Skip to content

Fix specialization constant patching on D3D12. #111356

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 13, 2025
Merged

Fix specialization constant patching on D3D12. #111356

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 15 additions & 17 deletions drivers/d3d12/rendering_shader_container_d3d12.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,28 +94,26 @@ uint32_t RenderingDXIL::patch_specialization_constant(
const uint64_t (&p_stages_bit_offsets)[D3D12_BITCODE_OFFSETS_NUM_STAGES],
HashMap<RenderingDeviceCommons::ShaderStage, Vector<uint8_t>> &r_stages_bytecodes,
bool p_is_first_patch) {
uint32_t patch_val = 0;
int64_t patch_val = 0;
switch (p_type) {
case RenderingDeviceCommons::PIPELINE_SPECIALIZATION_CONSTANT_TYPE_INT: {
uint32_t int_value = *((const int *)p_value);
ERR_FAIL_COND_V(int_value & (1 << 31), 0);
patch_val = int_value;
patch_val = *((const int32_t *)p_value);
} break;
case RenderingDeviceCommons::PIPELINE_SPECIALIZATION_CONSTANT_TYPE_BOOL: {
bool bool_value = *((const bool *)p_value);
patch_val = (uint32_t)bool_value;
patch_val = (int32_t)bool_value;
} break;
case RenderingDeviceCommons::PIPELINE_SPECIALIZATION_CONSTANT_TYPE_FLOAT: {
uint32_t int_value = *((const int *)p_value);
ERR_FAIL_COND_V(int_value & (1 << 31), 0);
patch_val = (int_value >> 1);
patch_val = *((const int32_t *)p_value);
} break;
}
// For VBR encoding to encode the number of bits we expect (32), we need to set the MSB unconditionally.
// However, signed VBR moves the MSB to the LSB, so setting the MSB to 1 wouldn't help. Therefore,
// the bit we set to 1 is the one at index 30.
patch_val |= (1 << 30);
patch_val <<= 1; // What signed VBR does.

// Encode to signed VBR.
if (patch_val >= 0) {
patch_val <<= 1;
} else {
patch_val = ((-patch_val) << 1) | 1;
}

auto tamper_bits = [](uint8_t *p_start, uint64_t p_bit_offset, uint64_t p_tb_value) -> uint64_t {
uint64_t original = 0;
Expand Down Expand Up @@ -169,13 +167,13 @@ uint32_t RenderingDXIL::patch_specialization_constant(

Vector<uint8_t> &bytecode = r_stages_bytecodes[(RenderingDeviceCommons::ShaderStage)stage];
#ifdef DEV_ENABLED
uint64_t orig_patch_val = tamper_bits(bytecode.ptrw(), offset, patch_val);
uint64_t orig_patch_val = tamper_bits(bytecode.ptrw(), offset, (uint64_t)patch_val);
// Checking against the value the NIR patch should have set.
DEV_ASSERT(!p_is_first_patch || ((orig_patch_val >> 1) & GODOT_NIR_SC_SENTINEL_MAGIC_MASK) == GODOT_NIR_SC_SENTINEL_MAGIC);
uint64_t readback_patch_val = tamper_bits(bytecode.ptrw(), offset, patch_val);
DEV_ASSERT(readback_patch_val == patch_val);
uint64_t readback_patch_val = tamper_bits(bytecode.ptrw(), offset, (uint64_t)patch_val);
DEV_ASSERT(readback_patch_val == (uint64_t)patch_val);
#else
tamper_bits(bytecode.ptrw(), offset, patch_val);
tamper_bits(bytecode.ptrw(), offset, (uint64_t)patch_val);
#endif

stages_patched_mask |= (1 << stage);
Expand Down