This is an archive of the discontinued LLVM Phabricator instance.

Differential D51498

[WebAssembly] Vector conversions
ClosedPublic

Authored by tlively on Aug 30 2018, 10:55 AM.

Details

Reviewers
aheejin
dschuff
Commits
rG80725808a3eb: [WebAssembly] Vector conversions
rL341128: [WebAssembly] Vector conversions
Summary

Lowers away bitconverts between vector types. This CL depends
on D51383.

Diff Detail

Event Timeline

tlively created this revision.Aug 30 2018, 10:55 AM
Herald added subscribers: llvm-commits, sunfish, jgravelle-google, sbc100. · View Herald TranscriptAug 30 2018, 10:55 AM
Harbormaster completed remote builds in B22107: Diff 163373.Aug 30 2018, 10:55 AM
tlively mentioned this in D51459: [WebAssembly][WIP] Vector conversions.Aug 30 2018, 10:59 AM
tlively added inline comments.Aug 30 2018, 11:20 AM
lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
283

@dschuff, FWIW, ARM has these same no-op patterns at ARMInstrNEON.td:7129 but without the nice loops

tlively updated this revision to Diff 163402.Aug 30 2018, 1:48 PM
  • Remove bogus opcode
dschuff accepted this revision.Aug 30 2018, 2:12 PM
This revision is now accepted and ready to land.Aug 30 2018, 2:12 PM
Closed by commit rL341128: [WebAssembly] Vector conversions (authored by tlively). · Explain WhyAug 30 2018, 2:45 PM
This revision was automatically updated to reflect the committed changes.
aheejin mentioned this in D51891: [WebAssembly][NFC] Enable explicit locals in simd-arith.ll.Sep 10 2018, 3:26 PM

Revision Contents

PathSize
lib/
Target/
WebAssembly/
2 lines
11 lines
test/
CodeGen/
WebAssembly/
324 lines

Diff 163373

lib/Target/WebAssembly/WebAssemblyInstrInfo.td

Show First 20 Lines • Show All 202 Lines • ▼ Show 20 Lineslet hasSideEffects = 0 in {
defm COPY_#vt : I<(outs vt:$res), (ins vt:$src), (outs), (ins), [], defm COPY_#vt : I<(outs vt:$res), (ins vt:$src), (outs), (ins), [],
"copy_local\t$res, $src", "copy_local">; "copy_local\t$res, $src", "copy_local">;
// TEE is similar to COPY, but writes two copies of its result. Typically // TEE is similar to COPY, but writes two copies of its result. Typically
// this would be used to stackify one result and write the other result to a // this would be used to stackify one result and write the other result to a
// local. // local.
let isAsCheapAsAMove = 1, isCodeGenOnly = 1 in let isAsCheapAsAMove = 1, isCodeGenOnly = 1 in
defm TEE_#vt : I<(outs vt:$res, vt:$also), (ins vt:$src), (outs), (ins), [], defm TEE_#vt : I<(outs vt:$res, vt:$also), (ins vt:$src), (outs), (ins), [],
"tee_local\t$res, $also, $src", "tee_local">; "tee_local\t$res, $also, $src", "tee_local", 0x22>;
// This is the actual get_local instruction in wasm. These are made explicit // This is the actual get_local instruction in wasm. These are made explicit
// by the ExplicitLocals pass. It has mayLoad because it reads from a wasm // by the ExplicitLocals pass. It has mayLoad because it reads from a wasm
// local, which is a side effect not otherwise modeled in LLVM. // local, which is a side effect not otherwise modeled in LLVM.
let mayLoad = 1, isAsCheapAsAMove = 1 in let mayLoad = 1, isAsCheapAsAMove = 1 in
defm GET_LOCAL_#vt : I<(outs vt:$res), (ins local_op:$local), defm GET_LOCAL_#vt : I<(outs vt:$res), (ins local_op:$local),
(outs), (ins local_op:$local), [], (outs), (ins local_op:$local), [],
"get_local\t$res, $local", "get_local\t$local", 0x20>; "get_local\t$res, $local", "get_local\t$local", 0x20>;
▲ Show 20 Lines • Show All 82 Lines • Show Last 20 Lines

lib/Target/WebAssembly/WebAssemblyInstrSIMD.td

Show First 20 Lines • Show All 265 Lines • ▼ Show 20 Lines
} }
// follow convention of making implicit expansions unsigned // follow convention of making implicit expansions unsigned
def : Pat<(i32 (vector_extract (v16i8 V128:$vec), (i32 LaneIdx16:$idx))), def : Pat<(i32 (vector_extract (v16i8 V128:$vec), (i32 LaneIdx16:$idx))),
(EXTRACT_LANE_v16i8_u V128:$vec, (i32 LaneIdx16:$idx))>; (EXTRACT_LANE_v16i8_u V128:$vec, (i32 LaneIdx16:$idx))>;
def : Pat<(i32 (vector_extract (v8i16 V128:$vec), (i32 LaneIdx8:$idx))), def : Pat<(i32 (vector_extract (v8i16 V128:$vec), (i32 LaneIdx8:$idx))),
(EXTRACT_LANE_v8i16_u V128:$vec, (i32 LaneIdx8:$idx))>; (EXTRACT_LANE_v8i16_u V128:$vec, (i32 LaneIdx8:$idx))>;
// Bitcasts are nops
// Matching bitcast t1 to t1 causes strange errors, so avoid repeating types
foreach t1 = [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64] in
foreach t2 = !foldl(
[]<ValueType>, [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
acc, cur, !if(!eq(!cast<string>(t1), !cast<string>(cur)),
acc, !listconcat(acc, [cur])
)
) in
def : Pat<(t1 (bitconvert (t2 V128:$v))), (t1 V128:$v)>;
tlivelyAuthorUnsubmitted
Not Done
ReplyInline Actions

@dschuff, FWIW, ARM has these same no-op patterns at ARMInstrNEON.td:7129 but without the nice loops

tlively: @dschuff, FWIW, ARM has these same no-op patterns at ARMInstrNEON.td:7129 but without the nice…
// arbitrary other BUILD_VECTOR patterns // arbitrary other BUILD_VECTOR patterns
def : Pat<(v16i8 (build_vector def : Pat<(v16i8 (build_vector
(i32 I32:$x0), (i32 I32:$x1), (i32 I32:$x2), (i32 I32:$x3), (i32 I32:$x0), (i32 I32:$x1), (i32 I32:$x2), (i32 I32:$x3),
(i32 I32:$x4), (i32 I32:$x5), (i32 I32:$x6), (i32 I32:$x7), (i32 I32:$x4), (i32 I32:$x5), (i32 I32:$x6), (i32 I32:$x7),
(i32 I32:$x8), (i32 I32:$x9), (i32 I32:$x10), (i32 I32:$x11), (i32 I32:$x8), (i32 I32:$x9), (i32 I32:$x10), (i32 I32:$x11),
(i32 I32:$x12), (i32 I32:$x13), (i32 I32:$x14), (i32 I32:$x15) (i32 I32:$x12), (i32 I32:$x13), (i32 I32:$x14), (i32 I32:$x15)
)), )),
(v16i8 (REPLACE_LANE_v16i8 (v16i8 (REPLACE_LANE_v16i8
▲ Show 20 Lines • Show All 103 Lines • Show Last 20 Lines

test/CodeGen/WebAssembly/simd-conversions.ll

  • This file was added.
; RUN: llc < %s -disable-wasm-fallthrough-return-opt -wasm-keep-registers -wasm-disable-explicit-locals -wasm-enable-unimplemented-simd -mattr=+simd128,+sign-ext | FileCheck %s --check-prefixes CHECK,SIMD128
; RUN: llc < %s -disable-wasm-fallthrough-return-opt -wasm-keep-registers -wasm-disable-explicit-locals -mattr=+simd128,+sign-ext | FileCheck %s --check-prefixes CHECK,SIMD128-VM
; RUN: llc < %s -disable-wasm-fallthrough-return-opt -wasm-keep-registers -wasm-disable-explicit-locals -mattr=-simd128,+sign-ext | FileCheck %s --check-prefixes CHECK,NO-SIMD128
; Test that bitcasts between vector types are lowered to zero instructions
target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
target triple = "wasm32-unknown-unknown"
; CHECK-LABEL: v16i8_to_v16i8:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <16 x i8> @v16i8_to_v16i8(<16 x i8> %v) {
%res = bitcast <16 x i8> %v to <16 x i8>
ret <16 x i8> %res
}
; CHECK-LABEL: v16i8_to_v8i16:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <8 x i16> @v16i8_to_v8i16(<16 x i8> %v) {
%res = bitcast <16 x i8> %v to <8 x i16>
ret <8 x i16> %res
}
; CHECK-LABEL: v16i8_to_v4i32:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <4 x i32> @v16i8_to_v4i32(<16 x i8> %v) {
%res = bitcast <16 x i8> %v to <4 x i32>
ret <4 x i32> %res
}
; CHECK-LABEL: v16i8_to_v2i64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM: v128.store
; SIMD128-VM-NEXT: return #
; SIMD128: return $0
define <2 x i64> @v16i8_to_v2i64(<16 x i8> %v) {
%res = bitcast <16 x i8> %v to <2 x i64>
ret <2 x i64> %res
}
; CHECK-LABEL: v16i8_to_v4f32:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <4 x float> @v16i8_to_v4f32(<16 x i8> %v) {
%res = bitcast <16 x i8> %v to <4 x float>
ret <4 x float> %res
}
; CHECK-LABEL: v16i8_to_v2f64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM: v128.store
; SIMD128-VM-NEXT: return #
; SIMD128: return $0
define <2 x double> @v16i8_to_v2f64(<16 x i8> %v) {
%res = bitcast <16 x i8> %v to <2 x double>
ret <2 x double> %res
}
; CHECK-LABEL: v8i16_to_v16i8:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <16 x i8> @v8i16_to_v16i8(<8 x i16> %v) {
%res = bitcast <8 x i16> %v to <16 x i8>
ret <16 x i8> %res
}
; CHECK-LABEL: v8i16_to_v8i16:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <8 x i16> @v8i16_to_v8i16(<8 x i16> %v) {
%res = bitcast <8 x i16> %v to <8 x i16>
ret <8 x i16> %res
}
; CHECK-LABEL: v8i16_to_v4i32:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <4 x i32> @v8i16_to_v4i32(<8 x i16> %v) {
%res = bitcast <8 x i16> %v to <4 x i32>
ret <4 x i32> %res
}
; CHECK-LABEL: v8i16_to_v2i64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM: v128.store
; SIMD128-VM-NEXT: return #
; SIMD128: return $0
define <2 x i64> @v8i16_to_v2i64(<8 x i16> %v) {
%res = bitcast <8 x i16> %v to <2 x i64>
ret <2 x i64> %res
}
; CHECK-LABEL: v8i16_to_v4f32:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <4 x float> @v8i16_to_v4f32(<8 x i16> %v) {
%res = bitcast <8 x i16> %v to <4 x float>
ret <4 x float> %res
}
; CHECK-LABEL: v8i16_to_v2f64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM: v128.store
; SIMD128-VM-NEXT: return #
; SIMD128: return $0
define <2 x double> @v8i16_to_v2f64(<8 x i16> %v) {
%res = bitcast <8 x i16> %v to <2 x double>
ret <2 x double> %res
}
; CHECK-LABEL: v4i32_to_v16i8:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <16 x i8> @v4i32_to_v16i8(<4 x i32> %v) {
%res = bitcast <4 x i32> %v to <16 x i8>
ret <16 x i8> %res
}
; CHECK-LABEL: v4i32_to_v8i16:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <8 x i16> @v4i32_to_v8i16(<4 x i32> %v) {
%res = bitcast <4 x i32> %v to <8 x i16>
ret <8 x i16> %res
}
; CHECK-LABEL: v4i32_to_v4i32:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <4 x i32> @v4i32_to_v4i32(<4 x i32> %v) {
%res = bitcast <4 x i32> %v to <4 x i32>
ret <4 x i32> %res
}
; CHECK-LABEL: v4i32_to_v2i64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM: v128.store
; SIMD128-VM-NEXT: return #
; SIMD128: return $0
define <2 x i64> @v4i32_to_v2i64(<4 x i32> %v) {
%res = bitcast <4 x i32> %v to <2 x i64>
ret <2 x i64> %res
}
; CHECK-LABEL: v4i32_to_v4f32:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <4 x float> @v4i32_to_v4f32(<4 x i32> %v) {
%res = bitcast <4 x i32> %v to <4 x float>
ret <4 x float> %res
}
; CHECK-LABEL: v4i32_to_v2f64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM: v128.store
; SIMD128-VM-NEXT: return #
; SIMD128: return $0
define <2 x double> @v4i32_to_v2f64(<4 x i32> %v) {
%res = bitcast <4 x i32> %v to <2 x double>
ret <2 x double> %res
}
; CHECK-LABEL: v2i64_to_v16i8:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <16 x i8> @v2i64_to_v16i8(<2 x i64> %v) {
%res = bitcast <2 x i64> %v to <16 x i8>
ret <16 x i8> %res
}
; CHECK-LABEL: v2i64_to_v8i16:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <8 x i16> @v2i64_to_v8i16(<2 x i64> %v) {
%res = bitcast <2 x i64> %v to <8 x i16>
ret <8 x i16> %res
}
; CHECK-LABEL: v2i64_to_v4i32:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <4 x i32> @v2i64_to_v4i32(<2 x i64> %v) {
%res = bitcast <2 x i64> %v to <4 x i32>
ret <4 x i32> %res
}
; CHECK-LABEL: v2i64_to_v2i64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <2 x i64> @v2i64_to_v2i64(<2 x i64> %v) {
%res = bitcast <2 x i64> %v to <2 x i64>
ret <2 x i64> %res
}
; CHECK-LABEL: v2i64_to_v4f32:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <4 x float> @v2i64_to_v4f32(<2 x i64> %v) {
%res = bitcast <2 x i64> %v to <4 x float>
ret <4 x float> %res
}
; CHECK-LABEL: v2i64_to_v2f64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <2 x double> @v2i64_to_v2f64(<2 x i64> %v) {
%res = bitcast <2 x i64> %v to <2 x double>
ret <2 x double> %res
}
; CHECK-LABEL: v4f32_to_v16i8:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <16 x i8> @v4f32_to_v16i8(<4 x float> %v) {
%res = bitcast <4 x float> %v to <16 x i8>
ret <16 x i8> %res
}
; CHECK-LABEL: v4f32_to_v8i16:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <8 x i16> @v4f32_to_v8i16(<4 x float> %v) {
%res = bitcast <4 x float> %v to <8 x i16>
ret <8 x i16> %res
}
; CHECK-LABEL: v4f32_to_v4i32:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <4 x i32> @v4f32_to_v4i32(<4 x float> %v) {
%res = bitcast <4 x float> %v to <4 x i32>
ret <4 x i32> %res
}
; CHECK-LABEL: v4f32_to_v2i64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM: v128.store
; SIMD128-VM-NEXT: return #
; SIMD128: return $0
define <2 x i64> @v4f32_to_v2i64(<4 x float> %v) {
%res = bitcast <4 x float> %v to <2 x i64>
ret <2 x i64> %res
}
; CHECK-LABEL: v4f32_to_v4f32:
; NO-SIMD128-NOT: return $0
; SIMD128: return $0
define <4 x float> @v4f32_to_v4f32(<4 x float> %v) {
%res = bitcast <4 x float> %v to <4 x float>
ret <4 x float> %res
}
; CHECK-LABEL: v4f32_to_v2f64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM: v128.store
; SIMD128-VM-NEXT: return #
; SIMD128: return $0
define <2 x double> @v4f32_to_v2f64(<4 x float> %v) {
%res = bitcast <4 x float> %v to <2 x double>
ret <2 x double> %res
}
; CHECK-LABEL: v2f64_to_v16i8:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <16 x i8> @v2f64_to_v16i8(<2 x double> %v) {
%res = bitcast <2 x double> %v to <16 x i8>
ret <16 x i8> %res
}
; CHECK-LABEL: v2f64_to_v8i16:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <8 x i16> @v2f64_to_v8i16(<2 x double> %v) {
%res = bitcast <2 x double> %v to <8 x i16>
ret <8 x i16> %res
}
; CHECK-LABEL: v2f64_to_v4i32:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <4 x i32> @v2f64_to_v4i32(<2 x double> %v) {
%res = bitcast <2 x double> %v to <4 x i32>
ret <4 x i32> %res
}
; CHECK-LABEL: v2f64_to_v2i64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <2 x i64> @v2f64_to_v2i64(<2 x double> %v) {
%res = bitcast <2 x double> %v to <2 x i64>
ret <2 x i64> %res
}
; CHECK-LABEL: v2f64_to_v4f32:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <4 x float> @v2f64_to_v4f32(<2 x double> %v) {
%res = bitcast <2 x double> %v to <4 x float>
ret <4 x float> %res
}
; CHECK-LABEL: v2f64_to_v2f64:
; NO-SIMD128-NOT: return $0
; SIMD128-VM-NOT: return $0
; SIMD128: return $0
define <2 x double> @v2f64_to_v2f64(<2 x double> %v) {
%res = bitcast <2 x double> %v to <2 x double>
ret <2 x double> %res
}