[PATCH] D14478: AMDGPU: Set isAllocatable = 0 on VS_32/VS_64
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Fri Nov 6 23:18:17 PST 2015
arsenm created this revision.
arsenm added a reviewer: tstellarAMD.
arsenm added a subscriber: llvm-commits.
Herald added a subscriber: arsenm.
http://reviews.llvm.org/D14478
Files:
lib/Target/AMDGPU/SIRegisterInfo.cpp
lib/Target/AMDGPU/SIRegisterInfo.h
lib/Target/AMDGPU/SIRegisterInfo.td
test/CodeGen/AMDGPU/ftrunc.f64.ll
test/CodeGen/AMDGPU/llvm.round.f64.ll
Index: test/CodeGen/AMDGPU/llvm.round.f64.ll
===================================================================
--- test/CodeGen/AMDGPU/llvm.round.f64.ll
+++ test/CodeGen/AMDGPU/llvm.round.f64.ll
@@ -21,7 +21,7 @@
; SI-DAG: v_cmp_eq_i32
; SI-DAG: s_mov_b32 [[BFIMASK:s[0-9]+]], 0x7fffffff
-; SI-DAG: v_cmp_gt_i32_e32
+; SI-DAG: v_cmp_gt_i32
; SI-DAG: v_bfi_b32 [[COPYSIGN:v[0-9]+]], [[BFIMASK]]
; SI: buffer_store_dwordx2
Index: test/CodeGen/AMDGPU/ftrunc.f64.ll
===================================================================
--- test/CodeGen/AMDGPU/ftrunc.f64.ll
+++ test/CodeGen/AMDGPU/ftrunc.f64.ll
@@ -27,8 +27,8 @@
; SI: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000000
; SI: s_add_i32 s{{[0-9]+}}, [[SEXP]], 0xfffffc01
; SI: s_lshr_b64
-; SI: s_not_b64
-; SI: s_and_b64
+; SI-DAG: s_not_b64
+; SI-DAG: s_and_b64
; SI-DAG: cmp_gt_i32
; SI-DAG: cndmask_b32
; SI-DAG: cndmask_b32
Index: lib/Target/AMDGPU/SIRegisterInfo.td
===================================================================
--- lib/Target/AMDGPU/SIRegisterInfo.td
+++ lib/Target/AMDGPU/SIRegisterInfo.td
@@ -272,9 +272,12 @@
// VSrc_* Operands with an SGPR, VGPR or a 32-bit immediate
//===----------------------------------------------------------------------===//
-def VS_32 : RegisterClass<"AMDGPU", [i32, f32], 32, (add VGPR_32, SReg_32)>;
+def VS_32 : RegisterClass<"AMDGPU", [i32, f32], 32, (add SReg_32, VGPR_32)> {
+ let isAllocatable = 0;
+}
-def VS_64 : RegisterClass<"AMDGPU", [i64, f64], 32, (add VReg_64, SReg_64)> {
+def VS_64 : RegisterClass<"AMDGPU", [i64, f64], 32, (add SReg_64, VReg_64)> {
+ let isAllocatable = 0;
let CopyCost = 2;
}
Index: lib/Target/AMDGPU/SIRegisterInfo.h
===================================================================
--- lib/Target/AMDGPU/SIRegisterInfo.h
+++ lib/Target/AMDGPU/SIRegisterInfo.h
@@ -59,13 +59,6 @@
/// \returns true if this class contains VGPR registers.
bool hasVGPRs(const TargetRegisterClass *RC) const;
- /// returns true if this is a pseudoregister class combination of VGPRs and
- /// SGPRs for operand modeling. FIXME: We should set isAllocatable = 0 on
- /// them.
- static bool isPseudoRegClass(const TargetRegisterClass *RC) {
- return RC == &AMDGPU::VS_32RegClass || RC == &AMDGPU::VS_64RegClass;
- }
-
/// \returns A VGPR reg class with the same width as \p SRC
const TargetRegisterClass *getEquivalentVGPRClass(
const TargetRegisterClass *SRC) const;
Index: lib/Target/AMDGPU/SIRegisterInfo.cpp
===================================================================
--- lib/Target/AMDGPU/SIRegisterInfo.cpp
+++ lib/Target/AMDGPU/SIRegisterInfo.cpp
@@ -79,20 +79,14 @@
STI.getMaxWavesPerCU());
unsigned VGPRLimit = getNumVGPRsAllowed(STI.getMaxWavesPerCU());
- unsigned VSLimit = SGPRLimit + VGPRLimit;
-
for (regclass_iterator I = regclass_begin(), E = regclass_end();
I != E; ++I) {
const TargetRegisterClass *RC = *I;
unsigned NumSubRegs = std::max((int)RC->getSize() / 4, 1);
unsigned Limit;
- if (isPseudoRegClass(RC)) {
- // FIXME: This is a hack. We should never be considering the pressure of
- // these since no virtual register should ever have this class.
- Limit = VSLimit;
- } else if (isSGPRClass(RC)) {
+ if (isSGPRClass(RC)) {
Limit = SGPRLimit / NumSubRegs;
} else {
Limit = VGPRLimit / NumSubRegs;
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D14478.39633.patch
Type: text/x-patch
Size: 3478 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151107/1438d464/attachment.bin>
More information about the llvm-commits
mailing list