Cold-Air-Networks/mirror-llvm-project
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 7
mirror-llvm-project/llvm/test/CodeGen/WebAssembly/wide-arithmetic.ll
Alex Crichton b47267441e
[WebAssembly] Don't stackify multi-def instructions (#200429) 
This commit updates the `WebAssemblyRegStackify.cpp` pass to
specifically exclude attempting to stackify the first def of a multi-def
instruction. As the previous comments indicate this is possible to do in
some situations, but the current logic is incomplete and has led to
miscompilations such as #98323 and #199910. One option would be to make
the logic more robust, but in lieu of that in the meantime the change
here is to completely disable stackification in these situations. This
provides at least a "known working" base to build on later and fixes the
known regressions around this.

Closes #98323
Closes #199910
2026-06-08 16:44:11 -07:00

299 lines
9.2 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mattr=+wide-arithmetic < %s -O2 | FileCheck %s
target triple = "wasm32-unknown-unknown"
define i128 @add_i128(i128 %a, i128 %b) {
; CHECK-LABEL: add_i128:
; CHECK: .functype add_i128 (i32, i64, i64, i64, i64) -> ()
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 2
; CHECK-NEXT: local.get 3
; CHECK-NEXT: local.get 4
; CHECK-NEXT: i64.add128
; CHECK-NEXT: local.set 3
; CHECK-NEXT: local.set 4
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 3
; CHECK-NEXT: i64.store 8
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 4
; CHECK-NEXT: i64.store 0
; CHECK-NEXT: # fallthrough-return
%c = add i128 %a, %b
ret i128 %c
}
define i128 @sub_i128(i128 %a, i128 %b) {
; CHECK-LABEL: sub_i128:
; CHECK: .functype sub_i128 (i32, i64, i64, i64, i64) -> ()
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 2
; CHECK-NEXT: local.get 3
; CHECK-NEXT: local.get 4
; CHECK-NEXT: i64.sub128
; CHECK-NEXT: local.set 3
; CHECK-NEXT: local.set 4
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 3
; CHECK-NEXT: i64.store 8
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 4
; CHECK-NEXT: i64.store 0
; CHECK-NEXT: # fallthrough-return
%c = sub i128 %a, %b
ret i128 %c
}
define i128 @mul_i128(i128 %a, i128 %b) {
; CHECK-LABEL: mul_i128:
; CHECK: .functype mul_i128 (i32, i64, i64, i64, i64) -> ()
; CHECK-NEXT: .local i64, i64
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 3
; CHECK-NEXT: i64.mul_wide_u
; CHECK-NEXT: local.set 6
; CHECK-NEXT: local.set 5
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 5
; CHECK-NEXT: i64.store 0
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 6
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 4
; CHECK-NEXT: i64.mul
; CHECK-NEXT: i64.add
; CHECK-NEXT: local.get 2
; CHECK-NEXT: local.get 3
; CHECK-NEXT: i64.mul
; CHECK-NEXT: i64.add
; CHECK-NEXT: i64.store 8
; CHECK-NEXT: # fallthrough-return
%c = mul i128 %a, %b
ret i128 %c
}
define i128 @i64_mul_wide_s(i64 %a, i64 %b) {
; CHECK-LABEL: i64_mul_wide_s:
; CHECK: .functype i64_mul_wide_s (i32, i64, i64) -> ()
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.mul_wide_s
; CHECK-NEXT: local.set 1
; CHECK-NEXT: local.set 2
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.store 8
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.store 0
; CHECK-NEXT: # fallthrough-return
%a128 = sext i64 %a to i128
%b128 = sext i64 %b to i128
%c = mul i128 %a128, %b128
ret i128 %c
}
define i128 @i64_mul_wide_u(i64 %a, i64 %b) {
; CHECK-LABEL: i64_mul_wide_u:
; CHECK: .functype i64_mul_wide_u (i32, i64, i64) -> ()
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.mul_wide_u
; CHECK-NEXT: local.set 1
; CHECK-NEXT: local.set 2
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.store 8
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.store 0
; CHECK-NEXT: # fallthrough-return
%a128 = zext i64 %a to i128
%b128 = zext i64 %b to i128
%c = mul i128 %a128, %b128
ret i128 %c
}
define i64 @mul_i128_only_lo(i128 %a, i128 %b) {
; CHECK-LABEL: mul_i128_only_lo:
; CHECK: .functype mul_i128_only_lo (i64, i64, i64, i64) -> (i64)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.mul
; CHECK-NEXT: # fallthrough-return
%c = mul i128 %a, %b
%d = trunc i128 %c to i64
ret i64 %d
}
declare { i64, i1 } @llvm.sadd.with.overflow.i64(i64, i64)
declare { i64, i1 } @llvm.uadd.with.overflow.i64(i64, i64)
; This is a codegen test to see the effect of overflowing adds on signed
; integers with wide-arithmetic enabled. At this time it doesn't actually
; generate anything differently than without wide-arithmetic but this has also
; been useful for evaluating the proposal.
define { i64, i1 } @add_wide_s(i64 %a, i64 %b) {
; CHECK-LABEL: add_wide_s:
; CHECK: .functype add_wide_s (i32, i64, i64) -> ()
; CHECK-NEXT: .local i64
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.add
; CHECK-NEXT: local.tee 3
; CHECK-NEXT: i64.store 0
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: i64.lt_s
; CHECK-NEXT: local.get 3
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.lt_s
; CHECK-NEXT: i32.xor
; CHECK-NEXT: i32.store8 8
; CHECK-NEXT: # fallthrough-return
%pair = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
ret { i64, i1 } %pair
}
define { i64, i1 } @add_wide_u(i64 %a, i64 %b) {
; CHECK-LABEL: add_wide_u:
; CHECK: .functype add_wide_u (i32, i64, i64) -> ()
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: i64.add128
; CHECK-NEXT: local.set 1
; CHECK-NEXT: local.set 2
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.store8 8
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.store 0
; CHECK-NEXT: # fallthrough-return
%pair = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
ret { i64, i1 } %pair
}
; This is a model of a hypothetical `i64.add_wide3_u` instruction using LLVM
; intrinsics. In theory this should optimize better (to the equivalent below)
; but it doesn't currently.
define { i64, i64 } @add_wide3_u_via_intrinsics(i64 %a, i64 %b, i64 %c) {
; CHECK-LABEL: add_wide3_u_via_intrinsics:
; CHECK: .functype add_wide3_u_via_intrinsics (i32, i64, i64, i64) -> ()
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: i64.add128
; CHECK-NEXT: local.set 1
; CHECK-NEXT: local.set 2
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: local.get 3
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: i64.add128
; CHECK-NEXT: local.set 3
; CHECK-NEXT: local.set 2
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.store 0
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 3
; CHECK-NEXT: i64.add
; CHECK-NEXT: i64.store 8
; CHECK-NEXT: # fallthrough-return
%pair = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
%t0 = extractvalue { i64, i1 } %pair, 0
%carry1 = extractvalue { i64, i1 } %pair, 1
%pair2 = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %t0, i64 %c)
%ret1 = extractvalue { i64, i1 } %pair2, 0
%carry2 = extractvalue { i64, i1 } %pair2, 1
%carry1_64 = zext i1 %carry1 to i64
%carry2_64 = zext i1 %carry2 to i64
%ret2 = add i64 %carry1_64, %carry2_64
%r0 = insertvalue { i64, i64 } poison, i64 %ret1, 0
%r1 = insertvalue { i64, i64 } %r0, i64 %ret2, 1
ret { i64, i64 } %r1
}
; This is a model of a hypothetical `i64.add_wide3_u` instruction using 128-bit
; integer addition. This optimizes better than the above currently.
define { i64, i64 } @add_wide3_u_via_i128(i64 %a, i64 %b, i64 %c) {
; CHECK-LABEL: add_wide3_u_via_i128:
; CHECK: .functype add_wide3_u_via_i128 (i32, i64, i64, i64) -> ()
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: i64.add128
; CHECK-NEXT: local.set 1
; CHECK-NEXT: local.set 2
; CHECK-NEXT: local.get 2
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 3
; CHECK-NEXT: i64.const 0
; CHECK-NEXT: i64.add128
; CHECK-NEXT: local.set 1
; CHECK-NEXT: local.set 2
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.store 8
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i64.store 0
; CHECK-NEXT: # fallthrough-return
%a128 = zext i64 %a to i128
%b128 = zext i64 %b to i128
%c128 = zext i64 %c to i128
%t0 = add i128 %a128, %b128
%t1 = add i128 %t0, %c128
%result = trunc i128 %t1 to i64
%t2 = lshr i128 %t1, 64
%carry = trunc i128 %t2 to i64
%ret0 = insertvalue { i64, i64 } poison, i64 %result, 0
%ret1 = insertvalue { i64, i64 } %ret0, i64 %carry, 1
ret { i64, i64 } %ret1
}
define i1 @smul64_with_overflow(i64 %a, i64 %b) {
; CHECK-LABEL: smul64_with_overflow:
; CHECK: .functype smul64_with_overflow (i64, i64) -> (i32)
; CHECK-NEXT: # %bb.0: # %entry
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.mul_wide_s
; CHECK-NEXT: local.set 0
; CHECK-NEXT: local.set 1
; CHECK-NEXT: local.get 0
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i64.const 63
; CHECK-NEXT: i64.shr_s
; CHECK-NEXT: i64.ne
; CHECK-NEXT: # fallthrough-return
entry:
%res = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
%ov = extractvalue { i64, i1 } %res, 1
ret i1 %ov
}
declare { i64, i1 } @llvm.smul.with.overflow.i64(i64, i64)
Reference in a new issue View git blame Copy permalink