mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2025-01-31 20:51:52 +01:00
bc658bf60a
This adds support for the new 128-bit vector float instructions of z14. Note that these instructions actually only operate on the f128 type, since only each 128-bit vector register can hold only one 128-bit float value. However, this is still preferable to the legacy 128-bit float instructions, since those operate on pairs of floating-point registers (so we can hold at most 8 values in registers), while the new instructions use single vector registers (so we hold up to 32 value in registers). Adding support includes: - Enabling the instructions for the assembler/disassembler. - CodeGen for the instructions. This includes allocating the f128 type now to the VR128BitRegClass instead of FP128BitRegClass. - Scheduler description support for the instructions. Note that for a small number of operations, we have no new vector instructions (like integer <-> 128-bit float conversions), and so we use the legacy instruction and then reformat the operand (i.e. copy between a pair of floating-point registers and a vector register). llvm-svn: 308196
73 lines
2.5 KiB
LLVM
73 lines
2.5 KiB
LLVM
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z14 | FileCheck %s
|
|
|
|
declare fp128 @llvm.fma.f128(fp128 %f1, fp128 %f2, fp128 %f3)
|
|
|
|
define void @f1(fp128 *%ptr1, fp128 *%ptr2, fp128 *%ptr3, fp128 *%dst) {
|
|
; CHECK-LABEL: f1:
|
|
; CHECK-DAG: vl [[REG1:%v[0-9]+]], 0(%r2)
|
|
; CHECK-DAG: vl [[REG2:%v[0-9]+]], 0(%r3)
|
|
; CHECK-DAG: vl [[REG3:%v[0-9]+]], 0(%r4)
|
|
; CHECK: wfmaxb [[RES:%v[0-9]+]], [[REG1]], [[REG2]], [[REG3]]
|
|
; CHECK: vst [[RES]], 0(%r5)
|
|
; CHECK: br %r14
|
|
%f1 = load fp128, fp128 *%ptr1
|
|
%f2 = load fp128, fp128 *%ptr2
|
|
%f3 = load fp128, fp128 *%ptr3
|
|
%res = call fp128 @llvm.fma.f128 (fp128 %f1, fp128 %f2, fp128 %f3)
|
|
store fp128 %res, fp128 *%dst
|
|
ret void
|
|
}
|
|
|
|
define void @f2(fp128 *%ptr1, fp128 *%ptr2, fp128 *%ptr3, fp128 *%dst) {
|
|
; CHECK-LABEL: f2:
|
|
; CHECK-DAG: vl [[REG1:%v[0-9]+]], 0(%r2)
|
|
; CHECK-DAG: vl [[REG2:%v[0-9]+]], 0(%r3)
|
|
; CHECK-DAG: vl [[REG3:%v[0-9]+]], 0(%r4)
|
|
; CHECK: wfmsxb [[RES:%v[0-9]+]], [[REG1]], [[REG2]], [[REG3]]
|
|
; CHECK: vst [[RES]], 0(%r5)
|
|
; CHECK: br %r14
|
|
%f1 = load fp128, fp128 *%ptr1
|
|
%f2 = load fp128, fp128 *%ptr2
|
|
%f3 = load fp128, fp128 *%ptr3
|
|
%neg = fsub fp128 0xL00000000000000008000000000000000, %f3
|
|
%res = call fp128 @llvm.fma.f128 (fp128 %f1, fp128 %f2, fp128 %neg)
|
|
store fp128 %res, fp128 *%dst
|
|
ret void
|
|
}
|
|
|
|
define void @f3(fp128 *%ptr1, fp128 *%ptr2, fp128 *%ptr3, fp128 *%dst) {
|
|
; CHECK-LABEL: f3:
|
|
; CHECK-DAG: vl [[REG1:%v[0-9]+]], 0(%r2)
|
|
; CHECK-DAG: vl [[REG2:%v[0-9]+]], 0(%r3)
|
|
; CHECK-DAG: vl [[REG3:%v[0-9]+]], 0(%r4)
|
|
; CHECK: wfnmaxb [[RES:%v[0-9]+]], [[REG1]], [[REG2]], [[REG3]]
|
|
; CHECK: vst [[RES]], 0(%r5)
|
|
; CHECK: br %r14
|
|
%f1 = load fp128, fp128 *%ptr1
|
|
%f2 = load fp128, fp128 *%ptr2
|
|
%f3 = load fp128, fp128 *%ptr3
|
|
%res = call fp128 @llvm.fma.f128 (fp128 %f1, fp128 %f2, fp128 %f3)
|
|
%negres = fsub fp128 0xL00000000000000008000000000000000, %res
|
|
store fp128 %negres, fp128 *%dst
|
|
ret void
|
|
}
|
|
|
|
define void @f4(fp128 *%ptr1, fp128 *%ptr2, fp128 *%ptr3, fp128 *%dst) {
|
|
; CHECK-LABEL: f4:
|
|
; CHECK-DAG: vl [[REG1:%v[0-9]+]], 0(%r2)
|
|
; CHECK-DAG: vl [[REG2:%v[0-9]+]], 0(%r3)
|
|
; CHECK-DAG: vl [[REG3:%v[0-9]+]], 0(%r4)
|
|
; CHECK: wfnmsxb [[RES:%v[0-9]+]], [[REG1]], [[REG2]], [[REG3]]
|
|
; CHECK: vst [[RES]], 0(%r5)
|
|
; CHECK: br %r14
|
|
%f1 = load fp128, fp128 *%ptr1
|
|
%f2 = load fp128, fp128 *%ptr2
|
|
%f3 = load fp128, fp128 *%ptr3
|
|
%neg = fsub fp128 0xL00000000000000008000000000000000, %f3
|
|
%res = call fp128 @llvm.fma.f128 (fp128 %f1, fp128 %f2, fp128 %neg)
|
|
%negres = fsub fp128 0xL00000000000000008000000000000000, %res
|
|
store fp128 %negres, fp128 *%dst
|
|
ret void
|
|
}
|
|
|