shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/CodeGen/WebAssembly/simd-bitmask.ll
Folkert de Vries b861a289d7
[WebAssembly] combine bitmask with setcc <X>, 0, setlt (#179065) 
The rust `simd_bitmask` intrinsic is UB when the lanes of its input are
not either `0` or `!0`, presumably so that the implementation can be
more efficient because it could look at any bit. To get the "mask of
MSB" behavior of webassembly's `bitmask`, we would like to simply first
compare with a zero vector.

```llvm
define i32 @example(<2 x i64> noundef %v) {
entry:
  %1 = icmp slt <16 x i8> %v, zeroinitializer
  %2 = bitcast <16 x i1> %1 to i16
  %3 = zext i16 %2 to i32
  ret i32 %3
}
```

On x86_64, this additional comparison optimizes away, but for wasm it
does not.

https://godbolt.org/z/T5sPejocs

This PR adds a new combine, so that instead of emitting 

```asm
example:
        local.get       0
        v128.const      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
        i8x16.lt_s
        i8x16.bitmask
        end_function
```

we just emit 

```
example:
        local.get       0
        i8x16.bitmask
        end_function
```
2026-03-17 11:21:39 +00:00

319 lines
9.6 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mattr=+simd128 | FileCheck %s
;; Test that SIMD bitmask instruction can be selected
target triple = "wasm32-unknown-unknown"
define i16 @bitmask_v16i8(<16 x i8> %v) {
; CHECK-LABEL: bitmask_v16i8:
; CHECK: .functype bitmask_v16i8 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: v128.const 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
; CHECK-NEXT: i8x16.eq
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: # fallthrough-return
%cmp = icmp eq <16 x i8> %v, zeroinitializer
%bitmask = bitcast <16 x i1> %cmp to i16
ret i16 %bitmask
}
define i8 @bitmask_v8i16(<8 x i16> %v) {
; CHECK-LABEL: bitmask_v8i16:
; CHECK: .functype bitmask_v8i16 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: v128.const 0, 0, 0, 0, 0, 0, 0, 0
; CHECK-NEXT: i16x8.eq
; CHECK-NEXT: i16x8.bitmask
; CHECK-NEXT: # fallthrough-return
%cmp = icmp eq <8 x i16> %v, zeroinitializer
%bitmask = bitcast <8 x i1> %cmp to i8
ret i8 %bitmask
}
define i8 @bitmask_v4i32(<4 x i32> %v) {
; CHECK-LABEL: bitmask_v4i32:
; CHECK: .functype bitmask_v4i32 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: v128.const 0, 0, 0, 0
; CHECK-NEXT: i32x4.eq
; CHECK-NEXT: i32x4.bitmask
; CHECK-NEXT: # fallthrough-return
%cmp = icmp eq <4 x i32> %v, zeroinitializer
%bitmask = bitcast <4 x i1> %cmp to i4
%ext = zext i4 %bitmask to i8
ret i8 %ext
}
define i8 @bitmask_v2i64(<2 x i64> %v) {
; CHECK-LABEL: bitmask_v2i64:
; CHECK: .functype bitmask_v2i64 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: v128.const 0, 0
; CHECK-NEXT: i64x2.eq
; CHECK-NEXT: i64x2.bitmask
; CHECK-NEXT: # fallthrough-return
%cmp = icmp eq <2 x i64> %v, zeroinitializer
%bitmask = bitcast <2 x i1> %cmp to i2
%ext = zext i2 %bitmask to i8
ret i8 %ext
}
;; Test unusual vectors
define i1 @bitmask_v1i8(<1 x i8> %v) {
; CHECK-LABEL: bitmask_v1i8:
; CHECK: .functype bitmask_v1i8 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i8x16.extract_lane_u 0
; CHECK-NEXT: i32.eqz
; CHECK-NEXT: # fallthrough-return
%cmp = icmp eq <1 x i8> %v, zeroinitializer
%bitmask = bitcast <1 x i1> %cmp to i1
ret i1 %bitmask
}
define i7 @bitmask_v7i8(<7 x i8> %v) {
; CHECK-LABEL: bitmask_v7i8:
; CHECK: .functype bitmask_v7i8 (i32, i32, i32, i32, i32, i32, i32) -> (i32)
; CHECK-NEXT: .local v128
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: global.get __stack_pointer
; CHECK-NEXT: i32.const 16
; CHECK-NEXT: i32.sub
; CHECK-NEXT: drop
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i8x16.splat
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.replace_lane 1
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i8x16.replace_lane 2
; CHECK-NEXT: local.get 3
; CHECK-NEXT: i8x16.replace_lane 3
; CHECK-NEXT: local.get 4
; CHECK-NEXT: i8x16.replace_lane 4
; CHECK-NEXT: local.get 5
; CHECK-NEXT: i8x16.replace_lane 5
; CHECK-NEXT: local.get 6
; CHECK-NEXT: i8x16.replace_lane 6
; CHECK-NEXT: v128.const 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
; CHECK-NEXT: i8x16.eq
; CHECK-NEXT: local.tee 7
; CHECK-NEXT: i16x8.extract_lane_u 0
; CHECK-NEXT: i32.const 1
; CHECK-NEXT: i32.and
; CHECK-NEXT: local.get 7
; CHECK-NEXT: i16x8.extend_low_i8x16_s
; CHECK-NEXT: local.tee 7
; CHECK-NEXT: i16x8.extract_lane_u 1
; CHECK-NEXT: i32.const 1
; CHECK-NEXT: i32.and
; CHECK-NEXT: i32.const 1
; CHECK-NEXT: i32.shl
; CHECK-NEXT: i32.or
; CHECK-NEXT: local.get 7
; CHECK-NEXT: i16x8.extract_lane_u 2
; CHECK-NEXT: i32.const 1
; CHECK-NEXT: i32.and
; CHECK-NEXT: i32.const 2
; CHECK-NEXT: i32.shl
; CHECK-NEXT: i32.or
; CHECK-NEXT: local.get 7
; CHECK-NEXT: i16x8.extract_lane_u 3
; CHECK-NEXT: i32.const 1
; CHECK-NEXT: i32.and
; CHECK-NEXT: i32.const 3
; CHECK-NEXT: i32.shl
; CHECK-NEXT: i32.or
; CHECK-NEXT: local.get 7
; CHECK-NEXT: i16x8.extract_lane_u 4
; CHECK-NEXT: i32.const 1
; CHECK-NEXT: i32.and
; CHECK-NEXT: i32.const 4
; CHECK-NEXT: i32.shl
; CHECK-NEXT: i32.or
; CHECK-NEXT: local.get 7
; CHECK-NEXT: i16x8.extract_lane_u 5
; CHECK-NEXT: i32.const 1
; CHECK-NEXT: i32.and
; CHECK-NEXT: i32.const 5
; CHECK-NEXT: i32.shl
; CHECK-NEXT: i32.or
; CHECK-NEXT: local.get 7
; CHECK-NEXT: i16x8.extract_lane_u 6
; CHECK-NEXT: i32.const 6
; CHECK-NEXT: i32.shl
; CHECK-NEXT: i32.or
; CHECK-NEXT: i32.const 127
; CHECK-NEXT: i32.and
; CHECK-NEXT: # fallthrough-return
%cmp = icmp eq <7 x i8> %v, zeroinitializer
%bitmask = bitcast <7 x i1> %cmp to i7
ret i7 %bitmask
}
define i8 @bitmask_v8i8(<8 x i8> %v) {
; CHECK-LABEL: bitmask_v8i8:
; CHECK: .functype bitmask_v8i8 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: v128.const 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
; CHECK-NEXT: i8x16.eq
; CHECK-NEXT: i16x8.extend_low_i8x16_s
; CHECK-NEXT: i16x8.bitmask
; CHECK-NEXT: # fallthrough-return
%cmp = icmp eq <8 x i8> %v, zeroinitializer
%bitmask = bitcast <8 x i1> %cmp to i8
ret i8 %bitmask
}
define i32 @bitmask_v32i8(<32 x i8> %v) {
; CHECK-LABEL: bitmask_v32i8:
; CHECK: .functype bitmask_v32i8 (v128, v128) -> (i32)
; CHECK-NEXT: .local v128
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: i32.const 16
; CHECK-NEXT: local.get 0
; CHECK-NEXT: v128.const 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
; CHECK-NEXT: local.tee 2
; CHECK-NEXT: i8x16.eq
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: i32.const 16
; CHECK-NEXT: i32.add
; CHECK-NEXT: i32.shl
; CHECK-NEXT: local.get 1
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i8x16.eq
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: i32.add
; CHECK-NEXT: # fallthrough-return
%cmp = icmp eq <32 x i8> %v, zeroinitializer
%bitmask = bitcast <32 x i1> %cmp to i32
ret i32 %bitmask
}
define i32 @manual_bitmask_i8x16(<16 x i8> %v) {
; CHECK-LABEL: manual_bitmask_i8x16:
; CHECK: .functype manual_bitmask_i8x16 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: # fallthrough-return
%1 = icmp slt <16 x i8> %v, zeroinitializer
%2 = bitcast <16 x i1> %1 to i16
%3 = zext i16 %2 to i32
ret i32 %3
}
define i32 @manual_bitmask_i16x8(<8 x i16> %v) {
; CHECK-LABEL: manual_bitmask_i16x8:
; CHECK: .functype manual_bitmask_i16x8 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i16x8.bitmask
; CHECK-NEXT: # fallthrough-return
%1 = icmp slt <8 x i16> %v, zeroinitializer
%2 = bitcast <8 x i1> %1 to i8
%3 = zext i8 %2 to i32
ret i32 %3
}
define i32 @manual_bitmask_i32x4(<4 x i32> %v) {
; CHECK-LABEL: manual_bitmask_i32x4:
; CHECK: .functype manual_bitmask_i32x4 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i32x4.bitmask
; CHECK-NEXT: # fallthrough-return
%1 = icmp slt <4 x i32> %v, zeroinitializer
%2 = bitcast <4 x i1> %1 to i4
%3 = zext i4 %2 to i32
ret i32 %3
}
define i32 @manual_bitmask_i64x2(<2 x i64> %v) {
; CHECK-LABEL: manual_bitmask_i64x2:
; CHECK: .functype manual_bitmask_i64x2 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i64x2.bitmask
; CHECK-NEXT: # fallthrough-return
%1 = icmp slt <2 x i64> %v, zeroinitializer
%2 = bitcast <2 x i1> %1 to i2
%3 = zext i2 %2 to i32
ret i32 %3
}
define i32 @manual_bitmask_v8i8(<8 x i8> %v) {
; CHECK-LABEL: manual_bitmask_v8i8:
; CHECK: .functype manual_bitmask_v8i8 (v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: local.get 0
; CHECK-NEXT: v128.const 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
; CHECK-NEXT: i8x16.lt_s
; CHECK-NEXT: i16x8.extend_low_i8x16_s
; CHECK-NEXT: i16x8.bitmask
; CHECK-NEXT: # fallthrough-return
%1 = icmp slt <8 x i8> %v, zeroinitializer
%2 = bitcast <8 x i1> %1 to i8
%3 = zext i8 %2 to i32
ret i32 %3
}
define i32 @manual_bitmask_v32i8(<32 x i8> %v) {
; CHECK-LABEL: manual_bitmask_v32i8:
; CHECK: .functype manual_bitmask_v32i8 (v128, v128) -> (i32)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: i32.const 16
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: i32.const 16
; CHECK-NEXT: i32.add
; CHECK-NEXT: i32.shl
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: i32.add
; CHECK-NEXT: # fallthrough-return
%1 = icmp slt <32 x i8> %v, zeroinitializer
%2 = bitcast <32 x i1> %1 to i32
ret i32 %2
}
define i64 @manual_bitmask_v64i8(<64 x i8> %v) {
; CHECK-LABEL: manual_bitmask_v64i8:
; CHECK: .functype manual_bitmask_v64i8 (v128, v128, v128, v128) -> (i64)
; CHECK-NEXT: # %bb.0:
; CHECK-NEXT: i64.const 16
; CHECK-NEXT: i64.const 16
; CHECK-NEXT: i64.const 16
; CHECK-NEXT: local.get 0
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: i64.extend_i32_u
; CHECK-NEXT: i64.const 16
; CHECK-NEXT: i64.add
; CHECK-NEXT: i64.shl
; CHECK-NEXT: local.get 1
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: i64.extend_i32_u
; CHECK-NEXT: i64.add
; CHECK-NEXT: i64.shl
; CHECK-NEXT: local.get 2
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: i64.extend_i32_u
; CHECK-NEXT: i64.add
; CHECK-NEXT: i64.shl
; CHECK-NEXT: local.get 3
; CHECK-NEXT: i8x16.bitmask
; CHECK-NEXT: i64.extend_i32_u
; CHECK-NEXT: i64.add
; CHECK-NEXT: # fallthrough-return
%1 = icmp slt <64 x i8> %v, zeroinitializer
%2 = bitcast <64 x i1> %1 to i64
ret i64 %2
}
Reference in New Issue View Git Blame Copy Permalink