[llvm] [RISCV] add more generic macrofusions (PR #151140)
Daniel Henrique Barboza via llvm-commits
llvm-commits at lists.llvm.org
Wed Jul 30 05:17:44 PDT 2025
https://github.com/danielhb updated https://github.com/llvm/llvm-project/pull/151140
>From 51d5c2f78e076ead6e2f43179bb3a83f6143fb32 Mon Sep 17 00:00:00 2001
From: Daniel Henrique Barboza <dbarboza at ventanamicro.com>
Date: Tue, 4 Mar 2025 10:22:17 -0800
Subject: [PATCH 1/2] [RISCV] add more generic macrofusions
These are some macrofusions that are used internally in Ventana in an
yet not upstreamed processor. Figured it would be good to contribute
them ahead of the processor to allow the community to also use them in
their own processors, while also alleaviting our own downstream upkeep.
The macrofusions being added are:
- add+lw
- addi+ld, addi+lw
- adduw+lw
- auipc+ld
- bfext (slli+srli)
- lui+ld
- shXadd+load, where X=1,2,3 and load=lb,lh,lw,ld
- shXadduw+load, where X=1,2,3 and load=lb,lh,lw,ld
---
llvm/lib/Target/RISCV/RISCVMacroFusion.td | 91 ++++
llvm/test/CodeGen/RISCV/features-info.ll | 9 +
llvm/test/CodeGen/RISCV/macro-fusions.mir | 560 ++++++++++++++++++++++
3 files changed, 660 insertions(+)
diff --git a/llvm/lib/Target/RISCV/RISCVMacroFusion.td b/llvm/lib/Target/RISCV/RISCVMacroFusion.td
index 875a93d09a2c6..34ad042c565d7 100644
--- a/llvm/lib/Target/RISCV/RISCVMacroFusion.td
+++ b/llvm/lib/Target/RISCV/RISCVMacroFusion.td
@@ -91,3 +91,94 @@ def TuneLDADDFusion
CheckIsImmOperand<2>,
CheckImmOperand<2, 0>
]>>;
+
+// Fuse add with lw:
+// add rd, rs1, rs2
+// lw rd, 0(rd)
+def TuneADDLWFusion
+ : SimpleFusion<"add-lw-fusion", "HasADDLWFusion", "Enable ADD+LW macrofusion",
+ CheckOpcode<[ADD]>,
+ CheckAll<[
+ CheckOpcode<[LW]>,
+ CheckIsImmOperand<2>,
+ CheckImmOperand<2, 0>
+ ]>>;
+
+// Fuse AUIPC followed by LD:
+// auipc rd, imm20
+// ld rd, imm12(rd)
+def TuneAUIPCLDFusion
+ : SimpleFusion<"auipc-ld-fusion", "HasAUIPCLDFusion",
+ "Enable AUIPC+LD macrofusion",
+ CheckOpcode<[AUIPC]>,
+ CheckOpcode<[LD]>>;
+
+// Fuse LUI followed by LD:
+// lui rd, imm[31:12]
+// ld rd, imm12(rd)
+def TuneLUILDFusion
+ : SimpleFusion<"lui-ld-fusion", "HasLUILDFusion",
+ "Enable LUI+LD macrofusion",
+ CheckOpcode<[LUI]>,
+ CheckOpcode<[LD]>>;
+
+// Bitfield extract fusion: similar to TuneShiftedZExtWFusion
+// but without the immediate restriction
+// slli rd, rs1, imm12
+// srli rd, rd, imm12
+def TuneBFExtFusion
+ : SimpleFusion<"bfext-fusion", "HasBFExtFusion",
+ "Enable SLLI+SRLI (bitfield extract) macrofusion",
+ CheckAll<[
+ CheckOpcode<[SLLI]>,
+ CheckIsImmOperand<2>,
+ ]>,
+ CheckAll<[
+ CheckOpcode<[SRLI]>,
+ CheckIsImmOperand<2>,
+ ]>>;
+
+// Fuse ADDI followed by LD
+// addi rd, rs1, imm12
+// ld rd, imm12(rd)
+def TuneADDILDFusion
+ : SimpleFusion<"addi-ld-fusion", "HasADDILDFusion",
+ "Enable ADDI+LD macrofusion",
+ CheckOpcode<[ADDI]>,
+ CheckOpcode<[LD]>>;
+
+// Fuse ADDI followed by LW
+// addi rd, rs1, imm12
+// lw rd, imm12(rd)
+def TuneADDILWFusion
+ : SimpleFusion<"addi-lw-fusion", "HasADDILWFusion",
+ "Enable ADDI+LW macrofusion",
+ CheckOpcode<[ADDI]>,
+ CheckOpcode<[LW]>>;
+
+// Fuse ADDUW followed by LW
+// adduw rd, rs1, rs2
+// lw rd, imm12(rd)
+def TuneADDUWLWFusion
+ : SimpleFusion<"adduw-lw-fusion", "HasADDUWLWFusion",
+ "Enable ADD_UW+LW macrofusion",
+ CheckOpcode<[ADD_UW]>,
+ CheckOpcode<[LW]>>;
+
+// Fuse SHXADD followed by a load (lb, lh, lw, ld)
+// shXadd rd, rs1, rs2
+// load rd, imm12(rd)
+def TuneSHXADDLoadFusion
+ : SimpleFusion<"shxadd-load-fusion", "HasSHXADDLoadFusion",
+ "Enable SH(1|2|3)ADD + load macrofusion",
+ CheckOpcode<[SH1ADD, SH2ADD, SH3ADD]>,
+ CheckOpcode<[LB, LH, LW, LD]>>;
+
+// Fuse SHXADD.UW followed by a load (lb, lh, lw, ld)
+// shXadd.uw rd, rs1, rs2
+// load rd, imm12(rd)
+def TuneSHXADDUWLoadFusion
+ : SimpleFusion<"shxadduw-load-fusion", "HasSHXADDUWLoadFusion",
+ "Enable SH(1|2|3)ADDUW + load macrofusion",
+ CheckOpcode<[SH1ADD_UW, SH2ADD_UW, SH3ADD_UW]>,
+ CheckOpcode<[LB, LH, LW, LD]>>;
diff --git a/llvm/test/CodeGen/RISCV/features-info.ll b/llvm/test/CodeGen/RISCV/features-info.ll
index b94665b718ae7..de14b3355ac07 100644
--- a/llvm/test/CodeGen/RISCV/features-info.ll
+++ b/llvm/test/CodeGen/RISCV/features-info.ll
@@ -6,9 +6,15 @@
; CHECK-NEXT: 32bit - Implements RV32.
; CHECK-NEXT: 64bit - Implements RV64.
; CHECK-NEXT: a - 'A' (Atomic Instructions).
+; CHECK-NEXT: add-lw-fusion - Enable ADD+LW macrofusion.
+; CHECK-NEXT: addi-ld-fusion - Enable ADDI+LD macrofusion.
+; CHECK-NEXT: addi-lw-fusion - Enable ADDI+LW macrofusion.
+; CHECK-NEXT: adduw-lw-fusion - Enable ADD_UW+LW macrofusion.
; CHECK-NEXT: andes45 - Andes 45-Series processors.
; CHECK-NEXT: auipc-addi-fusion - Enable AUIPC+ADDI macrofusion.
+; CHECK-NEXT: auipc-ld-fusion - Enable AUIPC+LD macrofusion.
; CHECK-NEXT: b - 'B' (the collection of the Zba, Zbb, Zbs extensions).
+; CHECK-NEXT: bfext-fusion - Enable SLLI+SRLI (bitfield extract) macrofusion.
; CHECK-NEXT: c - 'C' (Compressed Instructions).
; CHECK-NEXT: conditional-cmv-fusion - Enable branch+c.mv fusion.
; CHECK-NEXT: d - 'D' (Double-Precision Floating-Point).
@@ -58,6 +64,7 @@
; CHECK-NEXT: ld-add-fusion - Enable LD+ADD macrofusion.
; CHECK-NEXT: log-vrgather - Has vrgather.vv with LMUL*log2(LMUL) latency
; CHECK-NEXT: lui-addi-fusion - Enable LUI+ADDI macro fusion.
+; CHECK-NEXT: lui-ld-fusion - Enable LUI+LD macrofusion.
; CHECK-NEXT: m - 'M' (Integer Multiplication and Division).
; CHECK-NEXT: mips-p8700 - MIPS p8700 processor.
; CHECK-NEXT: no-default-unroll - Disable default unroll preference..
@@ -130,6 +137,8 @@
; CHECK-NEXT: shvsatpa - 'Shvsatpa' (vsatp supports all modes supported by satp).
; CHECK-NEXT: shvstvala - 'Shvstvala' (vstval provides all needed values).
; CHECK-NEXT: shvstvecd - 'Shvstvecd' (vstvec supports Direct mode).
+; CHECK-NEXT: shxadd-load-fusion - Enable SH(1|2|3)ADD + load macrofusion.
+; CHECK-NEXT: shxadduw-load-fusion - Enable SH(1|2|3)ADDUW + load macrofusion.
; CHECK-NEXT: sifive7 - SiFive 7-Series processors.
; CHECK-NEXT: smaia - 'Smaia' (Advanced Interrupt Architecture Machine Level).
; CHECK-NEXT: smcdeleg - 'Smcdeleg' (Counter Delegation Machine Level).
diff --git a/llvm/test/CodeGen/RISCV/macro-fusions.mir b/llvm/test/CodeGen/RISCV/macro-fusions.mir
index 13464141ce27e..a8e6d887d57f8 100644
--- a/llvm/test/CodeGen/RISCV/macro-fusions.mir
+++ b/llvm/test/CodeGen/RISCV/macro-fusions.mir
@@ -2,7 +2,12 @@
# RUN: llc -mtriple=riscv64-linux-gnu -x=mir < %s \
# RUN: -debug-only=machine-scheduler -start-before=machine-scheduler 2>&1 \
# RUN: -mattr=+lui-addi-fusion,+auipc-addi-fusion,+zexth-fusion,+zextw-fusion,+shifted-zextw-fusion,+ld-add-fusion \
+# RUN: -mattr=+add-lw-fusion,+auipc-ld-fusion,+lui-ld-fusion,+addi-ld-fusion,+addi-lw-fusion \
+# RUN: -mattr=+zba,+adduw-lw-fusion,+shxadd-load-fusion,+shxadduw-load-fusion \
# RUN: | FileCheck %s
+# RUN: llc -mtriple=riscv64-linux-gnu -x=mir < %s \
+# RUN: -debug-only=machine-scheduler -start-before=machine-scheduler 2>&1 \
+# RUN: -mattr=+zba,+bfext-fusion | FileCheck --check-prefixes=CHECK-BFEXT %s
# CHECK: lui_addi:%bb.0
# CHECK: Macro fuse: {{.*}}LUI - ADDI
@@ -174,3 +179,558 @@ body: |
$x11 = COPY %5
PseudoRET
...
+
+# CHECK: add_lw
+# CHECK: Macro fuse: {{.*}}ADD - LW
+---
+name: add_lw
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LW %3, 0
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: auipc_ld
+# CHECK: Macro fuse: {{.*}}AUIPC - LD
+---
+name: auipc_ld
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10
+ %1:gpr = COPY $x10
+ %2:gpr = AUIPC 1
+ %3:gpr = XORI %1, 2
+ %4:gpr = LD %2, 4
+ $x10 = COPY %3
+ $x11 = COPY %4
+ PseudoRET
+...
+
+# CHECK: lui_ld
+# CHECK: Macro fuse: {{.*}}LUI - LD
+---
+name: lui_ld
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10
+ %1:gpr = COPY $x10
+ %2:gpr = LUI 1
+ %3:gpr = XORI %1, 2
+ %4:gpr = LD %2, 4
+ $x10 = COPY %3
+ $x11 = COPY %4
+ PseudoRET
+...
+
+# CHECK-BFEXT: bitfield_extract
+# CHECK-BFEXT: Macro fuse: {{.*}}SLLI - SRLI
+---
+name: bitfield_extract
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10
+ %1:gpr = COPY $x10
+ %2:gpr = SLLI %1, 31
+ %3:gpr = XORI %1, 3
+ %4:gpr = SRLI %2, 48
+ $x10 = COPY %3
+ $x11 = COPY %4
+ PseudoRET
+...
+
+# CHECK: addi_ld
+# CHECK: Macro fuse: {{.*}}ADDI - LD
+---
+name: addi_ld
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = ADDI %1, 8
+ %4:gpr = XORI %2, 3
+ %5:gpr = LD %3, 0
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: addi_lw
+# CHECK: Macro fuse: {{.*}}ADDI - LW
+---
+name: addi_lw
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = ADDI %1, 8
+ %4:gpr = XORI %2, 3
+ %5:gpr = LW %3, 0
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: adduw_lw
+# CHECK: Macro fuse: {{.*}}ADD_UW - LW
+---
+name: adduw_lw
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LW %3, 0
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1add_lb
+# CHECK: Macro fuse: {{.*}}SH1ADD - LB
+---
+name: sh1add_lb
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LB %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2add_lb
+# CHECK: Macro fuse: {{.*}}SH2ADD - LB
+---
+name: sh2add_lb
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LB %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3add_lb
+# CHECK: Macro fuse: {{.*}}SH3ADD - LB
+---
+name: sh3add_lb
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LB %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1add_lh
+# CHECK: Macro fuse: {{.*}}SH1ADD - LH
+---
+name: sh1add_lh
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LH %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2add_lh
+# CHECK: Macro fuse: {{.*}}SH2ADD - LH
+---
+name: sh2add_lh
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LH %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3add_lh
+# CHECK: Macro fuse: {{.*}}SH3ADD - LH
+---
+name: sh3add_lh
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LH %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1add_lw
+# CHECK: Macro fuse: {{.*}}SH1ADD - LW
+---
+name: sh1add_lw
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LW %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2add_lw
+# CHECK: Macro fuse: {{.*}}SH2ADD - LW
+---
+name: sh2add_lw
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LW %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3add_lw
+# CHECK: Macro fuse: {{.*}}SH3ADD - LW
+---
+name: sh3add_lw
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LW %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1add_ld
+# CHECK: Macro fuse: {{.*}}SH1ADD - LD
+---
+name: sh1add_ld
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LD %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2add_ld
+# CHECK: Macro fuse: {{.*}}SH2ADD - LD
+---
+name: sh2add_ld
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LD %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3add_ld
+# CHECK: Macro fuse: {{.*}}SH3ADD - LD
+---
+name: sh3add_ld
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LD %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1adduw_lb
+# CHECK: Macro fuse: {{.*}}SH1ADD_UW - LB
+---
+name: sh1adduw_lb
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LB %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2adduw_lb
+# CHECK: Macro fuse: {{.*}}SH2ADD_UW - LB
+---
+name: sh2adduw_lb
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LB %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3adduw_lb
+# CHECK: Macro fuse: {{.*}}SH3ADD_UW - LB
+---
+name: sh3adduw_lb
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LB %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1adduw_lh
+# CHECK: Macro fuse: {{.*}}SH1ADD_UW - LH
+---
+name: sh1adduw_lh
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LH %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2adduw_lh
+# CHECK: Macro fuse: {{.*}}SH2ADD_UW - LH
+---
+name: sh2adduw_lh
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LH %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3adduw_lh
+# CHECK: Macro fuse: {{.*}}SH3ADD_UW - LH
+---
+name: sh3adduw_lh
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LH %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1adduw_lw
+# CHECK: Macro fuse: {{.*}}SH1ADD_UW - LW
+---
+name: sh1adduw_lw
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LW %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2adduw_lw
+# CHECK: Macro fuse: {{.*}}SH2ADD_UW - LW
+---
+name: sh2adduw_lw
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LW %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3adduw_lw
+# CHECK: Macro fuse: {{.*}}SH3ADD_UW - LW
+---
+name: sh3adduw_lw
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LW %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1adduw_ld
+# CHECK: Macro fuse: {{.*}}SH1ADD_UW - LD
+---
+name: sh1adduw_ld
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LD %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2adduw_ld
+# CHECK: Macro fuse: {{.*}}SH2ADD_UW - LD
+---
+name: sh2adduw_ld
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LD %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3adduw_ld
+# CHECK: Macro fuse: {{.*}}SH3ADD_UW - LD
+---
+name: sh3adduw_ld
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LD %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
>From 3a1f13c8acc1ed5fd23f35cdf8e6eaba5ed72207 Mon Sep 17 00:00:00 2001
From: Daniel Henrique Barboza <dbarboza at ventanamicro.com>
Date: Tue, 29 Jul 2025 13:23:16 -0700
Subject: [PATCH 2/2] Add extra load macrofusion cases and other fixes
- add missing macrofusions in veyron-v1 processor def;
- fix ADD_UW/add.uw comment;
- add ADD + lb/lh/lbu/lhu/lwu macrofusions;
- add shXADD + lbu/lhu/lwu macrofusions;
- add shXADD_UW + lbu/lhu/lwu macrofusions.
---
llvm/lib/Target/RISCV/RISCVMacroFusion.td | 24 +-
llvm/lib/Target/RISCV/RISCVProcessors.td | 5 +-
llvm/test/CodeGen/RISCV/features-info.ll | 1 +
llvm/test/CodeGen/RISCV/macro-fusions.mir | 416 +++++++++++++++++++++-
4 files changed, 438 insertions(+), 8 deletions(-)
diff --git a/llvm/lib/Target/RISCV/RISCVMacroFusion.td b/llvm/lib/Target/RISCV/RISCVMacroFusion.td
index 34ad042c565d7..8df756b44bf89 100644
--- a/llvm/lib/Target/RISCV/RISCVMacroFusion.td
+++ b/llvm/lib/Target/RISCV/RISCVMacroFusion.td
@@ -104,6 +104,18 @@ def TuneADDLWFusion
CheckImmOperand<2, 0>
]>>;
+// Fuse add followed by a load (lb, lh, lw, ld, lbu, lhu, lwu):
+// add rd, rs1, rs2
+// load rd, 0(rd)
+def TuneADDLoadFusion
+ : SimpleFusion<"add-load-fusion", "HasADDLoadFusion", "Enable ADD + load macrofusion",
+ CheckOpcode<[ADD]>,
+ CheckAll<[
+ CheckOpcode<[LB, LH, LW, LD, LBU, LHU, LWU]>,
+ CheckIsImmOperand<2>,
+ CheckImmOperand<2, 0>
+ ]>>;
+
// Fuse AUIPC followed by LD:
// auipc rd, imm20
// ld rd, imm12(rd)
@@ -156,8 +168,8 @@ def TuneADDILWFusion
CheckOpcode<[ADDI]>,
CheckOpcode<[LW]>>;
-// Fuse ADDUW followed by LW
-// adduw rd, rs1, rs2
+// Fuse ADD_UW followed by LW
+// add.uw rd, rs1, rs2
// lw rd, imm12(rd)
def TuneADDUWLWFusion
: SimpleFusion<"adduw-lw-fusion", "HasADDUWLWFusion",
@@ -165,20 +177,20 @@ def TuneADDUWLWFusion
CheckOpcode<[ADD_UW]>,
CheckOpcode<[LW]>>;
-// Fuse SHXADD followed by a load (lb, lh, lw, ld)
+// Fuse SHXADD followed by a load (lb, lh, lw, ld, lbu, lhu, lwu)
// shXadd rd, rs1, rs2
// load rd, imm12(rd)
def TuneSHXADDLoadFusion
: SimpleFusion<"shxadd-load-fusion", "HasSHXADDLoadFusion",
"Enable SH(1|2|3)ADD + load macrofusion",
CheckOpcode<[SH1ADD, SH2ADD, SH3ADD]>,
- CheckOpcode<[LB, LH, LW, LD]>>;
+ CheckOpcode<[LB, LH, LW, LD, LBU, LHU, LWU]>>;
-// Fuse SHXADD.UW followed by a load (lb, lh, lw, ld)
+// Fuse SHXADD_UW followed by a load (lb, lh, lw, ld, lbu, lhu, lwu)
// shXadd.uw rd, rs1, rs2
// load rd, imm12(rd)
def TuneSHXADDUWLoadFusion
: SimpleFusion<"shxadduw-load-fusion", "HasSHXADDUWLoadFusion",
"Enable SH(1|2|3)ADDUW + load macrofusion",
CheckOpcode<[SH1ADD_UW, SH2ADD_UW, SH3ADD_UW]>,
- CheckOpcode<[LB, LH, LW, LD]>>;
+ CheckOpcode<[LB, LH, LW, LD, LBU, LHU, LWU]>>;
diff --git a/llvm/lib/Target/RISCV/RISCVProcessors.td b/llvm/lib/Target/RISCV/RISCVProcessors.td
index 838edf6c57250..ab43d3fa3a472 100644
--- a/llvm/lib/Target/RISCV/RISCVProcessors.td
+++ b/llvm/lib/Target/RISCV/RISCVProcessors.td
@@ -595,7 +595,10 @@ def VENTANA_VEYRON_V1 : RISCVProcessorModel<"veyron-v1",
TuneZExtHFusion,
TuneZExtWFusion,
TuneShiftedZExtWFusion,
- TuneLDADDFusion]> {
+ TuneLDADDFusion,
+ TuneADDLWFusion,
+ TuneAUIPCLDFusion,
+ TuneLUILDFusion]> {
let MVendorID = 0x61f;
let MArchID = 0x8000000000010000;
let MImpID = 0x111;
diff --git a/llvm/test/CodeGen/RISCV/features-info.ll b/llvm/test/CodeGen/RISCV/features-info.ll
index de14b3355ac07..ada77e5fd2fbb 100644
--- a/llvm/test/CodeGen/RISCV/features-info.ll
+++ b/llvm/test/CodeGen/RISCV/features-info.ll
@@ -6,6 +6,7 @@
; CHECK-NEXT: 32bit - Implements RV32.
; CHECK-NEXT: 64bit - Implements RV64.
; CHECK-NEXT: a - 'A' (Atomic Instructions).
+; CHECK-NEXT: add-load-fusion - Enable ADD + load macrofusion.
; CHECK-NEXT: add-lw-fusion - Enable ADD+LW macrofusion.
; CHECK-NEXT: addi-ld-fusion - Enable ADDI+LD macrofusion.
; CHECK-NEXT: addi-lw-fusion - Enable ADDI+LW macrofusion.
diff --git a/llvm/test/CodeGen/RISCV/macro-fusions.mir b/llvm/test/CodeGen/RISCV/macro-fusions.mir
index a8e6d887d57f8..587830c5f2947 100644
--- a/llvm/test/CodeGen/RISCV/macro-fusions.mir
+++ b/llvm/test/CodeGen/RISCV/macro-fusions.mir
@@ -2,7 +2,7 @@
# RUN: llc -mtriple=riscv64-linux-gnu -x=mir < %s \
# RUN: -debug-only=machine-scheduler -start-before=machine-scheduler 2>&1 \
# RUN: -mattr=+lui-addi-fusion,+auipc-addi-fusion,+zexth-fusion,+zextw-fusion,+shifted-zextw-fusion,+ld-add-fusion \
-# RUN: -mattr=+add-lw-fusion,+auipc-ld-fusion,+lui-ld-fusion,+addi-ld-fusion,+addi-lw-fusion \
+# RUN: -mattr=+add-load-fusion,+add-lw-fusion,+auipc-ld-fusion,+lui-ld-fusion,+addi-ld-fusion,+addi-lw-fusion \
# RUN: -mattr=+zba,+adduw-lw-fusion,+shxadd-load-fusion,+shxadduw-load-fusion \
# RUN: | FileCheck %s
# RUN: llc -mtriple=riscv64-linux-gnu -x=mir < %s \
@@ -180,6 +180,42 @@ body: |
PseudoRET
...
+# CHECK: add_lb
+# CHECK: Macro fuse: {{.*}}ADD - LB
+---
+name: add_lb
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LB %3, 0
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: add_lh
+# CHECK: Macro fuse: {{.*}}ADD - LH
+---
+name: add_lh
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LH %3, 0
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
# CHECK: add_lw
# CHECK: Macro fuse: {{.*}}ADD - LW
---
@@ -198,6 +234,60 @@ body: |
PseudoRET
...
+# CHECK: add_lbu
+# CHECK: Macro fuse: {{.*}}ADD - LBU
+---
+name: add_lbu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LBU %3, 0
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: add_lhu
+# CHECK: Macro fuse: {{.*}}ADD - LHU
+---
+name: add_lhu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LHU %3, 0
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: add_lwu
+# CHECK: Macro fuse: {{.*}}ADD - LWU
+---
+name: add_lwu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LWU %3, 0
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
# CHECK: auipc_ld
# CHECK: Macro fuse: {{.*}}AUIPC - LD
---
@@ -519,6 +609,168 @@ body: |
PseudoRET
...
+# CHECK: sh1add_lbu
+# CHECK: Macro fuse: {{.*}}SH1ADD - LBU
+---
+name: sh1add_lbu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LBU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2add_lbu
+# CHECK: Macro fuse: {{.*}}SH2ADD - LBU
+---
+name: sh2add_lbu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LBU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3add_lbu
+# CHECK: Macro fuse: {{.*}}SH3ADD - LBU
+---
+name: sh3add_lbu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LBU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1add_lhu
+# CHECK: Macro fuse: {{.*}}SH1ADD - LHU
+---
+name: sh1add_lhu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LHU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2add_lhu
+# CHECK: Macro fuse: {{.*}}SH2ADD - LHU
+---
+name: sh2add_lhu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LHU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3add_lhu
+# CHECK: Macro fuse: {{.*}}SH3ADD - LHU
+---
+name: sh3add_lhu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LHU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1add_lwu
+# CHECK: Macro fuse: {{.*}}SH1ADD - LWU
+---
+name: sh1add_lwu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LWU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2add_lwu
+# CHECK: Macro fuse: {{.*}}SH2ADD - LWU
+---
+name: sh2add_lwu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LWU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3add_lwu
+# CHECK: Macro fuse: {{.*}}SH3ADD - LWU
+---
+name: sh3add_lwu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LWU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
# CHECK: sh1adduw_lb
# CHECK: Macro fuse: {{.*}}SH1ADD_UW - LB
---
@@ -734,3 +986,165 @@ body: |
$x11 = COPY %5
PseudoRET
...
+
+# CHECK: sh1adduw_lbu
+# CHECK: Macro fuse: {{.*}}SH1ADD_UW - LBU
+---
+name: sh1adduw_lbu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LBU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2adduw_lbu
+# CHECK: Macro fuse: {{.*}}SH2ADD_UW - LBU
+---
+name: sh2adduw_lbu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LBU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3adduw_lbu
+# CHECK: Macro fuse: {{.*}}SH3ADD_UW - LBU
+---
+name: sh3adduw_lbu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LBU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1adduw_lhu
+# CHECK: Macro fuse: {{.*}}SH1ADD_UW - LHU
+---
+name: sh1adduw_lhu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LHU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2adduw_lhu
+# CHECK: Macro fuse: {{.*}}SH2ADD_UW - LHU
+---
+name: sh2adduw_lhu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LHU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3adduw_lhu
+# CHECK: Macro fuse: {{.*}}SH3ADD_UW - LHU
+---
+name: sh3adduw_lhu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LHU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh1adduw_lwu
+# CHECK: Macro fuse: {{.*}}SH1ADD_UW - LWU
+---
+name: sh1adduw_lwu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH1ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LWU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh2adduw_lwu
+# CHECK: Macro fuse: {{.*}}SH2ADD_UW - LWU
+---
+name: sh2adduw_lwu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH2ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LWU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
+
+# CHECK: sh3adduw_lwu
+# CHECK: Macro fuse: {{.*}}SH3ADD_UW - LWU
+---
+name: sh3adduw_lwu
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $x11
+ %1:gpr = COPY $x10
+ %2:gpr = COPY $x11
+ %3:gpr = SH3ADD_UW %1, %2
+ %4:gpr = XORI %2, 3
+ %5:gpr = LWU %3, 8
+ $x10 = COPY %4
+ $x11 = COPY %5
+ PseudoRET
+...
More information about the llvm-commits
mailing list