[llvm] r295314 - [DAGCombiner] Support {a|s}ext, {a|z|s}ext load nodes in load combine
Artur Pilipenko via llvm-commits
llvm-commits at lists.llvm.org
Thu Feb 16 04:53:26 PST 2017
Author: apilipenko
Date: Thu Feb 16 06:53:26 2017
New Revision: 295314
URL: http://llvm.org/viewvc/llvm-project?rev=295314&view=rev
Log:
[DAGCombiner] Support {a|s}ext, {a|z|s}ext load nodes in load combine
Support {a|s}ext, {a|z|s}ext load nodes as a part of load combine patters.
Reviewed By: filcab
Differential Revision: https://reviews.llvm.org/D29591
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
llvm/trunk/test/CodeGen/AArch64/load-combine-big-endian.ll
llvm/trunk/test/CodeGen/AArch64/load-combine.ll
llvm/trunk/test/CodeGen/ARM/fp16-promote.ll
llvm/trunk/test/CodeGen/ARM/load-combine-big-endian.ll
llvm/trunk/test/CodeGen/ARM/load-combine.ll
llvm/trunk/test/CodeGen/X86/load-combine.ll
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=295314&r1=295313&r2=295314&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Thu Feb 16 06:53:26 2017
@@ -4446,6 +4446,8 @@ const Optional<ByteProvider> calculateBy
: calculateByteProvider(Op->getOperand(0), Index - ByteShift,
Depth + 1);
}
+ case ISD::ANY_EXTEND:
+ case ISD::SIGN_EXTEND:
case ISD::ZERO_EXTEND: {
SDValue NarrowOp = Op->getOperand(0);
unsigned NarrowBitWidth = NarrowOp.getScalarValueSizeInBits();
@@ -4453,22 +4455,32 @@ const Optional<ByteProvider> calculateBy
return None;
uint64_t NarrowByteWidth = NarrowBitWidth / 8;
- return Index >= NarrowByteWidth
- ? ByteProvider::getConstantZero()
- : calculateByteProvider(NarrowOp, Index, Depth + 1);
+ if (Index >= NarrowByteWidth)
+ return Op.getOpcode() == ISD::ZERO_EXTEND
+ ? Optional<ByteProvider>(ByteProvider::getConstantZero())
+ : None;
+ else
+ return calculateByteProvider(NarrowOp, Index, Depth + 1);
}
case ISD::BSWAP:
return calculateByteProvider(Op->getOperand(0), ByteWidth - Index - 1,
Depth + 1);
case ISD::LOAD: {
auto L = cast<LoadSDNode>(Op.getNode());
+ if (L->isVolatile() || L->isIndexed())
+ return None;
- // TODO: support ext loads
- if (L->isVolatile() || L->isIndexed() ||
- L->getExtensionType() != ISD::NON_EXTLOAD)
+ unsigned NarrowBitWidth = L->getMemoryVT().getSizeInBits();
+ if (NarrowBitWidth % 8 != 0)
return None;
+ uint64_t NarrowByteWidth = NarrowBitWidth / 8;
- return ByteProvider::getMemory(L, Index);
+ if (Index >= NarrowByteWidth)
+ return L->getExtensionType() == ISD::ZEXTLOAD
+ ? Optional<ByteProvider>(ByteProvider::getConstantZero())
+ : None;
+ else
+ return ByteProvider::getMemory(L, Index);
}
}
@@ -4548,7 +4560,6 @@ SDValue DAGCombiner::MatchLoadCombine(SD
LoadSDNode *L = P->Load;
assert(L->hasNUsesOfValue(1, 0) && !L->isVolatile() && !L->isIndexed() &&
- (L->getExtensionType() == ISD::NON_EXTLOAD) &&
"Must be enforced by calculateByteProvider");
assert(L->getOffset().isUndef() && "Unindexed load must have undef offset");
Modified: llvm/trunk/test/CodeGen/AArch64/load-combine-big-endian.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/load-combine-big-endian.ll?rev=295314&r1=295313&r2=295314&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AArch64/load-combine-big-endian.ll (original)
+++ llvm/trunk/test/CodeGen/AArch64/load-combine-big-endian.ll Thu Feb 16 06:53:26 2017
@@ -336,11 +336,8 @@ define i32 @load_i32_by_bswap_i16(i32* %
; (i32) p[1] | (sext(p[0] << 16) to i32)
define i32 @load_i32_by_sext_i16(i32* %arg) {
; CHECK-LABEL: load_i32_by_sext_i16:
-; CHECK: ldrh w8, [x0]
-; CHECK-NEXT: ldrh w0, [x0, #2]
-; CHECK-NEXT: bfi w0, w8, #16, #16
+; CHECK: ldr w0, [x0]
; CHECK-NEXT: ret
-
%tmp = bitcast i32* %arg to i16*
%tmp1 = load i16, i16* %tmp, align 4
%tmp2 = sext i16 %tmp1 to i32
@@ -399,7 +396,6 @@ define i32 @load_i32_by_i8_base_offset_i
; CHECK-NEXT: ldur w8, [x8, #13]
; CHECK-NEXT: rev w0, w8
; CHECK-NEXT: ret
-
%tmp = add nuw nsw i32 %i, 4
%tmp2 = add nuw nsw i32 %i, 3
%tmp3 = add nuw nsw i32 %i, 2
Modified: llvm/trunk/test/CodeGen/AArch64/load-combine.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/load-combine.ll?rev=295314&r1=295313&r2=295314&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AArch64/load-combine.ll (original)
+++ llvm/trunk/test/CodeGen/AArch64/load-combine.ll Thu Feb 16 06:53:26 2017
@@ -324,12 +324,8 @@ define i32 @load_i32_by_bswap_i16(i32* %
; (i32) p[0] | (sext(p[1] << 16) to i32)
define i32 @load_i32_by_sext_i16(i32* %arg) {
; CHECK-LABEL: load_i32_by_sext_i16:
-; CHECK: ldrh w8, [x0]
-; CHECK-NEXT: ldrh w9, [x0, #2]
-; CHECK-NEXT: bfi w8, w9, #16, #16
-; CHECK-NEXT: mov w0, w8
+; CHECK: ldr w0, [x0]
; CHECK-NEXT: ret
-
%tmp = bitcast i32* %arg to i16*
%tmp1 = load i16, i16* %tmp, align 4
%tmp2 = zext i16 %tmp1 to i32
@@ -386,7 +382,6 @@ define i32 @load_i32_by_i8_base_offset_i
; CHECK: add x8, x0, w1, uxtw
; CHECK-NEXT: ldur w0, [x8, #13]
; CHECK-NEXT: ret
-
%tmp = add nuw nsw i32 %i, 4
%tmp2 = add nuw nsw i32 %i, 3
%tmp3 = add nuw nsw i32 %i, 2
Modified: llvm/trunk/test/CodeGen/ARM/fp16-promote.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/fp16-promote.ll?rev=295314&r1=295313&r2=295314&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/ARM/fp16-promote.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/fp16-promote.ll Thu Feb 16 06:53:26 2017
@@ -847,21 +847,15 @@ define void @test_insertelement(half* %p
}
; CHECK-ALL-LABEL: test_extractelement:
+; CHECK-VFP: push {{{.*}}, lr}
; CHECK-VFP: sub sp, sp, #8
-; CHECK-VFP: ldrh
-; CHECK-VFP: ldrh
-; CHECK-VFP: orr
-; CHECK-VFP: str
-; CHECK-VFP: ldrh
-; CHECK-VFP: ldrh
-; CHECK-VFP: orr
-; CHECK-VFP: str
+; CHECK-VFP: ldrd
; CHECK-VFP: mov
; CHECK-VFP: orr
; CHECK-VFP: ldrh
; CHECK-VFP: strh
; CHECK-VFP: add sp, sp, #8
-; CHECK-VFP: bx lr
+; CHECK-VFP: pop {{{.*}}, pc}
; CHECK-NOVFP: ldrh
; CHECK-NOVFP: strh
; CHECK-NOVFP: ldrh
Modified: llvm/trunk/test/CodeGen/ARM/load-combine-big-endian.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/load-combine-big-endian.ll?rev=295314&r1=295313&r2=295314&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/ARM/load-combine-big-endian.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/load-combine-big-endian.ll Thu Feb 16 06:53:26 2017
@@ -456,17 +456,12 @@ define i32 @load_i32_by_bswap_i16(i32* %
; (i32) p[1] | (sext(p[0] << 16) to i32)
define i32 @load_i32_by_sext_i16(i32* %arg) {
; CHECK-LABEL: load_i32_by_sext_i16:
-; CHECK: ldrh r1, [r0]
-; CHECK-NEXT: ldrh r0, [r0, #2]
-; CHECK-NEXT: orr r0, r0, r1, lsl #16
+; CHECK: ldr r0, [r0]
; CHECK-NEXT: mov pc, lr
-
+;
; CHECK-ARMv6-LABEL: load_i32_by_sext_i16:
-; CHECK-ARMv6: ldrh r1, [r0]
-; CHECK-ARMv6-NEXT: ldrh r0, [r0, #2]
-; CHECK-ARMv6-NEXT: orr r0, r0, r1, lsl #16
+; CHECK-ARMv6: ldr r0, [r0]
; CHECK-ARMv6-NEXT: bx lr
-
%tmp = bitcast i32* %arg to i16*
%tmp1 = load i16, i16* %tmp, align 4
%tmp2 = sext i16 %tmp1 to i32
Modified: llvm/trunk/test/CodeGen/ARM/load-combine.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/load-combine.ll?rev=295314&r1=295313&r2=295314&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/ARM/load-combine.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/load-combine.ll Thu Feb 16 06:53:26 2017
@@ -414,17 +414,12 @@ define i32 @load_i32_by_bswap_i16(i32* %
; (i32) p[0] | (sext(p[1] << 16) to i32)
define i32 @load_i32_by_sext_i16(i32* %arg) {
; CHECK-LABEL: load_i32_by_sext_i16:
-; CHECK: ldrh r1, [r0, #2]
-; CHECK-NEXT: ldrh r0, [r0]
-; CHECK-NEXT: orr r0, r0, r1, lsl #16
+; CHECK: ldr r0, [r0]
; CHECK-NEXT: mov pc, lr
;
; CHECK-ARMv6-LABEL: load_i32_by_sext_i16:
-; CHECK-ARMv6: ldrh r1, [r0, #2]
-; CHECK-ARMv6-NEXT: ldrh r0, [r0]
-; CHECK-ARMv6-NEXT: orr r0, r0, r1, lsl #16
-; CHECK-ARMv6-NEXT: bx lr
-
+; CHECK-ARMv6: ldr r0, [r0]
+; CHECK-ARMv6-NEXT: bx lr
%tmp = bitcast i32* %arg to i16*
%tmp1 = load i16, i16* %tmp, align 4
%tmp2 = zext i16 %tmp1 to i32
@@ -492,7 +487,6 @@ define i32 @load_i32_by_i8_base_offset_i
; CHECK-ARMv6: add r0, r0, r1
; CHECK-ARMv6-NEXT: ldr r0, [r0, #13]
; CHECK-ARMv6-NEXT: bx lr
-
%tmp = add nuw nsw i32 %i, 4
%tmp2 = add nuw nsw i32 %i, 3
%tmp3 = add nuw nsw i32 %i, 2
Modified: llvm/trunk/test/CodeGen/X86/load-combine.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/load-combine.ll?rev=295314&r1=295313&r2=295314&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/load-combine.ll (original)
+++ llvm/trunk/test/CodeGen/X86/load-combine.ll Thu Feb 16 06:53:26 2017
@@ -733,16 +733,8 @@ define i32 @load_i32_by_i8_bswap_base_in
; CHECK64-LABEL: load_i32_by_i8_bswap_base_index_offset:
; CHECK64: # BB#0:
; CHECK64-NEXT: movslq %esi, %rax
-; CHECK64-NEXT: movzbl (%rdi,%rax), %ecx
-; CHECK64-NEXT: shll $24, %ecx
-; CHECK64-NEXT: movzbl 1(%rdi,%rax), %edx
-; CHECK64-NEXT: shll $16, %edx
-; CHECK64-NEXT: orl %ecx, %edx
-; CHECK64-NEXT: movzbl 2(%rdi,%rax), %ecx
-; CHECK64-NEXT: shll $8, %ecx
-; CHECK64-NEXT: orl %edx, %ecx
-; CHECK64-NEXT: movzbl 3(%rdi,%rax), %eax
-; CHECK64-NEXT: orl %ecx, %eax
+; CHECK64-NEXT: movl (%rdi,%rax), %eax
+; CHECK64-NEXT: bswapl %eax
; CHECK64-NEXT: retq
%tmp = bitcast i32* %arg to i8*
%tmp2 = getelementptr inbounds i8, i8* %tmp, i32 %arg1
@@ -835,18 +827,12 @@ define i32 @load_i32_by_sext_i16(i32* %a
; CHECK-LABEL: load_i32_by_sext_i16:
; CHECK: # BB#0:
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %eax
-; CHECK-NEXT: movzwl (%eax), %ecx
-; CHECK-NEXT: movzwl 2(%eax), %eax
-; CHECK-NEXT: shll $16, %eax
-; CHECK-NEXT: orl %ecx, %eax
+; CHECK-NEXT: movl (%eax), %eax
; CHECK-NEXT: retl
;
; CHECK64-LABEL: load_i32_by_sext_i16:
; CHECK64: # BB#0:
-; CHECK64-NEXT: movzwl (%rdi), %ecx
-; CHECK64-NEXT: movzwl 2(%rdi), %eax
-; CHECK64-NEXT: shll $16, %eax
-; CHECK64-NEXT: orl %ecx, %eax
+; CHECK64-NEXT: movl (%rdi), %eax
; CHECK64-NEXT: retq
%tmp = bitcast i32* %arg to i16*
%tmp1 = load i16, i16* %tmp, align 1
@@ -865,24 +851,9 @@ define i32 @load_i32_by_sext_i16(i32* %a
define i32 @load_i32_by_i8_base_offset_index(i8* %arg, i32 %i) {
; CHECK-LABEL: load_i32_by_i8_base_offset_index:
; CHECK: # BB#0:
-; CHECK-NEXT: pushl %esi
-; CHECK-NEXT: .Lcfi4:
-; CHECK-NEXT: .cfi_def_cfa_offset 8
-; CHECK-NEXT: .Lcfi5:
-; CHECK-NEXT: .cfi_offset %esi, -8
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %eax
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; CHECK-NEXT: movzbl 12(%eax,%ecx), %edx
-; CHECK-NEXT: movzbl 13(%eax,%ecx), %esi
-; CHECK-NEXT: shll $8, %esi
-; CHECK-NEXT: orl %edx, %esi
-; CHECK-NEXT: movzbl 14(%eax,%ecx), %edx
-; CHECK-NEXT: shll $16, %edx
-; CHECK-NEXT: orl %esi, %edx
-; CHECK-NEXT: movzbl 15(%eax,%ecx), %eax
-; CHECK-NEXT: shll $24, %eax
-; CHECK-NEXT: orl %edx, %eax
-; CHECK-NEXT: popl %esi
+; CHECK-NEXT: movl 12(%eax,%ecx), %eax
; CHECK-NEXT: retl
;
; CHECK64-LABEL: load_i32_by_i8_base_offset_index:
@@ -925,24 +896,9 @@ define i32 @load_i32_by_i8_base_offset_i
define i32 @load_i32_by_i8_base_offset_index_2(i8* %arg, i32 %i) {
; CHECK-LABEL: load_i32_by_i8_base_offset_index_2:
; CHECK: # BB#0:
-; CHECK-NEXT: pushl %esi
-; CHECK-NEXT: .Lcfi6:
-; CHECK-NEXT: .cfi_def_cfa_offset 8
-; CHECK-NEXT: .Lcfi7:
-; CHECK-NEXT: .cfi_offset %esi, -8
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %eax
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; CHECK-NEXT: movzbl 13(%eax,%ecx), %edx
-; CHECK-NEXT: movzbl 14(%eax,%ecx), %esi
-; CHECK-NEXT: shll $8, %esi
-; CHECK-NEXT: orl %edx, %esi
-; CHECK-NEXT: movzbl 15(%eax,%ecx), %edx
-; CHECK-NEXT: shll $16, %edx
-; CHECK-NEXT: orl %esi, %edx
-; CHECK-NEXT: movzbl 16(%eax,%ecx), %eax
-; CHECK-NEXT: shll $24, %eax
-; CHECK-NEXT: orl %edx, %eax
-; CHECK-NEXT: popl %esi
+; CHECK-NEXT: movl 13(%eax,%ecx), %eax
; CHECK-NEXT: retl
;
; CHECK64-LABEL: load_i32_by_i8_base_offset_index_2:
@@ -996,39 +952,15 @@ define i32 @load_i32_by_i8_base_offset_i
define i32 @load_i32_by_i8_zaext_loads(i8* %arg, i32 %arg1) {
; CHECK-LABEL: load_i32_by_i8_zaext_loads:
; CHECK: # BB#0:
-; CHECK-NEXT: pushl %esi
-; CHECK-NEXT: .Lcfi8:
-; CHECK-NEXT: .cfi_def_cfa_offset 8
-; CHECK-NEXT: .Lcfi9:
-; CHECK-NEXT: .cfi_offset %esi, -8
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %eax
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; CHECK-NEXT: movzbl 12(%eax,%ecx), %edx
-; CHECK-NEXT: movzbl 13(%eax,%ecx), %esi
-; CHECK-NEXT: shll $8, %esi
-; CHECK-NEXT: orl %edx, %esi
-; CHECK-NEXT: movzbl 14(%eax,%ecx), %edx
-; CHECK-NEXT: shll $16, %edx
-; CHECK-NEXT: orl %esi, %edx
-; CHECK-NEXT: movzbl 15(%eax,%ecx), %eax
-; CHECK-NEXT: shll $24, %eax
-; CHECK-NEXT: orl %edx, %eax
-; CHECK-NEXT: popl %esi
+; CHECK-NEXT: movl 12(%eax,%ecx), %eax
; CHECK-NEXT: retl
;
; CHECK64-LABEL: load_i32_by_i8_zaext_loads:
; CHECK64: # BB#0:
; CHECK64-NEXT: movl %esi, %eax
-; CHECK64-NEXT: movzbl 12(%rdi,%rax), %ecx
-; CHECK64-NEXT: movzbl 13(%rdi,%rax), %edx
-; CHECK64-NEXT: shll $8, %edx
-; CHECK64-NEXT: orl %ecx, %edx
-; CHECK64-NEXT: movzbl 14(%rdi,%rax), %ecx
-; CHECK64-NEXT: shll $16, %ecx
-; CHECK64-NEXT: orl %edx, %ecx
-; CHECK64-NEXT: movzbl 15(%rdi,%rax), %eax
-; CHECK64-NEXT: shll $24, %eax
-; CHECK64-NEXT: orl %ecx, %eax
+; CHECK64-NEXT: movl 12(%rdi,%rax), %eax
; CHECK64-NEXT: retq
%tmp = add nuw nsw i32 %arg1, 3
%tmp2 = add nuw nsw i32 %arg1, 2
@@ -1076,39 +1008,15 @@ define i32 @load_i32_by_i8_zaext_loads(i
define i32 @load_i32_by_i8_zsext_loads(i8* %arg, i32 %arg1) {
; CHECK-LABEL: load_i32_by_i8_zsext_loads:
; CHECK: # BB#0:
-; CHECK-NEXT: pushl %esi
-; CHECK-NEXT: .Lcfi10:
-; CHECK-NEXT: .cfi_def_cfa_offset 8
-; CHECK-NEXT: .Lcfi11:
-; CHECK-NEXT: .cfi_offset %esi, -8
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %eax
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; CHECK-NEXT: movzbl 12(%eax,%ecx), %edx
-; CHECK-NEXT: movzbl 13(%eax,%ecx), %esi
-; CHECK-NEXT: shll $8, %esi
-; CHECK-NEXT: orl %edx, %esi
-; CHECK-NEXT: movzbl 14(%eax,%ecx), %edx
-; CHECK-NEXT: shll $16, %edx
-; CHECK-NEXT: orl %esi, %edx
-; CHECK-NEXT: movsbl 15(%eax,%ecx), %eax
-; CHECK-NEXT: shll $24, %eax
-; CHECK-NEXT: orl %edx, %eax
-; CHECK-NEXT: popl %esi
+; CHECK-NEXT: movl 12(%eax,%ecx), %eax
; CHECK-NEXT: retl
;
; CHECK64-LABEL: load_i32_by_i8_zsext_loads:
; CHECK64: # BB#0:
; CHECK64-NEXT: movl %esi, %eax
-; CHECK64-NEXT: movzbl 12(%rdi,%rax), %ecx
-; CHECK64-NEXT: movzbl 13(%rdi,%rax), %edx
-; CHECK64-NEXT: shll $8, %edx
-; CHECK64-NEXT: orl %ecx, %edx
-; CHECK64-NEXT: movzbl 14(%rdi,%rax), %ecx
-; CHECK64-NEXT: shll $16, %ecx
-; CHECK64-NEXT: orl %edx, %ecx
-; CHECK64-NEXT: movsbl 15(%rdi,%rax), %eax
-; CHECK64-NEXT: shll $24, %eax
-; CHECK64-NEXT: orl %ecx, %eax
+; CHECK64-NEXT: movl 12(%rdi,%rax), %eax
; CHECK64-NEXT: retq
%tmp = add nuw nsw i32 %arg1, 3
%tmp2 = add nuw nsw i32 %arg1, 2
More information about the llvm-commits
mailing list