Files
NX2navi/runtime/vk_tonemap.c
chrg6 eed3e6b786 Decode the upscaler stand-in's picture with a square, emit SURED with its real opcode and keep the translator's warnings.
The post pass stores the square root of each channel, so the stand-in for Ocarina's untranslatable upscaler now
filters those values and squares them in its compute pass (shaders/fallback_tonemap.spvasm, vk_tonemap.c) instead of
decoding them with the sRGB curve, which darkened mid-tones and lifted the darkest values; the pass is compiled in again,
its round-24 device loss having been the push descriptor call outside the recorder. NX2NAVI_FALLBACK_EXPOSURE is gone.
AmpereLowering emitted SURED as 0xeb48, which matches no Maxwell instruction, so the translator skipped every image
reduction; with 0xeb58 the edge AA's blend-weight pass writes its image. The translator's own warnings go to res.json
notes, and shader dumps are named by program serial so they match the frame dumps' program numbers.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00

150 lines
10 KiB
C

// Decode-and-scale pass standing in for the scene upscaler (vk.h vk_tonemap_record; used by vk_exec_draw.c
// untranslated_dispatch_fallback).
#include "vk.h"
#include <stdlib.h>
#include <string.h>
#include "cfg.h"
#include "fallback_tonemap_spv.h"
#include "host.h"
// The game's upscaler (tensor-core, untranslatable) reads the post pass's picture, whose channels hold the square root of
// their linear value packed as RGB9E5 (prog 102), and writes the linear picture the HUD is drawn over; the frame's last
// pass encodes to sRGB. This pass (shaders/fallback_tonemap.spvasm) scales with linear filtering in the stored values
// and squares them into an RGBA16F image of the target's size, which the caller blits in 1:1. The game composites
// and tonemaps before the post pass, so nothing here changes exposure. (It was a tonemap once, when the input was
// taken for the raw HDR scene; the blit's sRGB decode that stood in since darkened mid-tones and lifted the darkest
// values.)
static struct {
int tried, ok;
PFN_vkCmdPushDescriptorSetKHR push;
VkDescriptorSetLayout dsl;
VkPipelineLayout layout;
VkPipeline pipe;
VkSampler sampler;
struct { VkImage img; VkFormat f; VkImageView view; } views[16];
int nviews;
VkImage out; // the pass's own output (RGBA16F, the target's size); the caller blits it in
VmaAllocation out_alloc;
VkImageView out_view;
uint32_t ow, oh;
} TM;
// The output image at w x h, made again when the size changes (the old one is left to the deferred release).
static int tm_out(VkCommandBuffer cb, uint32_t w, uint32_t h) {
if (TM.out && TM.ow == w && TM.oh == h) return 1;
VkCtx *c = vk_ctx();
if (TM.out) vk_defer_view(TM.out_view), vk_defer_image(TM.out, TM.out_alloc), TM.out = VK_NULL_HANDLE;
VkImageCreateInfo ii = {.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, .imageType = VK_IMAGE_TYPE_2D,
.format = VK_FORMAT_R16G16B16A16_SFLOAT, .extent = {w, h, 1}, .mipLevels = 1, .arrayLayers = 1,
.samples = VK_SAMPLE_COUNT_1_BIT, .tiling = VK_IMAGE_TILING_OPTIMAL,
.usage = VK_IMAGE_USAGE_STORAGE_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT,
.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED};
VmaAllocationCreateInfo ai = {.usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE};
if (vmaCreateImage(c->vma, &ii, &ai, &TM.out, &TM.out_alloc, NULL) != VK_SUCCESS) return TM.out = VK_NULL_HANDLE, 0;
VkImageViewCreateInfo vi = {.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .image = TM.out,
.viewType = VK_IMAGE_VIEW_TYPE_2D, .format = VK_FORMAT_R16G16B16A16_SFLOAT,
.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1}};
if (vkCreateImageView(c->device, &vi, NULL, &TM.out_view) != VK_SUCCESS) {
vmaDestroyImage(c->vma, TM.out, TM.out_alloc);
return TM.out = VK_NULL_HANDLE, 0;
}
VkImageMemoryBarrier2 b = {.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2, .dstStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
.dstAccessMask = VK_ACCESS_2_MEMORY_READ_BIT | VK_ACCESS_2_MEMORY_WRITE_BIT,
.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = TM.out, .subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1}};
VkDependencyInfo d = {.sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO, .imageMemoryBarrierCount = 1, .pImageMemoryBarriers = &b};
vkCmdPipelineBarrier2(cb, &d);
TM.ow = w, TM.oh = h;
return 1;
}
static int tm_init(void) {
if (TM.tried) return TM.ok;
TM.tried = 1;
VkCtx *c = vk_ctx();
TM.push = (PFN_vkCmdPushDescriptorSetKHR)vkGetDeviceProcAddr(c->device, "vkCmdPushDescriptorSetKHR");
VkDescriptorSetLayoutBinding b[2] = {
{.binding = 0, .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, .descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT},
{.binding = 1, .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, .descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT}};
VkDescriptorSetLayoutCreateInfo dci = {.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.flags = VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR,
.bindingCount = 2, .pBindings = b};
VkPushConstantRange pr = {VK_SHADER_STAGE_COMPUTE_BIT, 0, 24};
VkSamplerCreateInfo si = {.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, .magFilter = VK_FILTER_LINEAR,
.minFilter = VK_FILTER_LINEAR, .mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST,
.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE,
.addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE,
.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE, .maxLod = 0};
VkShaderModuleCreateInfo mi = {.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
.codeSize = sizeof fallback_tonemap_spv, .pCode = fallback_tonemap_spv};
VkShaderModule m = VK_NULL_HANDLE;
int ok = TM.push && vkCreateDescriptorSetLayout(c->device, &dci, NULL, &TM.dsl) == VK_SUCCESS;
VkPipelineLayoutCreateInfo li = {.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, .setLayoutCount = 1,
.pSetLayouts = &TM.dsl, .pushConstantRangeCount = 1, .pPushConstantRanges = &pr};
ok = ok && vkCreatePipelineLayout(c->device, &li, NULL, &TM.layout) == VK_SUCCESS &&
vkCreateSampler(c->device, &si, NULL, &TM.sampler) == VK_SUCCESS &&
vkCreateShaderModule(c->device, &mi, NULL, &m) == VK_SUCCESS;
if (ok) {
VkComputePipelineCreateInfo ci = {.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.stage = {.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.stage = VK_SHADER_STAGE_COMPUTE_BIT, .module = m, .pName = "main"},
.layout = TM.layout};
ok = vkCreateComputePipelines(c->device, VK_NULL_HANDLE, 1, &ci, NULL, &TM.pipe) == VK_SUCCESS;
}
if (m) vkDestroyShaderModule(c->device, m, NULL);
rt_log("vk_exec: upscaler stand-in: %s", ok ? "scale and square-decode pass" : "blit (its pass could not be created)");
return TM.ok = ok;
}
// Views of the scratch images the pass samples, kept for the process (the scratch images live as long).
static VkImageView tm_view(VkImage img, VkFormat f) {
for (int i = 0; i < TM.nviews; i++)
if (TM.views[i].img == img && TM.views[i].f == f) return TM.views[i].view;
if (TM.nviews == 16) return VK_NULL_HANDLE;
VkImageViewCreateInfo vi = {.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .image = img, .viewType = VK_IMAGE_VIEW_TYPE_2D,
.format = f, .subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1}};
VkImageView v;
if (vkCreateImageView(vk_ctx()->device, &vi, NULL, &v) != VK_SUCCESS) return VK_NULL_HANDLE;
TM.views[TM.nviews].img = img, TM.views[TM.nviews].f = f, TM.views[TM.nviews].view = v, TM.nviews++;
return v;
}
// Records the pass from src (sampled) into the pass's own RGBA16F image of dw x dh and returns it (the caller blits it
// into the target); VK_NULL_HANDLE when it cannot (the caller blits the source instead).
VkImage vk_tonemap_record(VkCommandBuffer cb, VkImage src, VkFormat sf, uint32_t dw, uint32_t dh) {
static int disp = -1; // NX2NAVI_TM_DISPATCH=0: everything but the dispatch, =2: not at all (diagnosis)
if (disp < 0) disp = rt_knob("TM_DISPATCH", "1");
if (disp == 2 || !tm_init() || !tm_out(cb, dw, dh)) return VK_NULL_HANDLE;
VkImageView sv = tm_view(src, sf), dv = TM.out_view;
if (!sv) return VK_NULL_HANDLE;
VkMemoryBarrier2 mb = {.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .srcStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
.srcAccessMask = VK_ACCESS_2_MEMORY_WRITE_BIT, .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT,
.dstAccessMask = VK_ACCESS_2_SHADER_READ_BIT | VK_ACCESS_2_SHADER_WRITE_BIT};
VkDependencyInfo dep = {.sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO, .memoryBarrierCount = 1, .pMemoryBarriers = &mb};
vkCmdPipelineBarrier2(cb, &dep);
vkCmdBindPipeline(cb, VK_PIPELINE_BIND_POINT_COMPUTE, TM.pipe);
VkDescriptorImageInfo ii[2] = {{TM.sampler, sv, VK_IMAGE_LAYOUT_GENERAL}, {VK_NULL_HANDLE, dv, VK_IMAGE_LAYOUT_GENERAL}};
VkWriteDescriptorSet w[2] = {
{.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, .dstBinding = 0, .descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, .pImageInfo = &ii[0]},
{.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, .dstBinding = 1, .descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, .pImageInfo = &ii[1]}};
rec_push(TM.push, cb, VK_PIPELINE_BIND_POINT_COMPUTE, TM.layout, 0, 2, w); // through the recorder (vk_rec.h), in order
struct { float inv[2], unused, pad; int32_t size[2]; } pc = {{1.0f / (float)dw, 1.0f / (float)dh}, 0, 0,
{(int32_t)dw, (int32_t)dh}};
vkCmdPushConstants(cb, TM.layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof pc, &pc);
if (disp) vkCmdDispatch(cb, (dw + 7) / 8, (dh + 7) / 8, 1);
VkMemoryBarrier2 after = {.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2, .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT,
.srcAccessMask = VK_ACCESS_2_SHADER_WRITE_BIT, .dstStageMask = VK_PIPELINE_STAGE_2_ALL_TRANSFER_BIT,
.dstAccessMask = VK_ACCESS_2_TRANSFER_READ_BIT};
VkDependencyInfo dep2 = {.sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO, .memoryBarrierCount = 1, .pMemoryBarriers = &after};
vkCmdPipelineBarrier2(cb, &dep2);
return TM.out;
}