RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2024-11-29 06:30:39 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

R600: Report that integer division is expensive.

Browse Source 
Divides by weird constants now emit much better code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@210995 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Matt Arsenault 2014-06-15 19:48:16 +00:00
parent 36b9c7c872
commit 61bfbc4d96
2 changed files with 61 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
lib/Target/R600/AMDGPUISelLowering.cpp
View File

@ -327,6 +327,13 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
setSchedulingPreference(Sched::RegPressure);
setJumpIsExpensive(true);
// There are no integer divide instructions, and these expand to a pretty
// large sequence of instructions.
setIntDivIsCheap(false);
// TODO: Investigate this when 64-bit divides are implemented.
addBypassSlowDiv(64, 32);
// FIXME: Need to really handle these.
MaxStoresPerMemcpy = 4096;
MaxStoresPerMemmove = 4096;

58
test/CodeGen/R600/sdiv.ll
View File

@ -1,4 +1,5 @@
; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
; RUN: llc -march=r600 -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
; The code generated by sdiv is long and complex and may frequently change.
; The goal of this test is to make sure the ISel doesn't fail.
@ -9,9 +10,9 @@
; This was fixed by adding an additional pattern in R600Instructions.td to
; match this pattern with a CNDGE_INT.
; CHECK: CF_END
define void @test(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
; FUNC-LABEL: @sdiv_i32
; EG: CF_END
define void @sdiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
%den_ptr = getelementptr i32 addrspace(1)* %in, i32 1
%num = load i32 addrspace(1) * %in
%den = load i32 addrspace(1) * %den_ptr
@ -19,3 +20,52 @@ define void @test(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
store i32 %result, i32 addrspace(1)* %out
ret void
}
; FUNC-LABEL: @sdiv_i32_4
define void @sdiv_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
%num = load i32 addrspace(1) * %in
%result = sdiv i32 %num, 4
store i32 %result, i32 addrspace(1)* %out
ret void
}
; Multiply by a weird constant to make sure setIntDivIsCheap is
; working.
; FUNC-LABEL: @slow_sdiv_i32_3435
; SI: BUFFER_LOAD_DWORD [[VAL:v[0-9]+]],
; SI: V_MOV_B32_e32 [[MAGIC:v[0-9]+]], 0x98a1930b
; SI: V_MUL_HI_I32 [[TMP:v[0-9]+]], [[VAL]], [[MAGIC]]
; SI: V_ADD_I32
; SI: V_LSHRREV_B32
; SI: V_ASHRREV_I32
; SI: V_ADD_I32
; SI: BUFFER_STORE_DWORD
; SI: S_ENDPGM
define void @slow_sdiv_i32_3435(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
%num = load i32 addrspace(1) * %in
%result = sdiv i32 %num, 3435
store i32 %result, i32 addrspace(1)* %out
ret void
}
; Tests for 64-bit divide bypass.
; define void @test_get_quotient(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind {
; %result = sdiv i64 %a, %b
; store i64 %result, i64 addrspace(1)* %out, align 8
; ret void
; }
; define void @test_get_remainder(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind {
; %result = srem i64 %a, %b
; store i64 %result, i64 addrspace(1)* %out, align 8
; ret void
; }
; define void @test_get_quotient_and_remainder(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind {
; %resultdiv = sdiv i64 %a, %b
; %resultrem = srem i64 %a, %b
; %result = add i64 %resultdiv, %resultrem
; store i64 %result, i64 addrspace(1)* %out, align 8
; ret void
; }