From a35e9abc76a44eaef5316231cd4dca3638d61f07 Mon Sep 17 00:00:00 2001 From: Farzon Lotfi <1802579+farzonl@users.noreply.github.com> Date: Fri, 19 Jul 2024 10:18:58 -0400 Subject: [PATCH] [WebAssembly] Add Support for Arc and Hyperbolic trig llvm intrinsics (#98755) ## Change: - WebAssemblyRuntimeLibcallSignatures.cpp: Expose the RTLIB's for use by WASM - Add trig specific test cases ## History This change is part of an implementation of https://github.com/llvm/llvm-project/issues/87367's investigation on supporting IEEE math operations as intrinsics. Which was discussed in this RFC: https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294 This change adds wasm lowering cases for `acos`, `asin`, `atan`, `cosh`, `sinh`, and `tanh`. https://github.com/llvm/llvm-project/issues/70079 https://github.com/llvm/llvm-project/issues/70080 https://github.com/llvm/llvm-project/issues/70081 https://github.com/llvm/llvm-project/issues/70083 https://github.com/llvm/llvm-project/issues/70084 https://github.com/llvm/llvm-project/issues/95966 ## Why Web Assembly? From past changes to try and support constraint intrinsics the changes to the trig builtins to emit intrinsics\constraint intrinsics broke the WASM build. This is an attempt to preempt any such build break. - https://github.com/llvm/llvm-project/pull/95082 - https://github.com/llvm/llvm-project/pull/94559#issuecomment-2159923215 --- .../WebAssemblyRuntimeLibcallSignatures.cpp | 18 ++ .../test/CodeGen/WebAssembly/libcalls-trig.ll | 234 ++++++++++++++++++ 2 files changed, 252 insertions(+) create mode 100644 llvm/test/CodeGen/WebAssembly/libcalls-trig.ll diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyRuntimeLibcallSignatures.cpp b/llvm/lib/Target/WebAssembly/WebAssemblyRuntimeLibcallSignatures.cpp index e7810e18d44d42..ba3ab5164af267 100644 --- a/llvm/lib/Target/WebAssembly/WebAssemblyRuntimeLibcallSignatures.cpp +++ b/llvm/lib/Target/WebAssembly/WebAssemblyRuntimeLibcallSignatures.cpp @@ -204,6 +204,24 @@ struct RuntimeLibcallSignatureTable { Table[RTLIB::TAN_F32] = f32_func_f32; Table[RTLIB::TAN_F64] = f64_func_f64; Table[RTLIB::TAN_F128] = i64_i64_func_i64_i64; + Table[RTLIB::ASIN_F32] = f32_func_f32; + Table[RTLIB::ASIN_F64] = f64_func_f64; + Table[RTLIB::ASIN_F128] = i64_i64_func_i64_i64; + Table[RTLIB::ACOS_F32] = f32_func_f32; + Table[RTLIB::ACOS_F64] = f64_func_f64; + Table[RTLIB::ACOS_F128] = i64_i64_func_i64_i64; + Table[RTLIB::ATAN_F32] = f32_func_f32; + Table[RTLIB::ATAN_F64] = f64_func_f64; + Table[RTLIB::ATAN_F128] = i64_i64_func_i64_i64; + Table[RTLIB::SINH_F32] = f32_func_f32; + Table[RTLIB::SINH_F64] = f64_func_f64; + Table[RTLIB::SINH_F128] = i64_i64_func_i64_i64; + Table[RTLIB::COSH_F32] = f32_func_f32; + Table[RTLIB::COSH_F64] = f64_func_f64; + Table[RTLIB::COSH_F128] = i64_i64_func_i64_i64; + Table[RTLIB::TANH_F32] = f32_func_f32; + Table[RTLIB::TANH_F64] = f64_func_f64; + Table[RTLIB::TANH_F128] = i64_i64_func_i64_i64; Table[RTLIB::SINCOS_F32] = func_f32_iPTR_iPTR; Table[RTLIB::SINCOS_F64] = func_f64_iPTR_iPTR; Table[RTLIB::SINCOS_F128] = func_i64_i64_iPTR_iPTR; diff --git a/llvm/test/CodeGen/WebAssembly/libcalls-trig.ll b/llvm/test/CodeGen/WebAssembly/libcalls-trig.ll new file mode 100644 index 00000000000000..8bc9c043fcf8ae --- /dev/null +++ b/llvm/test/CodeGen/WebAssembly/libcalls-trig.ll @@ -0,0 +1,234 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc < %s -disable-wasm-fallthrough-return-opt -wasm-keep-registers | FileCheck %s + +; Test a subset of compiler-rt/libm libcalls expected to be emitted by the wasm backend + +target triple = "wasm32-unknown-unknown" + +declare fp128 @llvm.sin.f128(fp128) +declare fp128 @llvm.cos.f128(fp128) +declare fp128 @llvm.tan.f128(fp128) +declare fp128 @llvm.asin.f128(fp128) +declare fp128 @llvm.acos.f128(fp128) +declare fp128 @llvm.atan.f128.i32(fp128) +declare fp128 @llvm.sinh.f128(fp128) +declare fp128 @llvm.cosh.f128(fp128) +declare fp128 @llvm.tanh.f128(fp128) + +declare double @llvm.sin.f64(double) +declare double @llvm.cos.f64(double) +declare double @llvm.tan.f64(double) +declare double @llvm.asin.f64(double) +declare double @llvm.acos.f64(double) +declare double @llvm.atan.f64(double) +declare double @llvm.sinh.f64(double) +declare double @llvm.cosh.f64(double) +declare double @llvm.tanh.f64(double) + +declare float @llvm.sin.f32(float) +declare float @llvm.cos.f32(float) +declare float @llvm.tan.f32(float) +declare float @llvm.asin.f32(float) +declare float @llvm.acos.f32(float) +declare float @llvm.atan.f32(float) +declare float @llvm.sinh.f32(float) +declare float @llvm.cosh.f32(float) +declare float @llvm.tanh.f32(float) + + +define fp128 @fp128libcalls(fp128 %x) { + ; compiler-rt call +; CHECK-LABEL: fp128libcalls: +; CHECK: .functype fp128libcalls (i32, i64, i64) -> () +; CHECK-NEXT: .local i32 +; CHECK-NEXT: # %bb.0: +; CHECK-NEXT: global.get $push28=, __stack_pointer +; CHECK-NEXT: i32.const $push29=, 144 +; CHECK-NEXT: i32.sub $push73=, $pop28, $pop29 +; CHECK-NEXT: local.tee $push72=, 3, $pop73 +; CHECK-NEXT: global.set __stack_pointer, $pop72 +; CHECK-NEXT: local.get $push74=, 3 +; CHECK-NEXT: i32.const $push62=, 128 +; CHECK-NEXT: i32.add $push63=, $pop74, $pop62 +; CHECK-NEXT: local.get $push76=, 1 +; CHECK-NEXT: local.get $push75=, 2 +; CHECK-NEXT: call sinl, $pop63, $pop76, $pop75 +; CHECK-NEXT: local.get $push77=, 3 +; CHECK-NEXT: i32.const $push58=, 112 +; CHECK-NEXT: i32.add $push59=, $pop77, $pop58 +; CHECK-NEXT: local.get $push78=, 3 +; CHECK-NEXT: i64.load $push3=, 128($pop78) +; CHECK-NEXT: local.get $push79=, 3 +; CHECK-NEXT: i32.const $push60=, 128 +; CHECK-NEXT: i32.add $push61=, $pop79, $pop60 +; CHECK-NEXT: i32.const $push0=, 8 +; CHECK-NEXT: i32.add $push1=, $pop61, $pop0 +; CHECK-NEXT: i64.load $push2=, 0($pop1) +; CHECK-NEXT: call cosl, $pop59, $pop3, $pop2 +; CHECK-NEXT: local.get $push80=, 3 +; CHECK-NEXT: i32.const $push54=, 96 +; CHECK-NEXT: i32.add $push55=, $pop80, $pop54 +; CHECK-NEXT: local.get $push81=, 3 +; CHECK-NEXT: i64.load $push6=, 112($pop81) +; CHECK-NEXT: local.get $push82=, 3 +; CHECK-NEXT: i32.const $push56=, 112 +; CHECK-NEXT: i32.add $push57=, $pop82, $pop56 +; CHECK-NEXT: i32.const $push71=, 8 +; CHECK-NEXT: i32.add $push4=, $pop57, $pop71 +; CHECK-NEXT: i64.load $push5=, 0($pop4) +; CHECK-NEXT: call tanl, $pop55, $pop6, $pop5 +; CHECK-NEXT: local.get $push83=, 3 +; CHECK-NEXT: i32.const $push50=, 80 +; CHECK-NEXT: i32.add $push51=, $pop83, $pop50 +; CHECK-NEXT: local.get $push84=, 3 +; CHECK-NEXT: i64.load $push9=, 96($pop84) +; CHECK-NEXT: local.get $push85=, 3 +; CHECK-NEXT: i32.const $push52=, 96 +; CHECK-NEXT: i32.add $push53=, $pop85, $pop52 +; CHECK-NEXT: i32.const $push70=, 8 +; CHECK-NEXT: i32.add $push7=, $pop53, $pop70 +; CHECK-NEXT: i64.load $push8=, 0($pop7) +; CHECK-NEXT: call asinl, $pop51, $pop9, $pop8 +; CHECK-NEXT: local.get $push86=, 3 +; CHECK-NEXT: i32.const $push46=, 64 +; CHECK-NEXT: i32.add $push47=, $pop86, $pop46 +; CHECK-NEXT: local.get $push87=, 3 +; CHECK-NEXT: i64.load $push12=, 80($pop87) +; CHECK-NEXT: local.get $push88=, 3 +; CHECK-NEXT: i32.const $push48=, 80 +; CHECK-NEXT: i32.add $push49=, $pop88, $pop48 +; CHECK-NEXT: i32.const $push69=, 8 +; CHECK-NEXT: i32.add $push10=, $pop49, $pop69 +; CHECK-NEXT: i64.load $push11=, 0($pop10) +; CHECK-NEXT: call acosl, $pop47, $pop12, $pop11 +; CHECK-NEXT: local.get $push89=, 3 +; CHECK-NEXT: i32.const $push42=, 48 +; CHECK-NEXT: i32.add $push43=, $pop89, $pop42 +; CHECK-NEXT: local.get $push90=, 3 +; CHECK-NEXT: i64.load $push15=, 64($pop90) +; CHECK-NEXT: local.get $push91=, 3 +; CHECK-NEXT: i32.const $push44=, 64 +; CHECK-NEXT: i32.add $push45=, $pop91, $pop44 +; CHECK-NEXT: i32.const $push68=, 8 +; CHECK-NEXT: i32.add $push13=, $pop45, $pop68 +; CHECK-NEXT: i64.load $push14=, 0($pop13) +; CHECK-NEXT: call atanl, $pop43, $pop15, $pop14 +; CHECK-NEXT: local.get $push92=, 3 +; CHECK-NEXT: i32.const $push38=, 32 +; CHECK-NEXT: i32.add $push39=, $pop92, $pop38 +; CHECK-NEXT: local.get $push93=, 3 +; CHECK-NEXT: i64.load $push18=, 48($pop93) +; CHECK-NEXT: local.get $push94=, 3 +; CHECK-NEXT: i32.const $push40=, 48 +; CHECK-NEXT: i32.add $push41=, $pop94, $pop40 +; CHECK-NEXT: i32.const $push67=, 8 +; CHECK-NEXT: i32.add $push16=, $pop41, $pop67 +; CHECK-NEXT: i64.load $push17=, 0($pop16) +; CHECK-NEXT: call sinhl, $pop39, $pop18, $pop17 +; CHECK-NEXT: local.get $push95=, 3 +; CHECK-NEXT: i32.const $push34=, 16 +; CHECK-NEXT: i32.add $push35=, $pop95, $pop34 +; CHECK-NEXT: local.get $push96=, 3 +; CHECK-NEXT: i64.load $push21=, 32($pop96) +; CHECK-NEXT: local.get $push97=, 3 +; CHECK-NEXT: i32.const $push36=, 32 +; CHECK-NEXT: i32.add $push37=, $pop97, $pop36 +; CHECK-NEXT: i32.const $push66=, 8 +; CHECK-NEXT: i32.add $push19=, $pop37, $pop66 +; CHECK-NEXT: i64.load $push20=, 0($pop19) +; CHECK-NEXT: call coshl, $pop35, $pop21, $pop20 +; CHECK-NEXT: local.get $push100=, 3 +; CHECK-NEXT: local.get $push98=, 3 +; CHECK-NEXT: i64.load $push24=, 16($pop98) +; CHECK-NEXT: local.get $push99=, 3 +; CHECK-NEXT: i32.const $push32=, 16 +; CHECK-NEXT: i32.add $push33=, $pop99, $pop32 +; CHECK-NEXT: i32.const $push65=, 8 +; CHECK-NEXT: i32.add $push22=, $pop33, $pop65 +; CHECK-NEXT: i64.load $push23=, 0($pop22) +; CHECK-NEXT: call tanhl, $pop100, $pop24, $pop23 +; CHECK-NEXT: local.get $push102=, 0 +; CHECK-NEXT: local.get $push101=, 3 +; CHECK-NEXT: i32.const $push64=, 8 +; CHECK-NEXT: i32.add $push25=, $pop101, $pop64 +; CHECK-NEXT: i64.load $push26=, 0($pop25) +; CHECK-NEXT: i64.store 8($pop102), $pop26 +; CHECK-NEXT: local.get $push104=, 0 +; CHECK-NEXT: local.get $push103=, 3 +; CHECK-NEXT: i64.load $push27=, 0($pop103) +; CHECK-NEXT: i64.store 0($pop104), $pop27 +; CHECK-NEXT: local.get $push105=, 3 +; CHECK-NEXT: i32.const $push30=, 144 +; CHECK-NEXT: i32.add $push31=, $pop105, $pop30 +; CHECK-NEXT: global.set __stack_pointer, $pop31 +; CHECK-NEXT: return + ; libm calls + %d = call fp128 @llvm.sin.f128(fp128 %x) + %e = call fp128 @llvm.cos.f128(fp128 %d) + %f = call fp128 @llvm.tan.f128(fp128 %e) + %g = call fp128 @llvm.asin.f128.i32(fp128 %f) + %h = call fp128 @llvm.acos.f128(fp128 %g) + %i = call fp128 @llvm.atan.f128(fp128 %h) + %a = call fp128 @llvm.sinh.f128(fp128 %i) + %b = call fp128 @llvm.cosh.f128(fp128 %a) + %c = call fp128 @llvm.tanh.f128(fp128 %b) + ret fp128 %c +} + +define double @f64libcalls(double %x) { +; CHECK-LABEL: f64libcalls: +; CHECK: .functype f64libcalls (f64) -> (f64) +; CHECK-NEXT: # %bb.0: +; CHECK-NEXT: local.get $push9=, 0 +; CHECK-NEXT: call $push0=, sin, $pop9 +; CHECK-NEXT: call $push1=, cos, $pop0 +; CHECK-NEXT: call $push2=, tan, $pop1 +; CHECK-NEXT: call $push3=, asin, $pop2 +; CHECK-NEXT: call $push4=, acos, $pop3 +; CHECK-NEXT: call $push5=, atan, $pop4 +; CHECK-NEXT: call $push6=, sinh, $pop5 +; CHECK-NEXT: call $push7=, cosh, $pop6 +; CHECK-NEXT: call $push8=, tanh, $pop7 +; CHECK-NEXT: return $pop8 + + + %k = call double @llvm.sin.f64(double %x) + %a = call double @llvm.cos.f64(double %k) + %b = call double @llvm.tan.f64(double %a) + %c = call double @llvm.asin.f64(double %b) + %d = call double @llvm.acos.f64(double %c) + %e = call double @llvm.atan.f64(double %d) + %f = call double @llvm.sinh.f64(double %e) + %g = call double @llvm.cosh.f64(double %f) + %h = call double @llvm.tanh.f64(double %g) + ret double %h +} + +define float @f32libcalls(float %x) { +; CHECK-LABEL: f32libcalls: +; CHECK: .functype f32libcalls (f32) -> (f32) +; CHECK-NEXT: # %bb.0: +; CHECK-NEXT: local.get $push9=, 0 +; CHECK-NEXT: call $push0=, sinf, $pop9 +; CHECK-NEXT: call $push1=, cosf, $pop0 +; CHECK-NEXT: call $push2=, tanf, $pop1 +; CHECK-NEXT: call $push3=, asinf, $pop2 +; CHECK-NEXT: call $push4=, acosf, $pop3 +; CHECK-NEXT: call $push5=, atanf, $pop4 +; CHECK-NEXT: call $push6=, sinhf, $pop5 +; CHECK-NEXT: call $push7=, coshf, $pop6 +; CHECK-NEXT: call $push8=, tanhf, $pop7 +; CHECK-NEXT: return $pop8 + + + %k = call float @llvm.sin.f32(float %x) + %a = call float @llvm.cos.f32(float %k) + %b = call float @llvm.tan.f32(float %a) + %c = call float @llvm.asin.f32(float %b) + %d = call float @llvm.acos.f32(float %c) + %e = call float @llvm.atan.f32(float %d) + %f = call float @llvm.sinh.f32(float %e) + %g = call float @llvm.cosh.f32(float %f) + %h = call float @llvm.tanh.f32(float %g) + ret float %h +}