[llvm] [RISCV][SLEEF]: Support SLEEF vector library for RISC-V target. (PR #114014)

[Mark Goncharov via llvm-commits]

================
@@ -0,0 +1,2541 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals --filter "call.*(cos|sin|tan|cbrt|erf|exp[^e]|gamma|log|sqrt|copysign|dim|min|mod|hypot|nextafter|pow|fma)" --version 2
+; RUN: opt -mattr=+v -vector-library=sleefgnuabi -passes=inject-tli-mappings,loop-vectorize,simplifycfg -S < %s | FileCheck %s
+
+target triple = "riscv64-unknown-linux-gnu"
+
+; We are checking whether loops containing function calls can be vectorized,
+; when the compiler provides TLI mappings to their vector variants.
+
+declare double @acos(double)
+declare float @acosf(float)
+
+;.
+; CHECK: @llvm.compiler.used = appending global [86 x ptr] [ptr @Sleef_acosdx_u10rvvm2, ptr @Sleef_acosfx_u10rvvm2, ptr @Sleef_acoshdx_u10rvvm2, ptr @Sleef_acoshfx_u10rvvm2, ptr @Sleef_asindx_u10rvvm2, ptr @Sleef_asinfx_u10rvvm2, ptr @Sleef_asinhdx_u10rvvm2, ptr @Sleef_asinhfx_u10rvvm2, ptr @Sleef_atandx_u10rvvm2, ptr @Sleef_atanfx_u10rvvm2, ptr @Sleef_atan2dx_u10rvvm2, ptr @Sleef_atan2fx_u10rvvm2, ptr @Sleef_atanhdx_u10rvvm2, ptr @Sleef_atanhfx_u10rvvm2, ptr @Sleef_cbrtdx_u10rvvm2, ptr @Sleef_cbrtfx_u10rvvm2, ptr @Sleef_copysigndx_rvvm2, ptr @Sleef_copysignfx_rvvm2, ptr @Sleef_cosdx_u10rvvm2, ptr @Sleef_cosfx_u10rvvm2, ptr @Sleef_coshdx_u10rvvm2, ptr @Sleef_coshfx_u10rvvm2, ptr @Sleef_cospidx_u05rvvm2, ptr @Sleef_cospifx_u05rvvm2, ptr @Sleef_erfdx_u10rvvm2, ptr @Sleef_erffx_u10rvvm2, ptr @Sleef_erfcdx_u15rvvm2, ptr @Sleef_erfcfx_u15rvvm2, ptr @Sleef_expdx_u10rvvm2, ptr @Sleef_expfx_u10rvvm2, ptr @Sleef_exp10dx_u10rvvm2, ptr @Sleef_exp10fx_u10rvvm2, ptr @Sleef_exp2dx_u10rvvm2, ptr @Sleef_exp2fx_u10rvvm2, ptr @Sleef_expm1dx_u10rvvm2, ptr @Sleef_expm1fx_u10rvvm2, ptr @Sleef_fdimdx_rvvm2, ptr @Sleef_fdimfx_rvvm2, ptr @Sleef_fmadx_rvvm2, ptr @Sleef_fmafx_rvvm2, ptr @Sleef_fmaxdx_rvvm2, ptr @Sleef_fmaxfx_rvvm2, ptr @Sleef_fmindx_u10rvvm2, ptr @Sleef_fminfx_u10rvvm2, ptr @Sleef_fmoddx_rvvm2, ptr @Sleef_fmodfx_rvvm2, ptr @Sleef_hypotdx_u05rvvm2, ptr @Sleef_hypotfx_u05rvvm2, ptr @Sleef_ilogbdx_rvvm2, ptr @Sleef_ilogbfx_rvvm2, ptr @Sleef_ldexpdx_rvvm2, ptr @Sleef_ldexpfx_rvvm2, ptr @Sleef_lgammadx_u10rvvm2, ptr @Sleef_lgammafx_u10rvvm2, ptr @Sleef_logdx_u10rvvm2, ptr @Sleef_logfx_u10rvvm2, ptr @Sleef_log10dx_u10rvvm2, ptr @Sleef_log10fx_u10rvvm2, ptr @Sleef_log1pdx_u10rvvm2, ptr @Sleef_log1pfx_u10rvvm2, ptr @Sleef_log2dx_u10rvvm2, ptr @Sleef_log2fx_u10rvvm2, ptr @Sleef_modfdx_rvvm2, ptr @Sleef_modffx_rvvm2, ptr @Sleef_nextafterdx_rvvm2, ptr @Sleef_nextafterfx_rvvm2, ptr @Sleef_powdx_u10rvvm2, ptr @Sleef_powfx_u10rvvm2, ptr @Sleef_sindx_u10rvvm2, ptr @Sleef_sinfx_u10rvvm2, ptr @Sleef_sincosdx_u10rvvm2, ptr @Sleef_sincosfx_u10rvvm2, ptr @Sleef_sincospidx_u10rvvm2, ptr @Sleef_sincospifx_u10rvvm2, ptr @Sleef_sinhdx_u10rvvm2, ptr @Sleef_sinhfx_u10rvvm2, ptr @Sleef_sinpidx_u05rvvm2, ptr @Sleef_sinpifx_u05rvvm2, ptr @Sleef_sqrtdx_u05rvvm2, ptr @Sleef_sqrtfx_u05rvvm2, ptr @Sleef_tandx_u10rvvm2, ptr @Sleef_tanfx_u10rvvm2, ptr @Sleef_tanhdx_u10rvvm2, ptr @Sleef_tanhfx_u10rvvm2, ptr @Sleef_tgammadx_u10rvvm2, ptr @Sleef_tgammafx_u10rvvm2], section "llvm.metadata"
+;.
+define void @acos_f64(ptr noalias %in.ptr, ptr noalias %out.ptr) {
+;
+; CHECK-LABEL: define void @acos_f64
+; CHECK-SAME: (ptr noalias [[IN_PTR:%.*]], ptr noalias [[OUT_PTR:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK:    [[TMP9:%.*]] = call <vscale x 2 x double> @Sleef_acosdx_u10rvvm2(<vscale x 2 x double> [[WIDE_LOAD:%.*]])
+; CHECK:    [[CALL:%.*]] = tail call double @acos(double [[IN:%.*]]) #[[ATTR2:[0-9]+]]
+;
+  entry:
+  br label %for.body
+
+  for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %in.gep = getelementptr inbounds double, ptr %in.ptr, i64 %iv
+  %in = load double, ptr %in.gep, align 8
+  %call = tail call double @acos(double %in)
+  %out.gep = getelementptr inbounds double, ptr %out.ptr, i64 %iv
+  store double %call, ptr %out.gep, align 8
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond = icmp eq i64 %iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+  for.end:
+  ret void
+}
+
+define void @acos_f32(ptr noalias %in.ptr, ptr noalias %out.ptr) {
+;
+; CHECK-LABEL: define void @acos_f32
+; CHECK-SAME: (ptr noalias [[IN_PTR:%.*]], ptr noalias [[OUT_PTR:%.*]]) #[[ATTR0]] {
+; CHECK:    [[TMP9:%.*]] = call <vscale x 4 x float> @Sleef_acosfx_u10rvvm2(<vscale x 4 x float> [[WIDE_LOAD:%.*]])
+; CHECK:    [[CALL:%.*]] = tail call float @acosf(float [[IN:%.*]]) #[[ATTR3:[0-9]+]]
+;
+  entry:
+  br label %for.body
+
+  for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %in.gep = getelementptr inbounds float, ptr %in.ptr, i64 %iv
+  %in = load float, ptr %in.gep, align 8
----------------
mga-sc wrote:

Not intentionally, copy pasted from ARM: ./llvm/test/Transforms/LoopVectorize/AArch64/veclib-function-calls.ll

https://github.com/llvm/llvm-project/pull/114014