SystemZ: Add more tests for fp128 atomics (#90269)

These did not have proper floating point uses so weren't representative
samples. The bitcast inserted by lowering could be absorbed by the
load/store on the source/use.

Author: arsenm

llvm/test/CodeGen/SystemZ/atomic-load-08.ll

@@ -2,8 +2,8 @@
 ; loads with a bitcast, and this test case gets converted into that form as
 ; well by the AtomicExpand pass.
 ;
-; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
-; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
+; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck -check-prefixes=CHECK,BASE %s
+; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck -check-prefixes=CHECK,Z13 %s
 
 define void @f1(ptr %ret, ptr %src) {
 ; CHECK-LABEL: f1:
@@ -17,10 +17,54 @@ define void @f1(ptr %ret, ptr %src) {
   ret void
 }
 
+define void @f1_fpuse(ptr %ret, ptr %src) {
+; CHECK-LABEL: f1_fpuse:
+; CHECK:       # %bb.0:
+; BASE-NEXT: aghi	%r15, -176
+; BASE-NEXT: .cfi_def_cfa_offset 336
+
+; CHECK-NEXT:	lpq	%r0, 0(%r3)
+
+; BASE-NEXT: stg %r1, 168(%r15)
+; BASE-NEXT: stg %r0, 160(%r15)
+; BASE-NEXT: ld	%f0, 160(%r15)
+; BASE-NEXT: ld	%f2, 168(%r15)
+
+; Z13-NEXT: vlvgp %v0, %r0, %r1
+; Z13-NEXT: vrepg %v2, %v0, 1
+
+; CHECK-NEXT:	axbr	%f0, %f0
+; CHECK-NEXT:	std	%f0, 0(%r2)
+; CHECK-NEXT:	std	%f2, 8(%r2)
+; BASE-NEXT:	aghi	%r15, 176
+; CHECK-NEXT:	br	%r14
+
+  %val = load atomic fp128, ptr %src seq_cst, align 16
+  %use = fadd fp128 %val, %val
+  store fp128 %use, ptr %ret, align 8
+  ret void
+}
+
 define void @f2(ptr %ret, ptr %src) {
 ; CHECK-LABEL: f2:
 ; CHECK: brasl %r14, __atomic_load@PLT
   %val = load atomic fp128, ptr %src seq_cst, align 8
   store fp128 %val, ptr %ret, align 8
   ret void
 }
+
+define void @f2_fpuse(ptr %ret, ptr %src) {
+; CHECK-LABEL: f2_fpuse:
+; CHECK: brasl %r14, __atomic_load@PLT
+; CHECK-NEXT:   ld	%f0, 160(%r15)
+; CHECK-NEXT:	ld	%f2, 168(%r15)
+; CHECK-NEXT:	axbr	%f0, %f0
+; CHECK-NEXT:	std	%f0, 0(%r13)
+; CHECK-NEXT:	std	%f2, 8(%r13)
+; CHECK-NEXT:	lmg	%r13, %r15, 280(%r15)
+; CHECK-NEXT:	br	%r14
+  %val = load atomic fp128, ptr %src seq_cst, align 8
+  %use = fadd fp128 %val, %val
+  store fp128 %use, ptr %ret, align 8
+  ret void
+}

llvm/test/CodeGen/SystemZ/atomic-store-08.ll

@@ -1,8 +1,8 @@
 ; Test long double atomic stores. The atomic store is converted to i128 by
 ; the AtomicExpand pass.
 ;
-; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
-; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
+; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck -check-prefixes=CHECK,BASE %s
+; xUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck -check-prefixes=CHECK,Z13 %s
 
 define void @f1(ptr %dst, ptr %src) {
 ; CHECK-LABEL: f1:
@@ -17,10 +17,57 @@ define void @f1(ptr %dst, ptr %src) {
   ret void
 }
 
+define void @f1_fpsrc(ptr %dst, ptr %src) {
+; CHECK-LABEL: f1_fpsrc:
+; CHECK:       # %bb.0:
+; CHECK-NEXT: ld	%f0, 0(%r3)
+; CHECK-NEXT: ld	%f2, 8(%r3)
+; CHECK-NEXT: axbr	%f0, %f0
+
+; BASE-NEXT: lgdr	%r1, %f2
+; BASE-NEXT: lgdr	%r0, %f0
+
+; Z13-NEXT: vmrhg	%v0, %v0, %v2
+; Z13-NEXT: vlgvg	%r1, %v0, 1
+; Z13-NEXT: vlgvg	%r0, %v0, 0
+
+; CHECK-NEXT: stpq	%r0, 0(%r2)
+; CHECK-NEXT: bcr	15, %r0
+; CHECK-NEXT: br	%r14
+  %val = load fp128, ptr %src, align 8
+  %add = fadd fp128 %val, %val
+  store atomic fp128 %add, ptr %dst seq_cst, align 16
+  ret void
+}
+
 define void @f2(ptr %dst, ptr %src) {
 ; CHECK-LABEL: f2:
 ; CHECK: brasl %r14, __atomic_store@PLT
   %val = load fp128, ptr %src, align 8
   store atomic fp128 %val, ptr %dst seq_cst, align 8
   ret void
 }
+
+define void @f2_fpuse(ptr %dst, ptr %src) {
+; CHECK-LABEL: f2_fpuse:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:	stmg	%r14, %r15, 112(%r15)
+; CHECK-NEXT:	.cfi_offset %r14, -48
+; CHECK-NEXT:	.cfi_offset %r15, -40
+; CHECK-NEXT:	aghi	%r15, -176
+; CHECK-NEXT:	.cfi_def_cfa_offset 336
+; CHECK-NEXT:	ld	%f0, 0(%r3)
+; CHECK-NEXT:	ld	%f2, 8(%r3)
+; CHECK-NEXT:	lgr	%r3, %r2
+; CHECK-NEXT:	axbr	%f0, %f0
+; CHECK-NEXT:	la	%r4, 160(%r15)
+; CHECK-NEXT:	lghi	%r2, 16
+; CHECK-NEXT:	lhi	%r5, 5
+; CHECK-NEXT:	std	%f0, 160(%r15)
+; CHECK-NEXT:	std	%f2, 168(%r15)
+; CHECK-NEXT: brasl %r14, __atomic_store@PLT
+  %val = load fp128, ptr %src, align 8
+  %add = fadd fp128 %val, %val
+  store atomic fp128 %add, ptr %dst seq_cst, align 8
+  ret void
+}

llvm/test/CodeGen/SystemZ/atomicrmw-xchg-07.ll

@@ -26,3 +26,45 @@ define void @f1(ptr align 16 %ret, ptr align 16 %src, ptr align 16 %b) {
   store fp128 %res, ptr %ret, align 16
   ret void
 }
+
+define void @f1_fpuse(ptr align 16 %ret, ptr align 16 %src, ptr align 16 %b) {
+; CHECK-LABEL: f1_fpuse:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    stmg %r12, %r15, 96(%r15)
+; CHECK-NEXT:    .cfi_offset %r12, -64
+; CHECK-NEXT:    .cfi_offset %r13, -56
+; CHECK-NEXT:    .cfi_offset %r15, -40
+; CHECK-NEXT:    aghi %r15, -176
+; CHECK-NEXT:    .cfi_def_cfa_offset 336
+; CHECK-NEXT:    ld %f0, 0(%r4)
+; CHECK-NEXT:    ld %f2, 8(%r4)
+; CHECK-NEXT:    lg %r0, 8(%r3)
+; CHECK-NEXT:    lg %r1, 0(%r3)
+; CHECK-NEXT:    axbr %f0, %f0
+; CHECK-NEXT:    lgdr %r5, %f2
+; CHECK-NEXT:    lgdr %r4, %f0
+; CHECK-NEXT:  .LBB1_1: # %atomicrmw.start
+; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT:    lgr %r12, %r1
+; CHECK-NEXT:    lgr %r13, %r0
+; CHECK-NEXT:    cdsg %r12, %r4, 0(%r3)
+; CHECK-NEXT:    lgr %r0, %r13
+; CHECK-NEXT:    lgr %r1, %r12
+; CHECK-NEXT:    jl .LBB1_1
+; CHECK-NEXT:  # %bb.2: # %atomicrmw.end
+; CHECK-NEXT:    stg %r1, 160(%r15)
+; CHECK-NEXT:    stg %r0, 168(%r15)
+; CHECK-NEXT:    ld %f0, 160(%r15)
+; CHECK-NEXT:    ld %f2, 168(%r15)
+; CHECK-NEXT:    axbr %f0, %f0
+; CHECK-NEXT:    std %f0, 0(%r2)
+; CHECK-NEXT:    std %f2, 8(%r2)
+; CHECK-NEXT:    lmg %r12, %r15, 272(%r15)
+; CHECK-NEXT:    br %r14
+  %val = load fp128, ptr %b, align 16
+  %add.src = fadd fp128 %val, %val
+  %res = atomicrmw xchg ptr %src, fp128 %add.src seq_cst
+  %res.x2 = fadd fp128 %res, %res
+  store fp128 %res.x2, ptr %ret, align 16
+  ret void
+}