[RISCV] Move strength reduction of mul X, 3/5/9*2^N to combine #89966
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This moves our last major category tablegen driven multiply strength reduction into the post legalize combine framework. The one slightly tricky bit is making sure that we use a leading shl if we can form a slli.uw, and trailing shl otherwise. Having the trailing shl is critical for shNadd matching, and folding any following sext.w. As can be seen in the TD deltas, this allows us to kill off both the actual multiply patterns and the explicit add (mul X, C) Y patterns. The later are now handled by the generic shNadd matching code, with the exception of the THead only C=200 case because we don't (yet) have a multiply expansion with two shNadd + a shift.
|
@llvm/pr-subscribers-backend-risc-v Author: Philip Reames (preames) ChangesThis moves our last major category tablegen driven multiply strength reduction into the post legalize combine framework. The one slightly tricky bit is making sure that we use a leading shl if we can form a slli.uw, and trailing shl otherwise. Having the trailing shl is critical for shNadd matching, and folding any following sext.w. As can be seen in the TD deltas, this allows us to kill off both the actual multiply patterns and the explicit add (mul X, C) Y patterns. The later are now handled by the generic shNadd matching code, with the exception of the THead only C=200 case because we don't (yet) have a multiply expansion with two shNadd + a shift. Full diff: https://github.com/llvm/llvm-project/pull/89966.diff 6 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 9c66f09a0cbc85..6710785e8d4da8 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -13420,10 +13420,29 @@ static SDValue expandMul(SDNode *N, SelectionDAG &DAG,
if (MulAmt % Divisor != 0)
continue;
uint64_t MulAmt2 = MulAmt / Divisor;
- // 3/5/9 * 2^N -> shXadd (sll X, C), (sll X, C)
- // Matched in tablegen, avoid perturbing patterns.
- if (isPowerOf2_64(MulAmt2))
- return SDValue();
+ // 3/5/9 * 2^N -> shl (shXadd X, X), N
+ if (isPowerOf2_64(MulAmt2)) {
+ SDLoc DL(N);
+ SDValue X = N->getOperand(0);
+ // Put the shift first if we can fold a zext into the
+ // shift forming a sll.uw.
+ if (X.getOpcode() == ISD::AND && isa<ConstantSDNode>(X.getOperand(1)) &&
+ X.getConstantOperandVal(1) == UINT64_C(0xffffffff)) {
+ SDValue Shl = DAG.getNode(ISD::SHL, DL, VT, X,
+ DAG.getConstant(Log2_64(MulAmt2), DL, VT));
+ Shl = DAG.getFreeze(Shl);
+ return DAG.getNode(RISCVISD::SHL_ADD, DL, VT, Shl,
+ DAG.getConstant(Log2_64(Divisor - 1), DL, VT), Shl);
+ }
+ // Otherwise, put rhe shl second so that it can fold with following
+ // instructions (e.g. sext or add).
+ X = DAG.getFreeze(X);
+ SDValue Mul359 =
+ DAG.getNode(RISCVISD::SHL_ADD, DL, VT, X,
+ DAG.getConstant(Log2_64(Divisor - 1), DL, VT), X);
+ return DAG.getNode(ISD::SHL, DL, VT, Mul359,
+ DAG.getConstant(Log2_64(MulAmt2), DL, VT));
+ }
// 3/5/9 * 3/5/9 -> shXadd (shYadd X, X), (shYadd X, X)
if (MulAmt2 == 3 || MulAmt2 == 5 || MulAmt2 == 9) {
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td b/llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td
index b398c5e7fec2ec..bc14f165d9622d 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td
@@ -549,40 +549,11 @@ def : Pat<(add_non_imm12 sh2add_op:$rs1, (XLenVT GPR:$rs2)),
def : Pat<(add_non_imm12 sh3add_op:$rs1, (XLenVT GPR:$rs2)),
(TH_ADDSL GPR:$rs2, sh3add_op:$rs1, 3)>;
-def : Pat<(add (mul_oneuse GPR:$rs1, (XLenVT 6)), GPR:$rs2),
- (TH_ADDSL GPR:$rs2, (XLenVT (TH_ADDSL GPR:$rs1, GPR:$rs1, 1)), 1)>;
-def : Pat<(add (mul_oneuse GPR:$rs1, (XLenVT 10)), GPR:$rs2),
- (TH_ADDSL GPR:$rs2, (XLenVT (TH_ADDSL GPR:$rs1, GPR:$rs1, 2)), 1)>;
-def : Pat<(add (mul_oneuse GPR:$rs1, (XLenVT 18)), GPR:$rs2),
- (TH_ADDSL GPR:$rs2, (XLenVT (TH_ADDSL GPR:$rs1, GPR:$rs1, 3)), 1)>;
-def : Pat<(add (mul_oneuse GPR:$rs1, (XLenVT 12)), GPR:$rs2),
- (TH_ADDSL GPR:$rs2, (XLenVT (TH_ADDSL GPR:$rs1, GPR:$rs1, 1)), 2)>;
-def : Pat<(add (mul_oneuse GPR:$rs1, (XLenVT 20)), GPR:$rs2),
- (TH_ADDSL GPR:$rs2, (XLenVT (TH_ADDSL GPR:$rs1, GPR:$rs1, 2)), 2)>;
-def : Pat<(add (mul_oneuse GPR:$rs1, (XLenVT 36)), GPR:$rs2),
- (TH_ADDSL GPR:$rs2, (XLenVT (TH_ADDSL GPR:$rs1, GPR:$rs1, 3)), 2)>;
-def : Pat<(add (mul_oneuse GPR:$rs1, (XLenVT 24)), GPR:$rs2),
- (TH_ADDSL GPR:$rs2, (XLenVT (TH_ADDSL GPR:$rs1, GPR:$rs1, 1)), 3)>;
-def : Pat<(add (mul_oneuse GPR:$rs1, (XLenVT 40)), GPR:$rs2),
- (TH_ADDSL GPR:$rs2, (XLenVT (TH_ADDSL GPR:$rs1, GPR:$rs1, 2)), 3)>;
-def : Pat<(add (mul_oneuse GPR:$rs1, (XLenVT 72)), GPR:$rs2),
- (TH_ADDSL GPR:$rs2, (XLenVT (TH_ADDSL GPR:$rs1, GPR:$rs1, 3)), 3)>;
-
def : Pat<(add (XLenVT GPR:$r), CSImm12MulBy4:$i),
(TH_ADDSL GPR:$r, (XLenVT (ADDI (XLenVT X0), (SimmShiftRightBy2XForm CSImm12MulBy4:$i))), 2)>;
def : Pat<(add (XLenVT GPR:$r), CSImm12MulBy8:$i),
(TH_ADDSL GPR:$r, (XLenVT (ADDI (XLenVT X0), (SimmShiftRightBy3XForm CSImm12MulBy8:$i))), 3)>;
-def : Pat<(mul (XLenVT GPR:$r), C3LeftShift:$i),
- (SLLI (XLenVT (TH_ADDSL GPR:$r, GPR:$r, 1)),
- (TrailingZeros C3LeftShift:$i))>;
-def : Pat<(mul (XLenVT GPR:$r), C5LeftShift:$i),
- (SLLI (XLenVT (TH_ADDSL GPR:$r, GPR:$r, 2)),
- (TrailingZeros C5LeftShift:$i))>;
-def : Pat<(mul (XLenVT GPR:$r), C9LeftShift:$i),
- (SLLI (XLenVT (TH_ADDSL GPR:$r, GPR:$r, 3)),
- (TrailingZeros C9LeftShift:$i))>;
-
def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 200)),
(SLLI (XLenVT (TH_ADDSL (XLenVT (TH_ADDSL GPR:$r, GPR:$r, 2)),
(XLenVT (TH_ADDSL GPR:$r, GPR:$r, 2)), 2)), 3)>;
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td b/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
index ffe2b7e2712084..8a0bbf6abd334d 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
@@ -173,42 +173,6 @@ def BCLRIANDIMaskLow : SDNodeXForm<imm, [{
SDLoc(N), N->getValueType(0));
}]>;
-def C3LeftShift : PatLeaf<(imm), [{
- uint64_t C = N->getZExtValue();
- return C > 3 && (C >> llvm::countr_zero(C)) == 3;
-}]>;
-
-def C5LeftShift : PatLeaf<(imm), [{
- uint64_t C = N->getZExtValue();
- return C > 5 && (C >> llvm::countr_zero(C)) == 5;
-}]>;
-
-def C9LeftShift : PatLeaf<(imm), [{
- uint64_t C = N->getZExtValue();
- return C > 9 && (C >> llvm::countr_zero(C)) == 9;
-}]>;
-
-// Constant of the form (3 << C) where C is less than 32.
-def C3LeftShiftUW : PatLeaf<(imm), [{
- uint64_t C = N->getZExtValue();
- unsigned Shift = llvm::countr_zero(C);
- return 1 <= Shift && Shift < 32 && (C >> Shift) == 3;
-}]>;
-
-// Constant of the form (5 << C) where C is less than 32.
-def C5LeftShiftUW : PatLeaf<(imm), [{
- uint64_t C = N->getZExtValue();
- unsigned Shift = llvm::countr_zero(C);
- return 1 <= Shift && Shift < 32 && (C >> Shift) == 5;
-}]>;
-
-// Constant of the form (9 << C) where C is less than 32.
-def C9LeftShiftUW : PatLeaf<(imm), [{
- uint64_t C = N->getZExtValue();
- unsigned Shift = llvm::countr_zero(C);
- return 1 <= Shift && Shift < 32 && (C >> Shift) == 9;
-}]>;
-
def CSImm12MulBy4 : PatLeaf<(imm), [{
if (!N->hasOneUse())
return false;
@@ -693,25 +657,6 @@ foreach i = {1,2,3} in {
(shxadd pat:$rs1, GPR:$rs2)>;
}
-def : Pat<(add_like (mul_oneuse GPR:$rs1, (XLenVT 6)), GPR:$rs2),
- (SH1ADD (XLenVT (SH1ADD GPR:$rs1, GPR:$rs1)), GPR:$rs2)>;
-def : Pat<(add_like (mul_oneuse GPR:$rs1, (XLenVT 10)), GPR:$rs2),
- (SH1ADD (XLenVT (SH2ADD GPR:$rs1, GPR:$rs1)), GPR:$rs2)>;
-def : Pat<(add_like (mul_oneuse GPR:$rs1, (XLenVT 18)), GPR:$rs2),
- (SH1ADD (XLenVT (SH3ADD GPR:$rs1, GPR:$rs1)), GPR:$rs2)>;
-def : Pat<(add_like (mul_oneuse GPR:$rs1, (XLenVT 12)), GPR:$rs2),
- (SH2ADD (XLenVT (SH1ADD GPR:$rs1, GPR:$rs1)), GPR:$rs2)>;
-def : Pat<(add_like (mul_oneuse GPR:$rs1, (XLenVT 20)), GPR:$rs2),
- (SH2ADD (XLenVT (SH2ADD GPR:$rs1, GPR:$rs1)), GPR:$rs2)>;
-def : Pat<(add_like (mul_oneuse GPR:$rs1, (XLenVT 36)), GPR:$rs2),
- (SH2ADD (XLenVT (SH3ADD GPR:$rs1, GPR:$rs1)), GPR:$rs2)>;
-def : Pat<(add_like (mul_oneuse GPR:$rs1, (XLenVT 24)), GPR:$rs2),
- (SH3ADD (XLenVT (SH1ADD GPR:$rs1, GPR:$rs1)), GPR:$rs2)>;
-def : Pat<(add_like (mul_oneuse GPR:$rs1, (XLenVT 40)), GPR:$rs2),
- (SH3ADD (XLenVT (SH2ADD GPR:$rs1, GPR:$rs1)), GPR:$rs2)>;
-def : Pat<(add_like (mul_oneuse GPR:$rs1, (XLenVT 72)), GPR:$rs2),
- (SH3ADD (XLenVT (SH3ADD GPR:$rs1, GPR:$rs1)), GPR:$rs2)>;
-
def : Pat<(add_like (XLenVT GPR:$r), CSImm12MulBy4:$i),
(SH2ADD (XLenVT (ADDI (XLenVT X0), (SimmShiftRightBy2XForm CSImm12MulBy4:$i))),
GPR:$r)>;
@@ -719,16 +664,6 @@ def : Pat<(add_like (XLenVT GPR:$r), CSImm12MulBy8:$i),
(SH3ADD (XLenVT (ADDI (XLenVT X0), (SimmShiftRightBy3XForm CSImm12MulBy8:$i))),
GPR:$r)>;
-def : Pat<(mul (XLenVT GPR:$r), C3LeftShift:$i),
- (SLLI (XLenVT (SH1ADD GPR:$r, GPR:$r)),
- (TrailingZeros C3LeftShift:$i))>;
-def : Pat<(mul (XLenVT GPR:$r), C5LeftShift:$i),
- (SLLI (XLenVT (SH2ADD GPR:$r, GPR:$r)),
- (TrailingZeros C5LeftShift:$i))>;
-def : Pat<(mul (XLenVT GPR:$r), C9LeftShift:$i),
- (SLLI (XLenVT (SH3ADD GPR:$r, GPR:$r)),
- (TrailingZeros C9LeftShift:$i))>;
-
} // Predicates = [HasStdExtZba]
let Predicates = [HasStdExtZba, IsRV64] in {
@@ -780,15 +715,6 @@ def : Pat<(i64 (add_like_non_imm12 (and GPR:$rs1, 0x3FFFFFFFC), (XLenVT GPR:$rs2
def : Pat<(i64 (add_like_non_imm12 (and GPR:$rs1, 0x7FFFFFFF8), (XLenVT GPR:$rs2))),
(SH3ADD_UW (XLenVT (SRLI GPR:$rs1, 3)), GPR:$rs2)>;
-def : Pat<(i64 (mul (and_oneuse GPR:$r, 0xFFFFFFFF), C3LeftShiftUW:$i)),
- (SH1ADD (XLenVT (SLLI_UW GPR:$r, (TrailingZeros C3LeftShiftUW:$i))),
- (XLenVT (SLLI_UW GPR:$r, (TrailingZeros C3LeftShiftUW:$i))))>;
-def : Pat<(i64 (mul (and_oneuse GPR:$r, 0xFFFFFFFF), C5LeftShiftUW:$i)),
- (SH2ADD (XLenVT (SLLI_UW GPR:$r, (TrailingZeros C5LeftShiftUW:$i))),
- (XLenVT (SLLI_UW GPR:$r, (TrailingZeros C5LeftShiftUW:$i))))>;
-def : Pat<(i64 (mul (and_oneuse GPR:$r, 0xFFFFFFFF), C9LeftShiftUW:$i)),
- (SH3ADD (XLenVT (SLLI_UW GPR:$r, (TrailingZeros C9LeftShiftUW:$i))),
- (XLenVT (SLLI_UW GPR:$r, (TrailingZeros C9LeftShiftUW:$i))))>;
} // Predicates = [HasStdExtZba, IsRV64]
let Predicates = [HasStdExtZbcOrZbkc] in {
diff --git a/llvm/test/CodeGen/RISCV/addimm-mulimm.ll b/llvm/test/CodeGen/RISCV/addimm-mulimm.ll
index 736c8e7d55c75b..f40444dbb206a5 100644
--- a/llvm/test/CodeGen/RISCV/addimm-mulimm.ll
+++ b/llvm/test/CodeGen/RISCV/addimm-mulimm.ll
@@ -555,8 +555,9 @@ define i64 @add_mul_combine_infinite_loop(i64 %x) {
; RV32IMB-NEXT: sh3add a1, a1, a2
; RV32IMB-NEXT: sh1add a0, a0, a0
; RV32IMB-NEXT: slli a2, a0, 3
-; RV32IMB-NEXT: addi a0, a2, 2047
-; RV32IMB-NEXT: addi a0, a0, 1
+; RV32IMB-NEXT: li a3, 1
+; RV32IMB-NEXT: slli a3, a3, 11
+; RV32IMB-NEXT: sh3add a0, a0, a3
; RV32IMB-NEXT: sltu a2, a0, a2
; RV32IMB-NEXT: add a1, a1, a2
; RV32IMB-NEXT: ret
@@ -565,8 +566,8 @@ define i64 @add_mul_combine_infinite_loop(i64 %x) {
; RV64IMB: # %bb.0:
; RV64IMB-NEXT: addi a0, a0, 86
; RV64IMB-NEXT: sh1add a0, a0, a0
-; RV64IMB-NEXT: li a1, -16
-; RV64IMB-NEXT: sh3add a0, a0, a1
+; RV64IMB-NEXT: slli a0, a0, 3
+; RV64IMB-NEXT: addi a0, a0, -16
; RV64IMB-NEXT: ret
%tmp0 = mul i64 %x, 24
%tmp1 = add i64 %tmp0, 2048
diff --git a/llvm/test/CodeGen/RISCV/rv64-legal-i32/rv64zba.ll b/llvm/test/CodeGen/RISCV/rv64-legal-i32/rv64zba.ll
index 9f06a9dd124cef..87343a07647a43 100644
--- a/llvm/test/CodeGen/RISCV/rv64-legal-i32/rv64zba.ll
+++ b/llvm/test/CodeGen/RISCV/rv64-legal-i32/rv64zba.ll
@@ -647,9 +647,8 @@ define i64 @zext_mul12884901888(i32 signext %a) {
;
; RV64ZBA-LABEL: zext_mul12884901888:
; RV64ZBA: # %bb.0:
-; RV64ZBA-NEXT: andi a0, a0, -1
-; RV64ZBA-NEXT: sh1add a0, a0, a0
; RV64ZBA-NEXT: slli a0, a0, 32
+; RV64ZBA-NEXT: sh1add a0, a0, a0
; RV64ZBA-NEXT: ret
%b = zext i32 %a to i64
%c = mul i64 %b, 12884901888
@@ -670,9 +669,8 @@ define i64 @zext_mul21474836480(i32 signext %a) {
;
; RV64ZBA-LABEL: zext_mul21474836480:
; RV64ZBA: # %bb.0:
-; RV64ZBA-NEXT: andi a0, a0, -1
-; RV64ZBA-NEXT: sh2add a0, a0, a0
; RV64ZBA-NEXT: slli a0, a0, 32
+; RV64ZBA-NEXT: sh2add a0, a0, a0
; RV64ZBA-NEXT: ret
%b = zext i32 %a to i64
%c = mul i64 %b, 21474836480
@@ -693,9 +691,8 @@ define i64 @zext_mul38654705664(i32 signext %a) {
;
; RV64ZBA-LABEL: zext_mul38654705664:
; RV64ZBA: # %bb.0:
-; RV64ZBA-NEXT: andi a0, a0, -1
-; RV64ZBA-NEXT: sh3add a0, a0, a0
; RV64ZBA-NEXT: slli a0, a0, 32
+; RV64ZBA-NEXT: sh3add a0, a0, a0
; RV64ZBA-NEXT: ret
%b = zext i32 %a to i64
%c = mul i64 %b, 38654705664
diff --git a/llvm/test/CodeGen/RISCV/rv64zba.ll b/llvm/test/CodeGen/RISCV/rv64zba.ll
index 4eb493d642e853..f4c7ff3a4a1bc6 100644
--- a/llvm/test/CodeGen/RISCV/rv64zba.ll
+++ b/llvm/test/CodeGen/RISCV/rv64zba.ll
@@ -843,9 +843,8 @@ define i64 @zext_mul12884901888(i32 signext %a) {
;
; RV64ZBA-LABEL: zext_mul12884901888:
; RV64ZBA: # %bb.0:
-; RV64ZBA-NEXT: andi a0, a0, -1
-; RV64ZBA-NEXT: sh1add a0, a0, a0
; RV64ZBA-NEXT: slli a0, a0, 32
+; RV64ZBA-NEXT: sh1add a0, a0, a0
; RV64ZBA-NEXT: ret
%b = zext i32 %a to i64
%c = mul i64 %b, 12884901888
@@ -866,9 +865,8 @@ define i64 @zext_mul21474836480(i32 signext %a) {
;
; RV64ZBA-LABEL: zext_mul21474836480:
; RV64ZBA: # %bb.0:
-; RV64ZBA-NEXT: andi a0, a0, -1
-; RV64ZBA-NEXT: sh2add a0, a0, a0
; RV64ZBA-NEXT: slli a0, a0, 32
+; RV64ZBA-NEXT: sh2add a0, a0, a0
; RV64ZBA-NEXT: ret
%b = zext i32 %a to i64
%c = mul i64 %b, 21474836480
@@ -889,9 +887,8 @@ define i64 @zext_mul38654705664(i32 signext %a) {
;
; RV64ZBA-LABEL: zext_mul38654705664:
; RV64ZBA: # %bb.0:
-; RV64ZBA-NEXT: andi a0, a0, -1
-; RV64ZBA-NEXT: sh3add a0, a0, a0
; RV64ZBA-NEXT: slli a0, a0, 32
+; RV64ZBA-NEXT: sh3add a0, a0, a0
; RV64ZBA-NEXT: ret
%b = zext i32 %a to i64
%c = mul i64 %b, 38654705664
@@ -2616,7 +2613,7 @@ define i64 @regression(i32 signext %x, i32 signext %y) {
;
; RV64ZBA-LABEL: regression:
; RV64ZBA: # %bb.0:
-; RV64ZBA-NEXT: subw a0, a0, a1
+; RV64ZBA-NEXT: sub a0, a0, a1
; RV64ZBA-NEXT: slli.uw a0, a0, 3
; RV64ZBA-NEXT: sh1add a0, a0, a0
; RV64ZBA-NEXT: ret
|
preames
commented
Apr 24, 2024
| ; RV64IMB-NEXT: li a1, -16 | ||
| ; RV64IMB-NEXT: sh3add a0, a0, a1 | ||
| ; RV64IMB-NEXT: slli a0, a0, 3 | ||
| ; RV64IMB-NEXT: addi a0, a0, -16 |
There was a problem hiding this comment.
This case was previously picked up by the add_like (mul_one_use X, 24), Y pattern which didn't check whether Y is an immediate or not.
dtcxzyw
added a commit
to dtcxzyw/llvm-codegen-benchmark
that referenced
this pull request
Apr 24, 2024
preames
commented
Apr 24, 2024
| ; RV32IMB-NEXT: addi a0, a0, 1 | ||
| ; RV32IMB-NEXT: li a3, 1 | ||
| ; RV32IMB-NEXT: slli a3, a3, 11 | ||
| ; RV32IMB-NEXT: sh3add a0, a0, a3 |
There was a problem hiding this comment.
This case was previously not caught by the "add mul X, 24, Y" pattern because there's two multiplies by 24, and thus it failed the one use check. Instead, it went through generic "mul X, 3 << 2" expansion, and thus ended with the shift/add.
With the change, we hit the "(add_like_non_imm12 (shl GPR:$rs1, (XLenVT i)), GPR:$rs2)" pattern - which critically doesn't check if the immediate could be split across two addi. We should probably adjust this, but it seems a) minor, and b) very very separate. (And if we had zbb, this would be a bseti anyways.)
There was a problem hiding this comment.
With the change, we hit the "(add_like_non_imm12 (shl GPR:$rs1, (XLenVT i)), GPR:$rs2)" pattern - which critically doesn't check if the immediate could be split across two addi. We should probably adjust this, but it seems a) minor, and b) very very separate. (And if we had zbb, this would be a bseti anyways.)
This explanation makes sense to me.
|
This patch breaks bext pattern: Before: After: |
DAGCombiner pushes freeze through the |
dtcxzyw
added a commit
that referenced
this pull request
Apr 25, 2024
Inspired by #89966, this patch handles the special case `binop_allwusers<and> GPR:$rs1, 0xffffffff -> copy $rs1` to avoid creating redundant `andi rd, rs1, -1` instructions.
|
@dtcxzyw Thank you for finding and reporting this code gen difference. And also, yuck! This is actually the same issue I hit in X86 when trying to use the generic SHL_ADD node, so I've already investigated it a bit. In short, it really doesn't look like we have any good answers here. There's existing discussion around this here: #84921 (comment) My takeaway is that freeze in DAG is only half implemented, and there's really no clear good path forward. I very reluctantly think it may be time to give up here. No other target is actually undef correct for these cases, so maybe we shouldn't bother to be either. I've posted a patch for that here: #90097 If by chance anyone has a better idea, I'm definitely open to it. |
dtcxzyw
added a commit
to dtcxzyw/llvm-codegen-benchmark
that referenced
this pull request
Apr 25, 2024
It has been fixed :) Another regression (extracted from qemu): Before: After: We may fix this by handling |
dtcxzyw
approved these changes
Apr 25, 2024
Comment on lines
+13435
to
+13436
| if (X.getOpcode() == ISD::AND && isa<ConstantSDNode>(X.getOperand(1)) && | ||
| X.getConstantOperandVal(1) == UINT64_C(0xffffffff)) { |
There was a problem hiding this comment.
Use SDPatternMatch?
Suggested change
| if (X.getOpcode() == ISD::AND && isa<ConstantSDNode>(X.getOperand(1)) && | |
| X.getConstantOperandVal(1) == UINT64_C(0xffffffff)) { | |
| if (sd_match(X, m_And(m_Value(), m_SpecificInt(0xffffffff)))) { |
There was a problem hiding this comment.
I'm going to leave this as is, and then doing a single NFC to replace several usage examples.
Co-authored-by: Yingwei Zheng <dtcxzyw@qq.com>
See #91626 for this one. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This moves our last major category tablegen driven multiply strength reduction into the post legalize combine framework. The one slightly tricky bit is making sure that we use a leading shl if we can form a slli.uw, and trailing shl otherwise. Having the trailing shl is critical for shNadd matching, and folding any following sext.w.
As can be seen in the TD deltas, this allows us to kill off both the actual multiply patterns and the explicit add (mul X, C) Y patterns. The later are now handled by the generic shNadd matching code, with the exception of the THead only C=200 case because we don't (yet) have a multiply expansion with two shNadd + a shift.