dawn-cmake/test/array/assign_to_storage_var.wgsl....

91 lines
3.7 KiB
HLSL
Raw Normal View History

[numthreads(1, 1, 1)]
void unused_entry_point() {
return;
}
struct tint_padded_array_element {
int el;
};
struct S {
tint_padded_array_element arr[4];
};
transform: Fixes for DecomposeMemoryAccess CloneContext::Replace(T* what, T* with) is bug-prone, as complex transforms may want to clone `what` multiple times, or not at all. In both cases, this will likely result in an ICE as either the replacement will be reachable multiple times, or not at all. The CTS test: webgpu:shader,execution,robust_access:linear_memory:storageClass="storage";storageMode="read_write";access="read";atomic=true;baseType="i32" Was triggering this brokenness with DecomposeMemoryAccess's use of CloneContext::Replace(T*, T*). Switch the usage of CloneContext::Replace(T*, T*) to the new function form. As std::function is copyable, it cannot hold a captured std::unique_ptr. This prevented the Replace() lambdas from capturing the necessary `BufferAccess` data, as this held a `std::unique_ptr<Offset>`. To fix this, use a `BlockAllocator` for Offsets, and use raw pointers instead. Because the function passed to Replace() is called just before the node is cloned, insertion of new functions will occur just before the currently evaluated module-scope entity. This allows us to remove the "insert_after" arguments to LoadFunc(), StoreFunc(), and AtomicFunc(). We can also kill the icky InsertGlobal() and TypeDeclOf() helpers. Bug: tint:993 Change-Id: I60972bc13a2fa819a163ee2671f61e82d0e68d2a Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58222 Kokoro: Kokoro <noreply+kokoro@google.com> Reviewed-by: James Price <jrprice@google.com> Commit-Queue: Ben Clayton <bclayton@google.com>
2021-07-15 20:29:09 +00:00
static tint_padded_array_element src_private[4] = (tint_padded_array_element[4])0;
groupshared tint_padded_array_element src_workgroup[4];
cbuffer cbuffer_src_uniform : register(b0, space0) {
uint4 src_uniform[4];
};
RWByteAddressBuffer src_storage : register(u1, space0);
RWByteAddressBuffer tint_symbol : register(u2, space0);
RWByteAddressBuffer dst_nested : register(u3, space0);
typedef tint_padded_array_element ret_arr_ret[4];
ret_arr_ret ret_arr() {
const tint_padded_array_element tint_symbol_11[4] = (tint_padded_array_element[4])0;
return tint_symbol_11;
}
transform: Fixes for DecomposeMemoryAccess CloneContext::Replace(T* what, T* with) is bug-prone, as complex transforms may want to clone `what` multiple times, or not at all. In both cases, this will likely result in an ICE as either the replacement will be reachable multiple times, or not at all. The CTS test: webgpu:shader,execution,robust_access:linear_memory:storageClass="storage";storageMode="read_write";access="read";atomic=true;baseType="i32" Was triggering this brokenness with DecomposeMemoryAccess's use of CloneContext::Replace(T*, T*). Switch the usage of CloneContext::Replace(T*, T*) to the new function form. As std::function is copyable, it cannot hold a captured std::unique_ptr. This prevented the Replace() lambdas from capturing the necessary `BufferAccess` data, as this held a `std::unique_ptr<Offset>`. To fix this, use a `BlockAllocator` for Offsets, and use raw pointers instead. Because the function passed to Replace() is called just before the node is cloned, insertion of new functions will occur just before the currently evaluated module-scope entity. This allows us to remove the "insert_after" arguments to LoadFunc(), StoreFunc(), and AtomicFunc(). We can also kill the icky InsertGlobal() and TypeDeclOf() helpers. Bug: tint:993 Change-Id: I60972bc13a2fa819a163ee2671f61e82d0e68d2a Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58222 Kokoro: Kokoro <noreply+kokoro@google.com> Reviewed-by: James Price <jrprice@google.com> Commit-Queue: Ben Clayton <bclayton@google.com>
2021-07-15 20:29:09 +00:00
S ret_struct_arr() {
const S tint_symbol_12 = (S)0;
return tint_symbol_12;
}
transform: Fixes for DecomposeMemoryAccess CloneContext::Replace(T* what, T* with) is bug-prone, as complex transforms may want to clone `what` multiple times, or not at all. In both cases, this will likely result in an ICE as either the replacement will be reachable multiple times, or not at all. The CTS test: webgpu:shader,execution,robust_access:linear_memory:storageClass="storage";storageMode="read_write";access="read";atomic=true;baseType="i32" Was triggering this brokenness with DecomposeMemoryAccess's use of CloneContext::Replace(T*, T*). Switch the usage of CloneContext::Replace(T*, T*) to the new function form. As std::function is copyable, it cannot hold a captured std::unique_ptr. This prevented the Replace() lambdas from capturing the necessary `BufferAccess` data, as this held a `std::unique_ptr<Offset>`. To fix this, use a `BlockAllocator` for Offsets, and use raw pointers instead. Because the function passed to Replace() is called just before the node is cloned, insertion of new functions will occur just before the currently evaluated module-scope entity. This allows us to remove the "insert_after" arguments to LoadFunc(), StoreFunc(), and AtomicFunc(). We can also kill the icky InsertGlobal() and TypeDeclOf() helpers. Bug: tint:993 Change-Id: I60972bc13a2fa819a163ee2671f61e82d0e68d2a Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58222 Kokoro: Kokoro <noreply+kokoro@google.com> Reviewed-by: James Price <jrprice@google.com> Commit-Queue: Ben Clayton <bclayton@google.com>
2021-07-15 20:29:09 +00:00
void tint_symbol_2(RWByteAddressBuffer buffer, uint offset, tint_padded_array_element value[4]) {
buffer.Store((offset + 0u), asuint(value[0u].el));
buffer.Store((offset + 16u), asuint(value[1u].el));
buffer.Store((offset + 32u), asuint(value[2u].el));
buffer.Store((offset + 48u), asuint(value[3u].el));
}
transform: Fixes for DecomposeMemoryAccess CloneContext::Replace(T* what, T* with) is bug-prone, as complex transforms may want to clone `what` multiple times, or not at all. In both cases, this will likely result in an ICE as either the replacement will be reachable multiple times, or not at all. The CTS test: webgpu:shader,execution,robust_access:linear_memory:storageClass="storage";storageMode="read_write";access="read";atomic=true;baseType="i32" Was triggering this brokenness with DecomposeMemoryAccess's use of CloneContext::Replace(T*, T*). Switch the usage of CloneContext::Replace(T*, T*) to the new function form. As std::function is copyable, it cannot hold a captured std::unique_ptr. This prevented the Replace() lambdas from capturing the necessary `BufferAccess` data, as this held a `std::unique_ptr<Offset>`. To fix this, use a `BlockAllocator` for Offsets, and use raw pointers instead. Because the function passed to Replace() is called just before the node is cloned, insertion of new functions will occur just before the currently evaluated module-scope entity. This allows us to remove the "insert_after" arguments to LoadFunc(), StoreFunc(), and AtomicFunc(). We can also kill the icky InsertGlobal() and TypeDeclOf() helpers. Bug: tint:993 Change-Id: I60972bc13a2fa819a163ee2671f61e82d0e68d2a Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58222 Kokoro: Kokoro <noreply+kokoro@google.com> Reviewed-by: James Price <jrprice@google.com> Commit-Queue: Ben Clayton <bclayton@google.com>
2021-07-15 20:29:09 +00:00
typedef tint_padded_array_element tint_symbol_4_ret[4];
tint_symbol_4_ret tint_symbol_4(uint4 buffer[4], uint offset) {
const uint scalar_offset = ((offset + 0u)) / 4;
const uint scalar_offset_1 = ((offset + 16u)) / 4;
const uint scalar_offset_2 = ((offset + 32u)) / 4;
const uint scalar_offset_3 = ((offset + 48u)) / 4;
const tint_padded_array_element tint_symbol_13[4] = {{asint(buffer[scalar_offset / 4][scalar_offset % 4])}, {asint(buffer[scalar_offset_1 / 4][scalar_offset_1 % 4])}, {asint(buffer[scalar_offset_2 / 4][scalar_offset_2 % 4])}, {asint(buffer[scalar_offset_3 / 4][scalar_offset_3 % 4])}};
return tint_symbol_13;
}
typedef tint_padded_array_element tint_symbol_6_ret[4];
tint_symbol_6_ret tint_symbol_6(RWByteAddressBuffer buffer, uint offset) {
const tint_padded_array_element tint_symbol_14[4] = {{asint(buffer.Load((offset + 0u)))}, {asint(buffer.Load((offset + 16u)))}, {asint(buffer.Load((offset + 32u)))}, {asint(buffer.Load((offset + 48u)))}};
return tint_symbol_14;
}
void tint_symbol_8(RWByteAddressBuffer buffer, uint offset, int value[2]) {
buffer.Store((offset + 0u), asuint(value[0u]));
buffer.Store((offset + 4u), asuint(value[1u]));
}
void tint_symbol_9(RWByteAddressBuffer buffer, uint offset, int value[3][2]) {
tint_symbol_8(buffer, (offset + 0u), value[0u]);
tint_symbol_8(buffer, (offset + 8u), value[1u]);
tint_symbol_8(buffer, (offset + 16u), value[2u]);
}
void tint_symbol_10(RWByteAddressBuffer buffer, uint offset, int value[4][3][2]) {
tint_symbol_9(buffer, (offset + 0u), value[0u]);
tint_symbol_9(buffer, (offset + 24u), value[1u]);
tint_symbol_9(buffer, (offset + 48u), value[2u]);
tint_symbol_9(buffer, (offset + 72u), value[3u]);
}
void foo(tint_padded_array_element src_param[4]) {
tint_padded_array_element src_function[4] = (tint_padded_array_element[4])0;
const tint_padded_array_element tint_symbol_15[4] = {{1}, {2}, {3}, {3}};
transform: Fixes for DecomposeMemoryAccess CloneContext::Replace(T* what, T* with) is bug-prone, as complex transforms may want to clone `what` multiple times, or not at all. In both cases, this will likely result in an ICE as either the replacement will be reachable multiple times, or not at all. The CTS test: webgpu:shader,execution,robust_access:linear_memory:storageClass="storage";storageMode="read_write";access="read";atomic=true;baseType="i32" Was triggering this brokenness with DecomposeMemoryAccess's use of CloneContext::Replace(T*, T*). Switch the usage of CloneContext::Replace(T*, T*) to the new function form. As std::function is copyable, it cannot hold a captured std::unique_ptr. This prevented the Replace() lambdas from capturing the necessary `BufferAccess` data, as this held a `std::unique_ptr<Offset>`. To fix this, use a `BlockAllocator` for Offsets, and use raw pointers instead. Because the function passed to Replace() is called just before the node is cloned, insertion of new functions will occur just before the currently evaluated module-scope entity. This allows us to remove the "insert_after" arguments to LoadFunc(), StoreFunc(), and AtomicFunc(). We can also kill the icky InsertGlobal() and TypeDeclOf() helpers. Bug: tint:993 Change-Id: I60972bc13a2fa819a163ee2671f61e82d0e68d2a Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58222 Kokoro: Kokoro <noreply+kokoro@google.com> Reviewed-by: James Price <jrprice@google.com> Commit-Queue: Ben Clayton <bclayton@google.com>
2021-07-15 20:29:09 +00:00
tint_symbol_2(tint_symbol, 0u, tint_symbol_15);
tint_symbol_2(tint_symbol, 0u, src_param);
tint_symbol_2(tint_symbol, 0u, ret_arr());
const tint_padded_array_element src_let[4] = (tint_padded_array_element[4])0;
transform: Fixes for DecomposeMemoryAccess CloneContext::Replace(T* what, T* with) is bug-prone, as complex transforms may want to clone `what` multiple times, or not at all. In both cases, this will likely result in an ICE as either the replacement will be reachable multiple times, or not at all. The CTS test: webgpu:shader,execution,robust_access:linear_memory:storageClass="storage";storageMode="read_write";access="read";atomic=true;baseType="i32" Was triggering this brokenness with DecomposeMemoryAccess's use of CloneContext::Replace(T*, T*). Switch the usage of CloneContext::Replace(T*, T*) to the new function form. As std::function is copyable, it cannot hold a captured std::unique_ptr. This prevented the Replace() lambdas from capturing the necessary `BufferAccess` data, as this held a `std::unique_ptr<Offset>`. To fix this, use a `BlockAllocator` for Offsets, and use raw pointers instead. Because the function passed to Replace() is called just before the node is cloned, insertion of new functions will occur just before the currently evaluated module-scope entity. This allows us to remove the "insert_after" arguments to LoadFunc(), StoreFunc(), and AtomicFunc(). We can also kill the icky InsertGlobal() and TypeDeclOf() helpers. Bug: tint:993 Change-Id: I60972bc13a2fa819a163ee2671f61e82d0e68d2a Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58222 Kokoro: Kokoro <noreply+kokoro@google.com> Reviewed-by: James Price <jrprice@google.com> Commit-Queue: Ben Clayton <bclayton@google.com>
2021-07-15 20:29:09 +00:00
tint_symbol_2(tint_symbol, 0u, src_let);
tint_symbol_2(tint_symbol, 0u, src_function);
tint_symbol_2(tint_symbol, 0u, src_private);
tint_symbol_2(tint_symbol, 0u, src_workgroup);
tint_symbol_2(tint_symbol, 0u, ret_struct_arr().arr);
tint_symbol_2(tint_symbol, 0u, tint_symbol_4(src_uniform, 0u));
tint_symbol_2(tint_symbol, 0u, tint_symbol_6(src_storage, 0u));
int src_nested[4][3][2] = (int[4][3][2])0;
tint_symbol_10(dst_nested, 0u, src_nested);
}