4a8c1f7d12
[WebAssembly] [Backend] Wasm optimize illegal bitmask for #131980. Currently, the case for illegal bitmask (v32i8 or v64i8) is that at the SelectionDag level, two (four) vectors of v128 will be concatenated together, then they'll all be SETCC by the same pseudo illegal instruction, which requires expansion later on. I opt for SETCC-ing them seperately, bitcast and zext them and then add them up together in the end. --------- Co-authored-by: badumbatish <--show-origin>
199 lines
6.0 KiB
LLVM
199 lines
6.0 KiB
LLVM
; 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
|
|
}
|