SVML support for log10, sqrt

Although LLVM supports vectorization of loops containing log10/sqrt, it did not support using SVML implementation of it. Added support so that when clang is invoked with -fveclib=SVML now an appropriate SVML library log2 implementation will be invoked.

Follow up on: https://reviews.llvm.org/D77114

Tests:
Added unit tests to svml-calls.ll, svml-calls-finite.ll. Can be run with llvm-lint.
Created a simple c++ file that tests log10/sqrt, and used clang+ to build it, and output final assembly.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D87169

diff --git a/llvm/include/llvm/Analysis/VecFuncs.def b/llvm/include/llvm/Analysis/VecFuncs.def
index 9fdbf63..a47ee3c 100644 (file)
--- a/llvm/include/llvm/Analysis/VecFuncs.def
+++ b/llvm/include/llvm/Analysis/VecFuncs.def
@@ -269,6 +269,54 @@ TLI_DEFINE_VECFUNC("llvm.log2.f32", "__svml_log2f4", 4)
 TLI_DEFINE_VECFUNC("llvm.log2.f32", "__svml_log2f8", 8)
 TLI_DEFINE_VECFUNC("llvm.log2.f32", "__svml_log2f16", 16)
 
+TLI_DEFINE_VECFUNC("log10", "__svml_log102", 2)
+TLI_DEFINE_VECFUNC("log10", "__svml_log104", 4)
+TLI_DEFINE_VECFUNC("log10", "__svml_log108", 8)
+
+TLI_DEFINE_VECFUNC("log10f", "__svml_log10f4", 4)
+TLI_DEFINE_VECFUNC("log10f", "__svml_log10f8", 8)
+TLI_DEFINE_VECFUNC("log10f", "__svml_log10f16", 16)
+
+TLI_DEFINE_VECFUNC("__log10_finite", "__svml_log102", 2)
+TLI_DEFINE_VECFUNC("__log10_finite", "__svml_log104", 4)
+TLI_DEFINE_VECFUNC("__log10_finite", "__svml_log108", 8)
+
+TLI_DEFINE_VECFUNC("__log10f_finite", "__svml_log10f4", 4)
+TLI_DEFINE_VECFUNC("__log10f_finite", "__svml_log10f8", 8)
+TLI_DEFINE_VECFUNC("__log10f_finite", "__svml_log10f16", 16)
+
+TLI_DEFINE_VECFUNC("llvm.log10.f64", "__svml_log102", 2)
+TLI_DEFINE_VECFUNC("llvm.log10.f64", "__svml_log104", 4)
+TLI_DEFINE_VECFUNC("llvm.log10.f64", "__svml_log108", 8)
+
+TLI_DEFINE_VECFUNC("llvm.log10.f32", "__svml_log10f4", 4)
+TLI_DEFINE_VECFUNC("llvm.log10.f32", "__svml_log10f8", 8)
+TLI_DEFINE_VECFUNC("llvm.log10.f32", "__svml_log10f16", 16)
+
+TLI_DEFINE_VECFUNC("sqrt", "__svml_sqrt2", 2)
+TLI_DEFINE_VECFUNC("sqrt", "__svml_sqrt4", 4)
+TLI_DEFINE_VECFUNC("sqrt", "__svml_sqrt8", 8)
+
+TLI_DEFINE_VECFUNC("sqrtf", "__svml_sqrtf4", 4)
+TLI_DEFINE_VECFUNC("sqrtf", "__svml_sqrtf8", 8)
+TLI_DEFINE_VECFUNC("sqrtf", "__svml_sqrtf16", 16)
+
+TLI_DEFINE_VECFUNC("__sqrt_finite", "__svml_sqrt2", 2)
+TLI_DEFINE_VECFUNC("__sqrt_finite", "__svml_sqrt4", 4)
+TLI_DEFINE_VECFUNC("__sqrt_finite", "__svml_sqrt8", 8)
+
+TLI_DEFINE_VECFUNC("__sqrtf_finite", "__svml_sqrtf4", 4)
+TLI_DEFINE_VECFUNC("__sqrtf_finite", "__svml_sqrtf8", 8)
+TLI_DEFINE_VECFUNC("__sqrtf_finite", "__svml_sqrtf16", 16)
+
+TLI_DEFINE_VECFUNC("llvm.sqrt.f64", "__svml_sqrt2", 2)
+TLI_DEFINE_VECFUNC("llvm.sqrt.f64", "__svml_sqrt4", 4)
+TLI_DEFINE_VECFUNC("llvm.sqrt.f64", "__svml_sqrt8", 8)
+
+TLI_DEFINE_VECFUNC("llvm.sqrt.f32", "__svml_sqrtf4", 4)
+TLI_DEFINE_VECFUNC("llvm.sqrt.f32", "__svml_sqrtf8", 8)
+TLI_DEFINE_VECFUNC("llvm.sqrt.f32", "__svml_sqrtf16", 16)
+
 TLI_DEFINE_VECFUNC("exp2", "__svml_exp22", 2)
 TLI_DEFINE_VECFUNC("exp2", "__svml_exp24", 4)
 TLI_DEFINE_VECFUNC("exp2", "__svml_exp28", 8)

diff --git a/llvm/test/Transforms/LoopVectorize/X86/svml-calls-finite.ll b/llvm/test/Transforms/LoopVectorize/X86/svml-calls-finite.ll
index dd6692d..a6e191c 100644 (file)
--- a/llvm/test/Transforms/LoopVectorize/X86/svml-calls-finite.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/svml-calls-finite.ll
@@ -300,3 +300,117 @@ for.end:                                          ; preds = %for.body
 !91 = distinct !{!31, !32, !33}
 !92 = !{!"llvm.loop.vectorize.width", i32 4}
 !93 = !{!"llvm.loop.vectorize.enable", i1 true}
+
+declare float @__log10f_finite(float) #0
+
+; CHECK-LABEL: @log10_f32
+; CHECK: <4 x float> @__svml_log10f4
+; CHECK: ret
+define void @log10_f32(float* nocapture %varray) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call fast float @__log10f_finite(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %indvars.iv
+  store float %call, float* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !21
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!101 = distinct !{!21, !22, !23}
+!102 = !{!"llvm.loop.vectorize.width", i32 4}
+!103 = !{!"llvm.loop.vectorize.enable", i1 true}
+
+
+declare double @__log10_finite(double) #0
+
+; CHECK-LABEL: @log10_f64
+; CHECK: <4 x double> @__svml_log104
+; CHECK: ret
+define void @log10_f64(double* nocapture %varray) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call fast double @__log10_finite(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %indvars.iv
+  store double %call, double* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !31
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!111 = distinct !{!31, !32, !33}
+!112 = !{!"llvm.loop.vectorize.width", i32 4}
+!113 = !{!"llvm.loop.vectorize.enable", i1 true}
+
+declare float @__sqrtf_finite(float) #0
+
+; CHECK-LABEL: @sqrt_f32
+; CHECK: <4 x float> @__svml_sqrtf4
+; CHECK: ret
+define void @sqrt_f32(float* nocapture %varray) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call fast float @__sqrtf_finite(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %indvars.iv
+  store float %call, float* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !21
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!121 = distinct !{!21, !22, !23}
+!122 = !{!"llvm.loop.vectorize.width", i32 4}
+!123 = !{!"llvm.loop.vectorize.enable", i1 true}
+
+
+declare double @__sqrt_finite(double) #0
+
+; CHECK-LABEL: @sqrt_f64
+; CHECK: <4 x double> @__svml_sqrt4
+; CHECK: ret
+define void @sqrt_f64(double* nocapture %varray) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %tmp = trunc i64 %indvars.iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call fast double @__sqrt_finite(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %indvars.iv
+  store double %call, double* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !31
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+!131 = distinct !{!31, !32, !33}
+!132 = !{!"llvm.loop.vectorize.width", i32 4}
+!133 = !{!"llvm.loop.vectorize.enable", i1 true}

diff --git a/llvm/test/Transforms/LoopVectorize/X86/svml-calls.ll b/llvm/test/Transforms/LoopVectorize/X86/svml-calls.ll
index c074830..da6b469 100644 (file)
--- a/llvm/test/Transforms/LoopVectorize/X86/svml-calls.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/svml-calls.ll
@@ -33,6 +33,16 @@ declare float @log2f(float) #0
 declare double @llvm.log2.f64(double) #0
 declare float @llvm.log2.f32(float) #0
 
+declare double @log10(double) #0
+declare float @log10f(float) #0
+declare double @llvm.log10.f64(double) #0
+declare float @llvm.log10.f32(float) #0
+
+declare double @sqrt(double) #0
+declare float @sqrtf(float) #0
+declare double @llvm.sqrt.f64(double) #0
+declare float @llvm.sqrt.f32(float) #0
+
 declare double @exp2(double) #0
 declare float @exp2f(float) #0
 declare double @llvm.exp2.f64(double) #0
@@ -598,6 +608,190 @@ for.end:
   ret void
 }
 
+define void @log10_f64(double* nocapture %varray) {
+; CHECK-LABEL: @log10_f64(
+; CHECK:    [[TMP5:%.*]] = call <4 x double> @__svml_log104(<4 x double> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call double @log10(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %iv
+  store double %call, double* %arrayidx, align 4
+  %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 @log10_f32(float* nocapture %varray) {
+; CHECK-LABEL: @log10_f32(
+; CHECK:    [[TMP5:%.*]] = call <4 x float> @__svml_log10f4(<4 x float> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call float @log10f(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %iv
+  store float %call, float* %arrayidx, align 4
+  %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 @log10_f64_intrinsic(double* nocapture %varray) {
+; CHECK-LABEL: @log10_f64_intrinsic(
+; CHECK:    [[TMP5:%.*]] = call <4 x double> @__svml_log104(<4 x double> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call double @llvm.log10.f64(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %iv
+  store double %call, double* %arrayidx, align 4
+  %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 @log10_f32_intrinsic(float* nocapture %varray) {
+; CHECK-LABEL: @log10_f32_intrinsic(
+; CHECK:    [[TMP5:%.*]] = call <4 x float> @__svml_log10f4(<4 x float> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call float @llvm.log10.f32(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %iv
+  store float %call, float* %arrayidx, align 4
+  %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 @sqrt_f64(double* nocapture %varray) {
+; CHECK-LABEL: @sqrt_f64(
+; CHECK:    [[TMP5:%.*]] = call <4 x double> @__svml_sqrt4(<4 x double> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call double @sqrt(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %iv
+  store double %call, double* %arrayidx, align 4
+  %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 @sqrt_f32(float* nocapture %varray) {
+; CHECK-LABEL: @sqrt_f32(
+; CHECK:    [[TMP5:%.*]] = call <4 x float> @__svml_sqrtf4(<4 x float> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call float @sqrtf(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %iv
+  store float %call, float* %arrayidx, align 4
+  %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 @sqrt_f64_intrinsic(double* nocapture %varray) {
+; CHECK-LABEL: @sqrt_f64_intrinsic(
+; CHECK:    [[TMP5:%.*]] = call <4 x double> @__svml_sqrt4(<4 x double> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to double
+  %call = tail call double @llvm.sqrt.f64(double %conv)
+  %arrayidx = getelementptr inbounds double, double* %varray, i64 %iv
+  store double %call, double* %arrayidx, align 4
+  %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 @sqrt_f32_intrinsic(float* nocapture %varray) {
+; CHECK-LABEL: @sqrt_f32_intrinsic(
+; CHECK:    [[TMP5:%.*]] = call <4 x float> @__svml_sqrtf4(<4 x float> [[TMP4:%.*]])
+; CHECK:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
+  %tmp = trunc i64 %iv to i32
+  %conv = sitofp i32 %tmp to float
+  %call = tail call float @llvm.sqrt.f32(float %conv)
+  %arrayidx = getelementptr inbounds float, float* %varray, i64 %iv
+  store float %call, float* %arrayidx, align 4
+  %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 @exp2_f64(double* nocapture %varray) {
 ; CHECK-LABEL: @exp2_f64(
 ; CHECK:    [[TMP5:%.*]] = call <4 x double> @__svml_exp24(<4 x double> [[TMP4:%.*]])

diff --git a/llvm/test/Transforms/Util/add-TLI-mappings.ll b/llvm/test/Transforms/Util/add-TLI-mappings.ll
index c68a9c9..75e3252 100644 (file)
--- a/llvm/test/Transforms/Util/add-TLI-mappings.ll
+++ b/llvm/test/Transforms/Util/add-TLI-mappings.ll
@@ -9,10 +9,13 @@ target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 ; COMMON-LABEL: @llvm.compiler.used = appending global
-; SVML-SAME:        [3 x i8*] [
+; SVML-SAME:        [6 x i8*] [
 ; SVML-SAME:          i8* bitcast (<2 x double> (<2 x double>)* @__svml_sin2 to i8*),
 ; SVML-SAME:          i8* bitcast (<4 x double> (<4 x double>)* @__svml_sin4 to i8*),
-; SVML-SAME:          i8* bitcast (<8 x double> (<8 x double>)* @__svml_sin8 to i8*)
+; SVML-SAME:          i8* bitcast (<8 x double> (<8 x double>)* @__svml_sin8 to i8*),
+; SVML-SAME:          i8* bitcast (<4 x float> (<4 x float>)* @__svml_log10f4 to i8*),
+; SVML-SAME:          i8* bitcast (<8 x float> (<8 x float>)* @__svml_log10f8 to i8*),
+; SVML-SAME:          i8* bitcast (<16 x float> (<16 x float>)* @__svml_log10f16 to i8*)
 ; MASSV-SAME:       [2 x i8*] [
 ; MASSV-SAME:         i8* bitcast (<2 x double> (<2 x double>)* @__sind2_massv to i8*),
 ; MASSV-SAME:         i8* bitcast (<4 x float> (<4 x float>)* @__log10f4_massv to i8*)