This patch combines several changes from Evan Cheng for rdar://8659675.

Making use of VFP / NEON floating point multiply-accumulate / subtraction is
difficult on current ARM implementations for a few reasons.
1. Even though a single vmla has latency that is one cycle shorter than a pair
   of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause
   additional pipeline stall. So it's frequently better to single codegen
   vmul + vadd.
2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to
   stall for 4 cycles. We need to schedule them apart.
3. A vmla followed vmla is a special case. Obvious issuing back to back RAW
   vmla + vmla is very bad. But this isn't ideal either:
     vmul
     vadd
     vmla
   Instead, we want to expand the second vmla:
     vmla
     vmul
     vadd
   Even with the 4 cycle vmul stall, the second sequence is still 2 cycles
   faster.

Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough
but it isn't the optimial solution. This patch attempts to make it possible to
use vmla / vmls in cases where it is profitable.

A. Add missing isel predicates which cause vmla to be codegen'ed.
B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to
   compute a fmul and a fmla.
C. Add additional isel checks for vmla, avoid cases where vmla is feeding into
   fp instructions (except for the #3 exceptional case).
D. Add ARM hazard recognizer to model the vmla / vmls hazards.
E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the
   vmla / vmls will trigger one of the special hazards.

Enable these fp vmlx codegen changes for Cortex-A9.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129775 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 53 insertions, 0 deletions

diff --git a/test/CodeGen/ARM/fmacs.ll b/test/CodeGen/ARM/fmacs.ll
index fb83ef6..b63f609 100644
--- a/test/CodeGen/ARM/fmacs.ll
+++ b/test/CodeGen/ARM/fmacs.ll
@@ -1,6 +1,8 @@
 ; RUN: llc < %s -march=arm -mattr=+vfp2 | FileCheck %s -check-prefix=VFP2
 ; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s -check-prefix=NEON
 ; RUN: llc < %s -march=arm -mcpu=cortex-a8 | FileCheck %s -check-prefix=A8
+; RUN: llc < %s -march=arm -mcpu=cortex-a9 | FileCheck %s -check-prefix=A9
+; RUN: llc < %s -mtriple=arm-linux-gnueabi -mcpu=cortex-a9 -float-abi=hard | FileCheck %s -check-prefix=HARD
 
 define float @t1(float %acc, float %a, float %b) {
 entry:
@@ -49,3 +51,54 @@ entry:
         %1 = fadd float %0, %acc
 	ret float %1
 }
+
+; It's possible to make use of fp vmla / vmls on Cortex-A9.
+; rdar://8659675
+define void @t4(float %acc1, float %a, float %b, float %acc2, float %c, float* %P1, float* %P2) {
+entry:
+; A8: t4:
+; A8: vmul.f32
+; A8: vmul.f32
+; A8: vadd.f32
+; A8: vadd.f32
+
+; Two vmla with now RAW hazard
+; A9: t4:
+; A9: vmla.f32
+; A9: vmla.f32
+
+; HARD: t4:
+; HARD: vmla.f32 s0, s1, s2
+; HARD: vmla.f32 s3, s1, s4
+  %0 = fmul float %a, %b
+  %1 = fadd float %acc1, %0
+  %2 = fmul float %a, %c
+  %3 = fadd float %acc2, %2
+  store float %1, float* %P1
+  store float %3, float* %P2
+  ret void
+}
+
+define float @t5(float %a, float %b, float %c, float %d, float %e) {
+entry:
+; A8: t5:
+; A8: vmul.f32
+; A8: vmul.f32
+; A8: vadd.f32
+; A8: vadd.f32
+
+; A9: t5:
+; A9: vmla.f32
+; A9: vmul.f32
+; A9: vadd.f32
+
+; HARD: t5:
+; HARD: vmla.f32 s4, s0, s1
+; HARD: vmul.f32 s0, s2, s3
+; HARD: vadd.f32 s0, s4, s0
+  %0 = fmul float %a, %b
+  %1 = fadd float %e, %0
+  %2 = fmul float %c, %d
+  %3 = fadd float %1, %2
+  ret float %3
+}