RX 6700 XT (driver 26.10.1, issue #5) parts the recorder
// recorded and the submitting thread ended broke when a submission was split (vk_sync.c vk_sync_split): the GPU hung
// or the scene turned green, then white. The handoff in rec_drain orders the two threads, so Vulkan allows the old
// split; the driver did not cope with it. TOTK_REC_END=0 brings the old split back.
typedef struct { int kind; uint64_t n; } A_EndCb;
int rec_end_cb(VkCommandBuffer cb, int crumb, uint64_t n) {
static int mode = -1;
if (mode < 0) {
mode = rt_knob("REC_END", "1");
rt_log("vk_rec: a part the recorder records is ended by %s (TOTK_REC_END=%d)",
mode ? "the recorder" : "the submitting thread", mode);
}
if (!rec_live || cb != rec_cb || !mode) {
rec_end();
if (crumb) vk_crumb(cb, crumb, n);
return 0;
}
A_EndCb *a = put(R_EndCb, sizeof *a);
*a = (A_EndCb){crumb, n};
pub();
rec_drain();
rec_live = 0;
return 1;
}
void rec_stats_log(void) {
if (!(rt_trace & RT_TRACE_PERF) || !RS_.parts) return;
rt_log("vk_rec: %llu of %llu parts recorded by the recorder: %llu packets (%.1f MiB), %llu drains (%llu for a call "
@@ -588,6 +615,12 @@ static void exec_packet(uint32_t op, void *p) {
case R_BlendEnable: { A_Targets *a = p; ((PFN_vkCmdSetColorBlendEnableEXT)a->fn)(cb, a->first, a->n, a->v); break; }
case R_BlendEquation: { A_Targets *a = p; ((PFN_vkCmdSetColorBlendEquationEXT)a->fn)(cb, a->first, a->n, a->v); break; }
case R_WriteMask: { A_Targets *a = p; ((PFN_vkCmdSetColorWriteMaskEXT)a->fn)(cb, a->first, a->n, a->v); break; }
case R_EndCb: {
A_EndCb *a = p;
if (a->kind) vk_crumb(cb, a->kind, a->n);
vkEndCommandBuffer(cb);
break;
}
default: rt_fatal("vk_rec: unknown packet %u", op);
}
}
+5 -1
View File
@@ -4,7 +4,8 @@
// ring, in call order, and the recorder makes the real call in that order; calls into any other command buffer (uploads,
// the window, probes) are made at once. A call the ring has no packet for (a structure with a pNext chain) waits for
// the recorder to empty the ring and is then made at once on the calling thread. rec_end returns once the recorder made
// every call, so ending, submitting and the command pool stay on the submitting thread.
// every call. vk_sync.c ends a part through rec_end_cb: the recorder makes the last calls into a part it recorded,
// vkEndCommandBuffer included; submitting and the command pool stay on the submitting thread.
//
// Every vkCmd* the runtime calls through vk.h goes through the rc_* wrappers below (the macros at the end); the
// extension commands called through function pointers have rec_* wrappers of their own. TOTK_VK_RECORDER=0 makes every
@@ -16,6 +17,9 @@
void rec_begin(VkCommandBuffer cb); // the submitting thread, sub_lock held
void rec_end(void); // waits for the recorder; no-op when rec_begin did not start it
// The end of part cb: the breadcrumb crumb (0: none, vk_crumbs.c), then vkEndCommandBuffer, both made by the recorder
// when it records cb, and the wait for it (1); otherwise rec_end and the breadcrumb, and the caller ends cb (0).
int rec_end_cb(VkCommandBuffer cb, int crumb, uint64_t n);
void rec_drain(void); // waits until the recorder made every call queued so far
void rec_stats_log(void); // TOTK_TRACE=perf: packets, bytes, drains and waits since the last call
+3 -5
View File
@@ -293,9 +293,8 @@ static VkCommandBuffer prologue(VkCtx *c, uint64_t fence) {
VkCommandBuffer vk_sync_split(void) {
VkCtx *c = vk_ctx();
VkCommandBuffer cb = slots[cur_slot].cb;
rec_end(); // every command of the part is in cb
ts_end(cb, cur_fence);
vkEndCommandBuffer(cb);
if (!rec_end_cb(cb, 0, 0)) vkEndCommandBuffer(cb); // every command of the part is in cb, ended by its recorder
vk_upload_flush(0); // the uploads this part asked for precede it in queue order
VkCommandBuffer pro = prologue(c, cur_fence);
VkCommandBufferSubmitInfo cbi[2] = {{.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO, .commandBuffer = pro},
@@ -348,11 +347,10 @@ uint64_t vk_sync_submit(const uint64_t *handles, int n) {
rec_begin(s->cb);
int recorded = vk_exec_record(s->cb, handles, n, fence);
s = &slots[cur_slot]; // the last part (vk_sync_split)
rec_end();
vk_crumb(s->cb, VK_CRUMB_SUBMIT_END, fence); // the recorder is done with the part
if (recorded) ts_end(s->cb, fence);
else ts_drop();
vkEndCommandBuffer(s->cb);
// The recorder is done with the part: the end-of-submission crumb, then the end, by the recorder when it records.
if (!rec_end_cb(s->cb, VK_CRUMB_SUBMIT_END, fence)) vkEndCommandBuffer(s->cb);
perf_replay(t0);
// After the recording: the uploads it asked for (and every earlier one) precede it in queue order.
vk_upload_flush(0);
+12 -1
View File
@@ -328,7 +328,18 @@ static Tex *create_tex(const tex_t *t, const TexKey *k, int defer_init) {
.samples = (VkSampleCountFlagBits)samples, .tiling = VK_IMAGE_TILING_OPTIMAL, .usage = usage,
.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED};
int bw, bh, block_views = 0;
if (!ds && samples == 1 && f.vk == f.ideal && nvn_fmt_bpb(d->format, &bw, &bh) && bw == 4 && bh == 4) {
// TOTK_BLOCK_VIEWS_3D (default 0): a 3D BC image also gets block views. On RDNA2 cards (driver 26.10.1, issue #5)
// the 32x32x32 BC4 volume the game creates at the logo, made BLOCK_TEXEL_VIEW_COMPATIBLE and 2D_ARRAY_COMPATIBLE,
// took a GPU write just past the end of its memory: device lost at the first submit after it when the image ended
// its memory block. Draws into a 3D BC image go through a stand-in instead (vk_exec.c bc_target).
static int bv3d = -1;
if (bv3d < 0) {
bv3d = rt_knob("BLOCK_VIEWS_3D", "0");
rt_log("vk: 3D BC textures %s (TOTK_BLOCK_VIEWS_3D=%d)", bv3d ? "get block views (renderable)" : "have no block views",
bv3d);
}
if (!ds && samples == 1 && f.vk == f.ideal && nvn_fmt_bpb(d->format, &bw, &bh) && bw == 4 && bh == 4 &&
(type != VK_IMAGE_TYPE_3D || bv3d)) {
VkImageCreateInfo bc = ci; // BC formats: renderable through views of their blocks (block_views_ok)
bc.usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
bc.flags |= VK_IMAGE_CREATE_BLOCK_TEXEL_VIEW_COMPATIBLE_BIT | VK_IMAGE_CREATE_EXTENDED_USAGE_BIT;