diff options
| author | Paul E. Murphy <murp@ibm.com> | 2024-05-13 11:56:42 -0500 |
|---|---|---|
| committer | Paul Murphy <murp@ibm.com> | 2024-05-15 13:55:28 +0000 |
| commit | 408739fc96f00a50642673544e803a3c3f3e27df (patch) | |
| tree | dfdefff56180661408202a77de0456af3324b47c /src/cmd/internal/obj | |
| parent | 6ccd8e4cf69efbc8983a9873a41158f554ea5363 (diff) | |
| download | go-408739fc96f00a50642673544e803a3c3f3e27df.tar.xz | |
cmd/compile,cmd/asm: on PPC64, generate compares against constant 0
Merge the handling of CMPx r,r,cr and CMPx r,i,cr when assembling.
This prevents generating machine code like cmpd rx,r0 when cmpdi rx,0
is preferred. The preferred form can be fused on Power10 for faster
execution of some instruction sequences.
Likewise, update a common case to use $0 instead of R0 to take
advantage of this.
Change-Id: If2549ca25a5f7d23001885ad444c70d829b3b066
Cq-Include-Trybots: luci.golang.try:gotip-linux-ppc64_power10,gotip-linux-ppc64_power8,gotip-linux-ppc64le_power10,gotip-linux-ppc64le_power8,gotip-linux-ppc64le_power9
Reviewed-on: https://go-review.googlesource.com/c/go/+/585137
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Diffstat (limited to 'src/cmd/internal/obj')
| -rw-r--r-- | src/cmd/internal/obj/ppc64/asm9.go | 28 |
1 files changed, 9 insertions, 19 deletions
diff --git a/src/cmd/internal/obj/ppc64/asm9.go b/src/cmd/internal/obj/ppc64/asm9.go index d9b7c2eed3..74f1772e3d 100644 --- a/src/cmd/internal/obj/ppc64/asm9.go +++ b/src/cmd/internal/obj/ppc64/asm9.go @@ -472,12 +472,12 @@ var optabBase = []Optab{ {as: ACMP, a1: C_REG, a6: C_REG, type_: 70, size: 4}, {as: ACMP, a1: C_REG, a2: C_CREG, a6: C_REG, type_: 70, size: 4}, - {as: ACMP, a1: C_REG, a6: C_S16CON, type_: 71, size: 4}, - {as: ACMP, a1: C_REG, a2: C_CREG, a6: C_S16CON, type_: 71, size: 4}, + {as: ACMP, a1: C_REG, a6: C_S16CON, type_: 70, size: 4}, + {as: ACMP, a1: C_REG, a2: C_CREG, a6: C_S16CON, type_: 70, size: 4}, {as: ACMPU, a1: C_REG, a6: C_REG, type_: 70, size: 4}, {as: ACMPU, a1: C_REG, a2: C_CREG, a6: C_REG, type_: 70, size: 4}, - {as: ACMPU, a1: C_REG, a6: C_U16CON, type_: 71, size: 4}, - {as: ACMPU, a1: C_REG, a2: C_CREG, a6: C_U16CON, type_: 71, size: 4}, + {as: ACMPU, a1: C_REG, a6: C_U16CON, type_: 70, size: 4}, + {as: ACMPU, a1: C_REG, a2: C_CREG, a6: C_U16CON, type_: 70, size: 4}, {as: AFCMPO, a1: C_FREG, a6: C_FREG, type_: 70, size: 4}, {as: AFCMPO, a1: C_FREG, a2: C_CREG, a6: C_FREG, type_: 70, size: 4}, {as: ATW, a1: C_32CON, a2: C_REG, a6: C_REG, type_: 60, size: 4}, @@ -3449,23 +3449,13 @@ func asmout(c *ctxt9, p *obj.Prog, o *Optab, out *[5]uint32) { o1 = AOP_RRR(OP_MTCRF, uint32(p.From.Reg), 0, 0) | uint32(v)<<12 - case 70: /* [f]cmp r,r,cr*/ - var r int - if p.Reg == 0 { - r = 0 - } else { - r = (int(p.Reg) & 7) << 2 - } - o1 = AOP_RRR(c.oprrr(p.As), uint32(r), uint32(p.From.Reg), uint32(p.To.Reg)) - - case 71: /* cmp[l] r,i,cr*/ - var r int - if p.Reg == 0 { - r = 0 + case 70: /* cmp* r,r,cr or cmp*i r,i,cr or fcmp f,f,cr or cmpeqb r,r */ + r := uint32(p.Reg&7) << 2 + if p.To.Type == obj.TYPE_CONST { + o1 = AOP_IRR(c.opirr(p.As), r, uint32(p.From.Reg), uint32(uint16(p.To.Offset))) } else { - r = (int(p.Reg) & 7) << 2 + o1 = AOP_RRR(c.oprrr(p.As), r, uint32(p.From.Reg), uint32(p.To.Reg)) } - o1 = AOP_RRR(c.opirr(p.As), uint32(r), uint32(p.From.Reg), 0) | uint32(c.regoff(&p.To))&0xffff case 72: /* slbmte (Rb+Rs -> slb[Rb]) -> Rs, Rb */ o1 = AOP_RRR(c.oprrr(p.As), uint32(p.From.Reg), 0, uint32(p.To.Reg)) |