Bump LLVM to llvm/llvm-project@ddda37a#18258
Conversation
Signed-off-by: Stanley Winata <stanley@nod-labs.com>
raikonenfnu
requested review from
Groverkss,
MaheshRavishankar,
antiagainst,
benvanik,
hanhanW,
kuhar and
qedawkins
as code owners
Signed-off-by: Stanley Winata <stanley@nod-labs.com>
Signed-off-by: Stanley Winata <stanley@nod-labs.com>
|
Do you want to wait for the upstream commit to be reviewed and landed. |
|
Currently llvm/llvm-project#104630 breaks 2 things: 1.vm.call + emitc to c code generation. error: null operand found
vm.check.eq %ref_dno, %res, "_v_r()=NULL" : !vm.ref<?>
%11 = "emitc.call_opaque"(%8, <<NULL VALUE>>) <{callee = "vm_cmp_eq_ref"}> : (!emitc.ptr<!emitc.opaque<"iree_vm_ref_t">>, <<NULL TYPE>>) -> i32
memref_util_test.mlir:1:79: error: failed to legalize unresolved materialization from ('i32') to 'i16' that remained live after conversion
util.func @load_store_i16(%buffer: memref<?xi16>, %idx0: index, %idx1: index, %value: i16) -> i16 {
^
memref_util_test.mlir:1:79: note: see current operation: %0 = "builtin.unrealized_conversion_cast"(%arg3) : (i32) -> i16
memref_util_test.mlir:7:3: note: see existing live user here: "util.buffer.store"(%0, %arg0, %4, %6, %5) : (i16, !util.buffer, index, index, index) -> ()
memref.store %value, %buffer[%idx0] : memref<?xi16>
^
memref_util_test.mlir:0:0: error: conversion to util failed
memref_util_test.mlir:0:0: note: see current operation:
"builtin.module"() ({
"util.func"() <{function_type = (memref<?xi16>, index, index, i16) -> i16, sym_name = "load_store_i16", tied_operands = [-1 : index]}> ({
^bb0(%arg0: memref<?xi16>, %arg1: index, %arg2: index, %arg3: i16):
"memref.store"(%arg3, %arg0, %arg1) : (i16, memref<?xi16>, index) -> ()
%0 = "memref.load"(%arg0, %arg2) : (memref<?xi16>, index) -> i16
"util.return"(%0) : (i16) -> ()
}) : () -> ()
}) : () -> ()llvm/llvm-project#104630 fixes issue no.1, however issue no.2 is still unsolved. I tried doing fixing it inside IREE by legalizing UnrealizedConversionCast, but seems like something a little more complex under the hood is going on. |
I'd like to land this for now since I don't want us to get too behind on the llvm upstreaming. If we do this, I will be sure to "ramp up" the next person on why we need this revert and when we can bring it back. But open to otherwise. What do you think @MaheshRavishankar? :) |
|
Sounds good. stamped |
Signed-off-by: Stanley Winata <stanley@nod-labs.com>
Signed-off-by: Stanley Winata <stanley@nod-labs.com>
|
Please put the reverted commits to the PR description. It is very helpful for tracking the local revert commits and cherry-picks. |
Done :) |
3 participants
Bump LLVM upstream while reverting llvm/llvm-project#104630 due to issues which will be detailed below.
Potential fixes:
llvm/llvm-project#104630 for issue no.1
llvm/llvm-project#104648 for issue no.2
llvm/llvm-project#104630 breaks 2 things:
1.vm.call + emitc to c code generation.
builtin.unrealized_conversion_cast, even though we expectedbuiltin.unrealized_conversion_cast.