From f2bbb4cb3e211de82987b280aa98565bd3ff6ce7 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng <dtcxzyw2333@gmail.com>
Date: Thu, 23 May 2024 01:26:48 +0800
Subject: [PATCH] [GISel][RISCV] Legalize `G_{U|S}DIVREM` (#93067)

This patch expands `G_{U|S}DIVREM` into `G_{U|S}DIV + G_{U|S}REM`.
`G_{U|S}DIVREM` is generated by the following fold:

https://github.com/llvm/llvm-project/blob/4ea21a0261cd8599a9ffa15f5c554ab0d4bbbe27/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp#L1410-L1471

It always folds `div + rem` pairs into `divrem` during pre-legalization.
I tried to change `isLegalOrBeforeLegalizer` to `isLegal`, but it
produced worse codegen on AArch64.
---
 .../Target/RISCV/GISel/RISCVLegalizerInfo.cpp |  3 +
 .../legalizer/legalize-div-rv32.mir           | 90 +++++++++++++++++++
 .../legalizer/legalize-div-rv64.mir           | 90 +++++++++++++++++++
 3 files changed, 183 insertions(+)

diff --git a/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp b/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
index a1d3aadb816a..c6d11b8a8bd7 100644
--- a/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
+++ b/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
@@ -349,6 +349,9 @@ RISCVLegalizerInfo::RISCVLegalizerInfo(const RISCVSubtarget &ST)
         .widenScalarToNextPow2(0);
   }
 
+  // TODO: Use libcall for sDoubleXLen.
+  getActionDefinitionsBuilder({G_UDIVREM, G_SDIVREM}).lower();
+
   auto &AbsActions = getActionDefinitionsBuilder(G_ABS);
   if (ST.hasStdExtZbb())
     AbsActions.customFor({s32, sXLen}).minScalar(0, sXLen);
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-div-rv32.mir b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-div-rv32.mir
index 4177a40e3826..26d8785afb47 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-div-rv32.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-div-rv32.mir
@@ -555,3 +555,93 @@ body:             |
     PseudoRET implicit $x10, implicit $x11
 
 ...
+---
+name:            udivrem_i32
+body:             |
+  bb.1.entry:
+    liveins: $x10, $x11
+
+    ; CHECK-I-LABEL: name: udivrem_i32
+    ; CHECK-I: liveins: $x10, $x11
+    ; CHECK-I-NEXT: {{  $}}
+    ; CHECK-I-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-I-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-I-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: $x10 = COPY [[COPY]](s32)
+    ; CHECK-I-NEXT: $x11 = COPY [[COPY1]](s32)
+    ; CHECK-I-NEXT: PseudoCALL target-flags(riscv-call) &__udivsi3, csr_ilp32_lp64, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: $x10 = COPY [[COPY]](s32)
+    ; CHECK-I-NEXT: $x11 = COPY [[COPY1]](s32)
+    ; CHECK-I-NEXT: PseudoCALL target-flags(riscv-call) &__umodsi3, csr_ilp32_lp64, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-I-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[COPY2]], [[COPY3]]
+    ; CHECK-I-NEXT: $x10 = COPY [[ADD]](s32)
+    ; CHECK-I-NEXT: PseudoRET implicit $x10
+    ;
+    ; CHECK-M-LABEL: name: udivrem_i32
+    ; CHECK-M: liveins: $x10, $x11
+    ; CHECK-M-NEXT: {{  $}}
+    ; CHECK-M-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-M-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-M-NEXT: [[UDIV:%[0-9]+]]:_(s32) = G_UDIV [[COPY]], [[COPY1]]
+    ; CHECK-M-NEXT: [[UREM:%[0-9]+]]:_(s32) = G_UREM [[COPY]], [[COPY1]]
+    ; CHECK-M-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[UDIV]], [[UREM]]
+    ; CHECK-M-NEXT: $x10 = COPY [[ADD]](s32)
+    ; CHECK-M-NEXT: PseudoRET implicit $x10
+    %0:_(s32) = COPY $x10
+    %1:_(s32) = COPY $x11
+    %2:_(s32), %3:_(s32) = G_UDIVREM %0, %1
+    %4:_(s32) = G_ADD %2, %3
+    $x10 = COPY %4(s32)
+    PseudoRET implicit $x10
+
+...
+---
+name:            sdivrem_i32
+body:             |
+  bb.1.entry:
+    liveins: $x10, $x11
+
+    ; CHECK-I-LABEL: name: sdivrem_i32
+    ; CHECK-I: liveins: $x10, $x11
+    ; CHECK-I-NEXT: {{  $}}
+    ; CHECK-I-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-I-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-I-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: $x10 = COPY [[COPY]](s32)
+    ; CHECK-I-NEXT: $x11 = COPY [[COPY1]](s32)
+    ; CHECK-I-NEXT: PseudoCALL target-flags(riscv-call) &__divsi3, csr_ilp32_lp64, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: $x10 = COPY [[COPY]](s32)
+    ; CHECK-I-NEXT: $x11 = COPY [[COPY1]](s32)
+    ; CHECK-I-NEXT: PseudoCALL target-flags(riscv-call) &__modsi3, csr_ilp32_lp64, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-I-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[COPY2]], [[COPY3]]
+    ; CHECK-I-NEXT: $x10 = COPY [[ADD]](s32)
+    ; CHECK-I-NEXT: PseudoRET implicit $x10
+    ;
+    ; CHECK-M-LABEL: name: sdivrem_i32
+    ; CHECK-M: liveins: $x10, $x11
+    ; CHECK-M-NEXT: {{  $}}
+    ; CHECK-M-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-M-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-M-NEXT: [[SDIV:%[0-9]+]]:_(s32) = G_SDIV [[COPY]], [[COPY1]]
+    ; CHECK-M-NEXT: [[SREM:%[0-9]+]]:_(s32) = G_SREM [[COPY]], [[COPY1]]
+    ; CHECK-M-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[SDIV]], [[SREM]]
+    ; CHECK-M-NEXT: $x10 = COPY [[ADD]](s32)
+    ; CHECK-M-NEXT: PseudoRET implicit $x10
+    %0:_(s32) = COPY $x10
+    %1:_(s32) = COPY $x11
+    %2:_(s32), %3:_(s32) = G_SDIVREM %0, %1
+    %4:_(s32) = G_ADD %2, %3
+    $x10 = COPY %4(s32)
+    PseudoRET implicit $x10
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-div-rv64.mir b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-div-rv64.mir
index 492f9530997c..bbbe38f695d2 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-div-rv64.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-div-rv64.mir
@@ -655,3 +655,93 @@ body:             |
     PseudoRET implicit $x10, implicit $x11
 
 ...
+---
+name:            udivrem_i64
+body:             |
+  bb.1.entry:
+    liveins: $x10, $x11
+
+    ; CHECK-I-LABEL: name: udivrem_i64
+    ; CHECK-I: liveins: $x10, $x11
+    ; CHECK-I-NEXT: {{  $}}
+    ; CHECK-I-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; CHECK-I-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x11
+    ; CHECK-I-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: $x10 = COPY [[COPY]](s64)
+    ; CHECK-I-NEXT: $x11 = COPY [[COPY1]](s64)
+    ; CHECK-I-NEXT: PseudoCALL target-flags(riscv-call) &__udivdi3, csr_ilp32_lp64, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: $x10 = COPY [[COPY]](s64)
+    ; CHECK-I-NEXT: $x11 = COPY [[COPY1]](s64)
+    ; CHECK-I-NEXT: PseudoCALL target-flags(riscv-call) &__umoddi3, csr_ilp32_lp64, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: [[COPY3:%[0-9]+]]:_(s64) = COPY $x10
+    ; CHECK-I-NEXT: [[ADD:%[0-9]+]]:_(s64) = G_ADD [[COPY2]], [[COPY3]]
+    ; CHECK-I-NEXT: $x10 = COPY [[ADD]](s64)
+    ; CHECK-I-NEXT: PseudoRET implicit $x10
+    ;
+    ; CHECK-M-LABEL: name: udivrem_i64
+    ; CHECK-M: liveins: $x10, $x11
+    ; CHECK-M-NEXT: {{  $}}
+    ; CHECK-M-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; CHECK-M-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x11
+    ; CHECK-M-NEXT: [[UDIV:%[0-9]+]]:_(s64) = G_UDIV [[COPY]], [[COPY1]]
+    ; CHECK-M-NEXT: [[UREM:%[0-9]+]]:_(s64) = G_UREM [[COPY]], [[COPY1]]
+    ; CHECK-M-NEXT: [[ADD:%[0-9]+]]:_(s64) = G_ADD [[UDIV]], [[UREM]]
+    ; CHECK-M-NEXT: $x10 = COPY [[ADD]](s64)
+    ; CHECK-M-NEXT: PseudoRET implicit $x10
+    %0:_(s64) = COPY $x10
+    %1:_(s64) = COPY $x11
+    %2:_(s64), %3:_(s64) = G_UDIVREM %0, %1
+    %4:_(s64) = G_ADD %2, %3
+    $x10 = COPY %4(s64)
+    PseudoRET implicit $x10
+
+...
+---
+name:            sdivrem_i64
+body:             |
+  bb.1.entry:
+    liveins: $x10, $x11
+
+    ; CHECK-I-LABEL: name: sdivrem_i64
+    ; CHECK-I: liveins: $x10, $x11
+    ; CHECK-I-NEXT: {{  $}}
+    ; CHECK-I-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; CHECK-I-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x11
+    ; CHECK-I-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: $x10 = COPY [[COPY]](s64)
+    ; CHECK-I-NEXT: $x11 = COPY [[COPY1]](s64)
+    ; CHECK-I-NEXT: PseudoCALL target-flags(riscv-call) &__divdi3, csr_ilp32_lp64, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: $x10 = COPY [[COPY]](s64)
+    ; CHECK-I-NEXT: $x11 = COPY [[COPY1]](s64)
+    ; CHECK-I-NEXT: PseudoCALL target-flags(riscv-call) &__moddi3, csr_ilp32_lp64, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; CHECK-I-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $x2, implicit $x2
+    ; CHECK-I-NEXT: [[COPY3:%[0-9]+]]:_(s64) = COPY $x10
+    ; CHECK-I-NEXT: [[ADD:%[0-9]+]]:_(s64) = G_ADD [[COPY2]], [[COPY3]]
+    ; CHECK-I-NEXT: $x10 = COPY [[ADD]](s64)
+    ; CHECK-I-NEXT: PseudoRET implicit $x10
+    ;
+    ; CHECK-M-LABEL: name: sdivrem_i64
+    ; CHECK-M: liveins: $x10, $x11
+    ; CHECK-M-NEXT: {{  $}}
+    ; CHECK-M-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; CHECK-M-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x11
+    ; CHECK-M-NEXT: [[SDIV:%[0-9]+]]:_(s64) = G_SDIV [[COPY]], [[COPY1]]
+    ; CHECK-M-NEXT: [[SREM:%[0-9]+]]:_(s64) = G_SREM [[COPY]], [[COPY1]]
+    ; CHECK-M-NEXT: [[ADD:%[0-9]+]]:_(s64) = G_ADD [[SDIV]], [[SREM]]
+    ; CHECK-M-NEXT: $x10 = COPY [[ADD]](s64)
+    ; CHECK-M-NEXT: PseudoRET implicit $x10
+    %0:_(s64) = COPY $x10
+    %1:_(s64) = COPY $x11
+    %2:_(s64), %3:_(s64) = G_SDIVREM %0, %1
+    %4:_(s64) = G_ADD %2, %3
+    $x10 = COPY %4(s64)
+    PseudoRET implicit $x10
+
+...