Updated per comments. Typedefs for intermediate short vectors moved into the bodies of the functions using them.

Uploaded the correct diff.

Changed the type checks to ensure that the input has a valid type, rather than guarding against specific invalid types.

Ping.

Updated the AVX512 isel pattern to account for the instruction record name change.

Updated to support D48067 instead of D45202. Removed IntrinsicsX86.td changes since the lowering is now in InstCombine, changed the masked scalar pattern tests to exclude mask extension that was in the previous versions.

Abandoning this due to D48067 being accepted instead.

Fixed the typo in the test name and added checks to make the transform stop if the rounding mode immediate and/or SAE are not constant.

Changes made per comments. Note that zext IR instructions have been fully excluded from all patterns, which will require altering vec_floor.ll tests in D45203 if this revision is accepted.

Fixed the error in the final check (it was from badly undone edits around there). Moved the early-exit check. Expanded the comment on the AVX512BW check for clarity. Some names changed per comments.

Added tests for floor intrinsics and masked scalar double patterns to cover all introduced isel patterns.

Closing this due to failure of D45723.

Closing this due to failure of D45721.

Corrected the scalar pattern predicates, added packed zero-masked instruction patterns and tests to cover zero-masking. Changed the RUN line of vec_floor.ll to give different results for AVX512F and AVX512VL where needed (e.g. in 128- and 256-bit masked operations).

Changed isSequentialOrUndefInRange to take an increment argument and added isAnyInRange.

Taking over at @GBuella's request. The patterns that are currently implemented will be finished, but I don't have much hope for the masked versions. Since the mask is only good for the lower elements and the upped elements must be zeroed out, lowering the masked versions of these intrinsics would require not simple selects (see PR34877), but patterns like

Added zero extension of mask to i32 in the masked scalar tests and added more ways to represent the mask, testing the 8-bit mask pattern among others. 16-bit mask patterns removed due to scalar_to_vector errors.

Changed the scalar intrinsic lowering to work via extract-insert. D45203 contains tests for folding the resulting IR patterns.

Changed the folding to use isel patterns from D47012.

Added an extra flag to the test, so that both flag-reading paths are covered. Since flags are module-wide, if there need to be tests covering them separately they would have to be put in new files.

Changes made per comments.

Replaced assertion with a more explicit error at the start of the code path. Changed the emission of the "GNU\0" byte sequence to use EmitBytes with the StringRef constructor specifically to include the trailing zero byte.

Removed the llvm-readobj part to release as a separate patch.

Made changes to the readobj part per comments. Will wait on comment from @craig.topper before splitting the patch.

Enabled readobj to parse sections that use 32 bits for the data on 64-bit targets (like gcc currently does) and added explicit alignment to the section generation.

The bug was in the emitLongJmpShadowStackFix and emitEHSjLjLongJmp. The longjmp builtin call is not a terminator, so it's followed by an "unreachable" IR instruction. The previous patch put the register restoration and the indirect branch at the end of the final basic block of the resulting code. It produced correct code on Linux, but on Mac where "unreachable" produces ud2 instructions this lead to a crash because the ud2 was left in place and ended up being before the longjmp code, regardless of whether the shadow stack fix was present. This patch moves the longjmp pseudo-instruction and the ud2 (if present) to the final basic block of the produced code and emits the longjmp logic before them, so the ud2 is correctly placed at the end of longjmp and is properly unreachable.

Updated to use full 8/4 byte integers to store the flags rather than using 1 byte and padding.

Uploaded the version without the test for the section emission. Corrected.

Removed the unused HasIBT variable declaration from X86.h.

Removed the missed FIXME.

Replaced FIXMEs with more appropriate cautionary comments. They do not need to be addressed directly by any patches, so much as they served to warn about using the instructions carelessly now that there is no longer a simple way to restrict their usage.

@craig.topper, please review the latest changes.

Restored shuffle combining tests. Fixed the folding to account for unary (A == B) patterns, added tests.

@RKSimon, please review the latest changes.

Test fix.

Style fix.

Updated per comments.

Changed the shuffle mask emission code to match D45721.

Tidied up the shuffle mask emission and checking and moved the PACKUSDW SSE4.1 check into tracePackVectorShuffle.

@RKSimon what's the motivation for leaving the tests for upgraded intrinsics in place instead of moving them to appropriate files?

Updated per comments.

Style fix.

Updated per comments in the clang part.

Updated per comments.

Fast-isel tests restored and updated.

Style fix.

Updated fast-isel tests for SSE and AVX2. There don't appear to be any AVX512 fast-isel tests for these intrinsics, will add them if required.

Removed incorrect checks from SSE tests.

Updated per comments.

Updated per comments.

Style update.

Updated per comments.