[llvm] SystemZ: Add sincos intrinsic test (PR #147473)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Mon Jul 14 23:40:48 PDT 2025
https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/147473
>From fdb7d2f1c5f36fa7521af0fea2aab69524eb242c Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Tue, 8 Jul 2025 16:40:46 +0900
Subject: [PATCH 1/2] SystemZ: Add sincos intrinsic test
The ZOS run line is mostly broken. update_test_checks seems
to not work on it and I have no idea what I'm looking at here.
It's not obvious to me what the calls are. I added some checks
for the references to the libcalls printed at the end of the module,
but didn't check anything in the function body. half also just
asserts somewhere.
---
llvm/test/CodeGen/SystemZ/llvm.sincos.ll | 222 +++++++++++++++++++++++
1 file changed, 222 insertions(+)
create mode 100644 llvm/test/CodeGen/SystemZ/llvm.sincos.ll
diff --git a/llvm/test/CodeGen/SystemZ/llvm.sincos.ll b/llvm/test/CodeGen/SystemZ/llvm.sincos.ll
new file mode 100644
index 0000000000000..1ff29ab7c4e8b
--- /dev/null
+++ b/llvm/test/CodeGen/SystemZ/llvm.sincos.ll
@@ -0,0 +1,222 @@
+; RUN: llc -mtriple=s390x-linux-gnu < %s | FileCheck -check-prefix=LINUX %s
+; RUN: llc -mtriple=s390x-ibm-zos < %s | FileCheck -check-prefix=ZOS %s
+
+; FIXME: Check ZOS function content
+
+; FIXME: half broken
+; define { half, half } @test_sincos_f16(half %a) #0 {
+; %result = call { half, half } @llvm.sincos.f16(half %a)
+; ret { half, half } %result
+; }
+
+; define half @test_sincos_f16_only_use_sin(half %a) #0 {
+; %result = call { half, half } @llvm.sincos.f16(half %a)
+; %result.0 = extractvalue { half, half } %result, 0
+; ret half %result.0
+; }
+
+; define half @test_sincos_f16_only_use_cos(half %a) #0 {
+; %result = call { half, half } @llvm.sincos.f16(half %a)
+; %result.1 = extractvalue { half, half } %result, 1
+; ret half %result.1
+; }
+
+; define { <2 x half>, <2 x half> } @test_sincos_v2f16(<2 x half> %a) #0 {
+; %result = call { <2 x half>, <2 x half> } @llvm.sincos.v2f16(<2 x half> %a)
+; ret { <2 x half>, <2 x half> } %result
+; }
+
+define { float, float } @test_sincos_f32(float %a) #0 {
+; LINUX-LABEL: test_sincos_f32:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r14, %r15, 112(%r15)
+; LINUX-NEXT: aghi %r15, -168
+; LINUX-NEXT: la %r2, 164(%r15)
+; LINUX-NEXT: la %r3, 160(%r15)
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: le %f0, 164(%r15)
+; LINUX-NEXT: le %f2, 160(%r15)
+; LINUX-NEXT: lmg %r14, %r15, 280(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { float, float } @llvm.sincos.f32(float %a)
+ ret { float, float } %result
+}
+
+define { <2 x float>, <2 x float> } @test_sincos_v2f32(<2 x float> %a) #0 {
+; LINUX-LABEL: test_sincos_v2f32:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r14, %r15, 112(%r15)
+; LINUX-NEXT: aghi %r15, -184
+; LINUX-NEXT: std %f8, 176(%r15) # 8-byte Spill
+; LINUX-NEXT: la %r2, 164(%r15)
+; LINUX-NEXT: la %r3, 160(%r15)
+; LINUX-NEXT: ler %f8, %f2
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: la %r2, 172(%r15)
+; LINUX-NEXT: la %r3, 168(%r15)
+; LINUX-NEXT: ler %f0, %f8
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: le %f0, 164(%r15)
+; LINUX-NEXT: le %f2, 172(%r15)
+; LINUX-NEXT: le %f4, 160(%r15)
+; LINUX-NEXT: le %f6, 168(%r15)
+; LINUX-NEXT: ld %f8, 176(%r15) # 8-byte Reload
+; LINUX-NEXT: lmg %r14, %r15, 296(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { <2 x float>, <2 x float> } @llvm.sincos.v2f32(<2 x float> %a)
+ ret { <2 x float>, <2 x float> } %result
+}
+
+define { <3 x float>, <3 x float> } @test_sincos_v3f32(<3 x float> %a) #0 {
+; LINUX-LABEL: test_sincos_v3f32:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r13, %r15, 104(%r15)
+; LINUX-NEXT: aghi %r15, -192
+; LINUX-NEXT: std %f8, 184(%r15) # 8-byte Spill
+; LINUX-NEXT: std %f9, 176(%r15) # 8-byte Spill
+; LINUX-NEXT: lgr %r13, %r2
+; LINUX-NEXT: la %r2, 164(%r15)
+; LINUX-NEXT: la %r3, 160(%r15)
+; LINUX-NEXT: ler %f8, %f4
+; LINUX-NEXT: ler %f9, %f2
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: la %r2, 172(%r15)
+; LINUX-NEXT: la %r3, 168(%r15)
+; LINUX-NEXT: ler %f0, %f9
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: la %r2, 8(%r13)
+; LINUX-NEXT: la %r3, 24(%r13)
+; LINUX-NEXT: ler %f0, %f8
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: le %f0, 164(%r15)
+; LINUX-NEXT: le %f1, 172(%r15)
+; LINUX-NEXT: le %f2, 160(%r15)
+; LINUX-NEXT: lgdr %r0, %f0
+; LINUX-NEXT: lgdr %r1, %f1
+; LINUX-NEXT: lgdr %r2, %f2
+; LINUX-NEXT: le %f0, 168(%r15)
+; LINUX-NEXT: nilf %r0, 0
+; LINUX-NEXT: srlg %r1, %r1, 32
+; LINUX-NEXT: nilf %r2, 0
+; LINUX-NEXT: lgdr %r3, %f0
+; LINUX-NEXT: srlg %r3, %r3, 32
+; LINUX-NEXT: lr %r0, %r1
+; LINUX-NEXT: lr %r2, %r3
+; LINUX-NEXT: stg %r2, 16(%r13)
+; LINUX-NEXT: stg %r0, 0(%r13)
+; LINUX-NEXT: ld %f8, 184(%r15) # 8-byte Reload
+; LINUX-NEXT: ld %f9, 176(%r15) # 8-byte Reload
+; LINUX-NEXT: lmg %r13, %r15, 296(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { <3 x float>, <3 x float> } @llvm.sincos.v3f32(<3 x float> %a)
+ ret { <3 x float>, <3 x float> } %result
+}
+
+define { double, double } @test_sincos_f64(double %a) #0 {
+; LINUX-LABEL: test_sincos_f64:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r14, %r15, 112(%r15)
+; LINUX-NEXT: aghi %r15, -176
+; LINUX-NEXT: la %r2, 168(%r15)
+; LINUX-NEXT: la %r3, 160(%r15)
+; LINUX-NEXT: brasl %r14, sincos at PLT
+; LINUX-NEXT: ld %f0, 168(%r15)
+; LINUX-NEXT: ld %f2, 160(%r15)
+; LINUX-NEXT: lmg %r14, %r15, 288(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { double, double } @llvm.sincos.f64(double %a)
+ ret { double, double } %result
+}
+
+define { <2 x double>, <2 x double> } @test_sincos_v2f64(<2 x double> %a) #0 {
+; LINUX-LABEL: test_sincos_v2f64:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r14, %r15, 112(%r15)
+; LINUX-NEXT: aghi %r15, -200
+; LINUX-NEXT: std %f8, 192(%r15) # 8-byte Spill
+; LINUX-NEXT: la %r2, 168(%r15)
+; LINUX-NEXT: la %r3, 160(%r15)
+; LINUX-NEXT: ldr %f8, %f2
+; LINUX-NEXT: brasl %r14, sincos at PLT
+; LINUX-NEXT: la %r2, 184(%r15)
+; LINUX-NEXT: la %r3, 176(%r15)
+; LINUX-NEXT: ldr %f0, %f8
+; LINUX-NEXT: brasl %r14, sincos at PLT
+; LINUX-NEXT: ld %f0, 168(%r15)
+; LINUX-NEXT: ld %f2, 184(%r15)
+; LINUX-NEXT: ld %f4, 160(%r15)
+; LINUX-NEXT: ld %f6, 176(%r15)
+; LINUX-NEXT: ld %f8, 192(%r15) # 8-byte Reload
+; LINUX-NEXT: lmg %r14, %r15, 312(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { <2 x double>, <2 x double> } @llvm.sincos.v2f64(<2 x double> %a)
+ ret { <2 x double>, <2 x double> } %result
+}
+
+define { fp128, fp128 } @test_sincos_f128(fp128 %a) #0 {
+; LINUX-LABEL: test_sincos_f128:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r14, %r15, 112(%r15)
+; LINUX-NEXT: aghi %r15, -176
+; LINUX-NEXT: ld %f0, 0(%r3)
+; LINUX-NEXT: ld %f2, 8(%r3)
+; LINUX-NEXT: lgr %r3, %r2
+; LINUX-NEXT: la %r4, 16(%r2)
+; LINUX-NEXT: la %r2, 160(%r15)
+; LINUX-NEXT: std %f0, 160(%r15)
+; LINUX-NEXT: std %f2, 168(%r15)
+; LINUX-NEXT: brasl %r14, sincosl at PLT
+; LINUX-NEXT: lmg %r14, %r15, 288(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { fp128, fp128 } @llvm.sincos.f128(fp128 %a)
+ ret { fp128, fp128 } %result
+}
+
+define { <2 x fp128>, <2 x fp128> } @test_sincos_v2f128(<2 x fp128> %a) #0 {
+; LINUX-LABEL: test_sincos_v2f128:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r13, %r15, 104(%r15)
+; LINUX-NEXT: aghi %r15, -208
+; LINUX-NEXT: std %f8, 200(%r15) # 8-byte Spill
+; LINUX-NEXT: std %f10, 192(%r15) # 8-byte Spill
+; LINUX-NEXT: lgr %r13, %r2
+; LINUX-NEXT: ld %f8, 0(%r3)
+; LINUX-NEXT: ld %f10, 8(%r3)
+; LINUX-NEXT: ld %f0, 16(%r3)
+; LINUX-NEXT: ld %f2, 24(%r3)
+; LINUX-NEXT: la %r3, 16(%r2)
+; LINUX-NEXT: la %r4, 48(%r2)
+; LINUX-NEXT: la %r2, 176(%r15)
+; LINUX-NEXT: std %f0, 176(%r15)
+; LINUX-NEXT: std %f2, 184(%r15)
+; LINUX-NEXT: brasl %r14, sincosl at PLT
+; LINUX-NEXT: la %r4, 32(%r13)
+; LINUX-NEXT: la %r2, 160(%r15)
+; LINUX-NEXT: std %f8, 160(%r15)
+; LINUX-NEXT: std %f10, 168(%r15)
+; LINUX-NEXT: lgr %r3, %r13
+; LINUX-NEXT: brasl %r14, sincosl at PLT
+; LINUX-NEXT: ld %f8, 200(%r15) # 8-byte Reload
+; LINUX-NEXT: ld %f10, 192(%r15) # 8-byte Reload
+; LINUX-NEXT: lmg %r13, %r15, 312(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { <2 x fp128>, <2 x fp128> } @llvm.sincos.v2f128(<2 x fp128> %a)
+ ret { <2 x fp128>, <2 x fp128> } %result
+}
+
+
+; ZOS: .quad R(@@FSIN at B) * Offset 0 function descriptor of @@FSIN at B
+; ZOS: .quad V(@@FSIN at B)
+; ZOS: .quad R(@@FCOS at B) * Offset 16 function descriptor of @@FCOS at B
+; ZOS: .quad V(@@FCOS at B)
+; ZOS: .quad R(@@SSIN at B) * Offset 32 function descriptor of @@SSIN at B
+; ZOS: .quad V(@@SSIN at B)
+; ZOS: .quad R(@@SCOS at B) * Offset 48 function descriptor of @@SCOS at B
+; ZOS: .quad V(@@SCOS at B)
+; ZOS: .quad R(@@LSIN at B) * Offset 64 function descriptor of @@LSIN at B
+; ZOS: .quad V(@@LSIN at B)
+; ZOS: .quad R(@@LCOS at B) * Offset 80 function descriptor of @@LCOS at B
+; ZOS: .quad V(@@LCOS at B)
+
+
+attributes #0 = { nounwind }
>From e6720bf63fdb5929ac7ff838fa0057a97c3f27d2 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Tue, 15 Jul 2025 15:40:31 +0900
Subject: [PATCH 2/2] Split out f16 tests
---
llvm/test/CodeGen/SystemZ/llvm.sincos.f16.ll | 126 +++++++++++++++++++
llvm/test/CodeGen/SystemZ/llvm.sincos.ll | 25 +---
2 files changed, 128 insertions(+), 23 deletions(-)
create mode 100644 llvm/test/CodeGen/SystemZ/llvm.sincos.f16.ll
diff --git a/llvm/test/CodeGen/SystemZ/llvm.sincos.f16.ll b/llvm/test/CodeGen/SystemZ/llvm.sincos.f16.ll
new file mode 100644
index 0000000000000..116320723ccd4
--- /dev/null
+++ b/llvm/test/CodeGen/SystemZ/llvm.sincos.f16.ll
@@ -0,0 +1,126 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=s390x-linux-gnu < %s | FileCheck -check-prefix=LINUX %s
+
+; FIXME: Merge with llvm.sincos.ll when zos is fixed for half.
+
+define { half, half } @test_sincos_f16(half %a) #0 {
+; LINUX-LABEL: test_sincos_f16:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r14, %r15, 112(%r15)
+; LINUX-NEXT: .cfi_offset %r14, -48
+; LINUX-NEXT: .cfi_offset %r15, -40
+; LINUX-NEXT: aghi %r15, -176
+; LINUX-NEXT: .cfi_def_cfa_offset 336
+; LINUX-NEXT: std %f8, 168(%r15) # 8-byte Spill
+; LINUX-NEXT: .cfi_offset %f8, -168
+; LINUX-NEXT: brasl %r14, __extendhfsf2 at PLT
+; LINUX-NEXT: la %r2, 164(%r15)
+; LINUX-NEXT: la %r3, 160(%r15)
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: le %f0, 164(%r15)
+; LINUX-NEXT: brasl %r14, __truncsfhf2 at PLT
+; LINUX-NEXT: le %f1, 160(%r15)
+; LINUX-NEXT: ler %f8, %f0
+; LINUX-NEXT: ler %f0, %f1
+; LINUX-NEXT: brasl %r14, __truncsfhf2 at PLT
+; LINUX-NEXT: ler %f2, %f0
+; LINUX-NEXT: ler %f0, %f8
+; LINUX-NEXT: ld %f8, 168(%r15) # 8-byte Reload
+; LINUX-NEXT: lmg %r14, %r15, 288(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { half, half } @llvm.sincos.f16(half %a)
+ ret { half, half } %result
+}
+
+define half @test_sincos_f16_only_use_sin(half %a) #0 {
+; LINUX-LABEL: test_sincos_f16_only_use_sin:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r14, %r15, 112(%r15)
+; LINUX-NEXT: .cfi_offset %r14, -48
+; LINUX-NEXT: .cfi_offset %r15, -40
+; LINUX-NEXT: aghi %r15, -168
+; LINUX-NEXT: .cfi_def_cfa_offset 328
+; LINUX-NEXT: brasl %r14, __extendhfsf2 at PLT
+; LINUX-NEXT: la %r2, 160(%r15)
+; LINUX-NEXT: la %r3, 164(%r15)
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: le %f0, 160(%r15)
+; LINUX-NEXT: brasl %r14, __truncsfhf2 at PLT
+; LINUX-NEXT: lmg %r14, %r15, 280(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { half, half } @llvm.sincos.f16(half %a)
+ %result.0 = extractvalue { half, half } %result, 0
+ ret half %result.0
+}
+
+define half @test_sincos_f16_only_use_cos(half %a) #0 {
+; LINUX-LABEL: test_sincos_f16_only_use_cos:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r14, %r15, 112(%r15)
+; LINUX-NEXT: .cfi_offset %r14, -48
+; LINUX-NEXT: .cfi_offset %r15, -40
+; LINUX-NEXT: aghi %r15, -168
+; LINUX-NEXT: .cfi_def_cfa_offset 328
+; LINUX-NEXT: brasl %r14, __extendhfsf2 at PLT
+; LINUX-NEXT: la %r2, 164(%r15)
+; LINUX-NEXT: la %r3, 160(%r15)
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: le %f0, 160(%r15)
+; LINUX-NEXT: brasl %r14, __truncsfhf2 at PLT
+; LINUX-NEXT: lmg %r14, %r15, 280(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { half, half } @llvm.sincos.f16(half %a)
+ %result.1 = extractvalue { half, half } %result, 1
+ ret half %result.1
+}
+
+define { <2 x half>, <2 x half> } @test_sincos_v2f16(<2 x half> %a) #0 {
+; LINUX-LABEL: test_sincos_v2f16:
+; LINUX: # %bb.0:
+; LINUX-NEXT: stmg %r14, %r15, 112(%r15)
+; LINUX-NEXT: .cfi_offset %r14, -48
+; LINUX-NEXT: .cfi_offset %r15, -40
+; LINUX-NEXT: aghi %r15, -200
+; LINUX-NEXT: .cfi_def_cfa_offset 360
+; LINUX-NEXT: std %f8, 192(%r15) # 8-byte Spill
+; LINUX-NEXT: std %f9, 184(%r15) # 8-byte Spill
+; LINUX-NEXT: std %f10, 176(%r15) # 8-byte Spill
+; LINUX-NEXT: .cfi_offset %f8, -168
+; LINUX-NEXT: .cfi_offset %f9, -176
+; LINUX-NEXT: .cfi_offset %f10, -184
+; LINUX-NEXT: ler %f8, %f2
+; LINUX-NEXT: brasl %r14, __extendhfsf2 at PLT
+; LINUX-NEXT: la %r2, 172(%r15)
+; LINUX-NEXT: la %r3, 168(%r15)
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: ler %f0, %f8
+; LINUX-NEXT: brasl %r14, __extendhfsf2 at PLT
+; LINUX-NEXT: la %r2, 164(%r15)
+; LINUX-NEXT: la %r3, 160(%r15)
+; LINUX-NEXT: brasl %r14, sincosf at PLT
+; LINUX-NEXT: le %f0, 172(%r15)
+; LINUX-NEXT: brasl %r14, __truncsfhf2 at PLT
+; LINUX-NEXT: le %f1, 164(%r15)
+; LINUX-NEXT: ler %f8, %f0
+; LINUX-NEXT: ler %f0, %f1
+; LINUX-NEXT: brasl %r14, __truncsfhf2 at PLT
+; LINUX-NEXT: le %f1, 168(%r15)
+; LINUX-NEXT: ler %f9, %f0
+; LINUX-NEXT: ler %f0, %f1
+; LINUX-NEXT: brasl %r14, __truncsfhf2 at PLT
+; LINUX-NEXT: le %f1, 160(%r15)
+; LINUX-NEXT: ler %f10, %f0
+; LINUX-NEXT: ler %f0, %f1
+; LINUX-NEXT: brasl %r14, __truncsfhf2 at PLT
+; LINUX-NEXT: ler %f6, %f0
+; LINUX-NEXT: ler %f0, %f8
+; LINUX-NEXT: ler %f2, %f9
+; LINUX-NEXT: ler %f4, %f10
+; LINUX-NEXT: ld %f8, 192(%r15) # 8-byte Reload
+; LINUX-NEXT: ld %f9, 184(%r15) # 8-byte Reload
+; LINUX-NEXT: ld %f10, 176(%r15) # 8-byte Reload
+; LINUX-NEXT: lmg %r14, %r15, 312(%r15)
+; LINUX-NEXT: br %r14
+ %result = call { <2 x half>, <2 x half> } @llvm.sincos.v2f16(<2 x half> %a)
+ ret { <2 x half>, <2 x half> } %result
+}
diff --git a/llvm/test/CodeGen/SystemZ/llvm.sincos.ll b/llvm/test/CodeGen/SystemZ/llvm.sincos.ll
index 1ff29ab7c4e8b..97980770cc7c9 100644
--- a/llvm/test/CodeGen/SystemZ/llvm.sincos.ll
+++ b/llvm/test/CodeGen/SystemZ/llvm.sincos.ll
@@ -1,30 +1,9 @@
; RUN: llc -mtriple=s390x-linux-gnu < %s | FileCheck -check-prefix=LINUX %s
; RUN: llc -mtriple=s390x-ibm-zos < %s | FileCheck -check-prefix=ZOS %s
-; FIXME: Check ZOS function content
-
-; FIXME: half broken
-; define { half, half } @test_sincos_f16(half %a) #0 {
-; %result = call { half, half } @llvm.sincos.f16(half %a)
-; ret { half, half } %result
-; }
-
-; define half @test_sincos_f16_only_use_sin(half %a) #0 {
-; %result = call { half, half } @llvm.sincos.f16(half %a)
-; %result.0 = extractvalue { half, half } %result, 0
-; ret half %result.0
-; }
+; FIXME: half cases split out since they are broken on zos
-; define half @test_sincos_f16_only_use_cos(half %a) #0 {
-; %result = call { half, half } @llvm.sincos.f16(half %a)
-; %result.1 = extractvalue { half, half } %result, 1
-; ret half %result.1
-; }
-
-; define { <2 x half>, <2 x half> } @test_sincos_v2f16(<2 x half> %a) #0 {
-; %result = call { <2 x half>, <2 x half> } @llvm.sincos.v2f16(<2 x half> %a)
-; ret { <2 x half>, <2 x half> } %result
-; }
+; FIXME: Check ZOS function content
define { float, float } @test_sincos_f32(float %a) #0 {
; LINUX-LABEL: test_sincos_f32:
More information about the llvm-commits
mailing list