[RISCV][GISel] Legalize G_SMULO/G_UMULO #67635

topperc · 2023-09-28T06:35:41Z

I'm unhappy with the XLen case when we need to make a libcall. We get two libcalls.

llvmbot · 2023-09-28T06:36:53Z

@llvm/pr-subscribers-backend-risc-v

@llvm/pr-subscribers-llvm-globalisel

Changes

I'm unhappy with the XLen case when we need to make a libcall. We get two libcalls.

Patch is 37.88 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/67635.diff

3 Files Affected:

(modified) llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp (+8)
(added) llvm/test/CodeGen/RISCV/GlobalISel/legalizer/rv32/legalize-mulo.mir (+340)
(added) llvm/test/CodeGen/RISCV/GlobalISel/legalizer/rv64/legalize-mulo.mir (+442)

diff --git a/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp b/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
index 313a8221ab1bd4a..a9d6f61541253f7 100644
--- a/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
+++ b/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
@@ -145,6 +145,10 @@ RISCVLegalizerInfo::RISCVLegalizerInfo(const RISCVSubtarget &ST) {
         .legalFor({XLenLLT})
         .lower();
     // clang-format on
+
+    getActionDefinitionsBuilder({G_SMULO, G_UMULO})
+        .minScalar(0, XLenLLT)
+        .lower();
   } else {
     getActionDefinitionsBuilder(G_MUL)
         .libcallFor({XLenLLT, DoubleXLenLLT})
@@ -152,6 +156,10 @@ RISCVLegalizerInfo::RISCVLegalizerInfo(const RISCVSubtarget &ST) {
         .clampScalar(0, XLenLLT, DoubleXLenLLT);
 
     getActionDefinitionsBuilder({G_SMULH, G_UMULH}).lowerFor({XLenLLT});
+
+    getActionDefinitionsBuilder({G_SMULO, G_UMULO})
+        .minScalar(0, XLenLLT)
+        .lower();
   }
 
   if (ST.hasStdExtM()) {
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/rv32/legalize-mulo.mir b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/rv32/legalize-mulo.mir
new file mode 100644
index 000000000000000..35bea08f426a3a3
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/rv32/legalize-mulo.mir
@@ -0,0 +1,340 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mattr=+m -mtriple=riscv32 -run-pass=legalizer %s -o - \
+# RUN: | FileCheck %s
+# RUN: llc -mattr=+zmmul -mtriple=riscv32 -run-pass=legalizer %s -o - \
+# RUN: | FileCheck %s
+# RUN: llc -mtriple=riscv32 -run-pass=legalizer %s -o - \
+# RUN: | FileCheck %s --check-prefix=LIBCALL
+
+---
+name:            smulo_i8
+body:             |
+  bb.1:
+    liveins: $x10, $x11
+
+    ; CHECK-LABEL: name: smulo_i8
+    ; CHECK: liveins: $x10, $x11
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CHECK-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY]], [[C]](s32)
+    ; CHECK-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32)
+    ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CHECK-NEXT: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY1]], [[C1]](s32)
+    ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C1]](s32)
+    ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[ASHR]], [[ASHR1]]
+    ; CHECK-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CHECK-NEXT: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[MUL]], [[C2]](s32)
+    ; CHECK-NEXT: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C2]](s32)
+    ; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[MUL]](s32), [[ASHR2]]
+    ; CHECK-NEXT: $x10 = COPY [[MUL]](s32)
+    ; CHECK-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; CHECK-NEXT: PseudoRET implicit $x10, implicit $x11
+    ;
+    ; LIBCALL-LABEL: name: smulo_i8
+    ; LIBCALL: liveins: $x10, $x11
+    ; LIBCALL-NEXT: {{  $}}
+    ; LIBCALL-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; LIBCALL-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; LIBCALL-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY]], [[C]](s32)
+    ; LIBCALL-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32)
+    ; LIBCALL-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; LIBCALL-NEXT: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY1]], [[C1]](s32)
+    ; LIBCALL-NEXT: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C1]](s32)
+    ; LIBCALL-NEXT: $x10 = COPY [[ASHR]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[ASHR1]](s32)
+    ; LIBCALL-NEXT: PseudoCALL target-flags(riscv-call) &__mulsi3, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; LIBCALL-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; LIBCALL-NEXT: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[COPY2]], [[C2]](s32)
+    ; LIBCALL-NEXT: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C2]](s32)
+    ; LIBCALL-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[COPY2]](s32), [[ASHR2]]
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY2]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; LIBCALL-NEXT: PseudoRET implicit $x10, implicit $x11
+    %2:_(s32) = COPY $x10
+    %0:_(s8) = G_TRUNC %2(s32)
+    %3:_(s32) = COPY $x11
+    %1:_(s8) = G_TRUNC %3(s32)
+    %4:_(s8), %5:_(s1) = G_SMULO %0, %1
+    %6:_(s32) = G_ANYEXT %4(s8)
+    %7:_(s32) = G_ANYEXT %5(s1)
+    $x10 = COPY %6(s32)
+    $x11 = COPY %7(s32)
+    PseudoRET implicit $x10, implicit $x11
+
+...
+---
+name:            smulo_i16
+body:             |
+  bb.1:
+    liveins: $x10, $x11
+
+    ; CHECK-LABEL: name: smulo_i16
+    ; CHECK: liveins: $x10, $x11
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY]], [[C]](s32)
+    ; CHECK-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32)
+    ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK-NEXT: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY1]], [[C1]](s32)
+    ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C1]](s32)
+    ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[ASHR]], [[ASHR1]]
+    ; CHECK-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK-NEXT: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[MUL]], [[C2]](s32)
+    ; CHECK-NEXT: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C2]](s32)
+    ; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[MUL]](s32), [[ASHR2]]
+    ; CHECK-NEXT: $x10 = COPY [[MUL]](s32)
+    ; CHECK-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; CHECK-NEXT: PseudoRET implicit $x10, implicit $x11
+    ;
+    ; LIBCALL-LABEL: name: smulo_i16
+    ; LIBCALL: liveins: $x10, $x11
+    ; LIBCALL-NEXT: {{  $}}
+    ; LIBCALL-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; LIBCALL-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; LIBCALL-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY]], [[C]](s32)
+    ; LIBCALL-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32)
+    ; LIBCALL-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; LIBCALL-NEXT: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY1]], [[C1]](s32)
+    ; LIBCALL-NEXT: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C1]](s32)
+    ; LIBCALL-NEXT: $x10 = COPY [[ASHR]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[ASHR1]](s32)
+    ; LIBCALL-NEXT: PseudoCALL target-flags(riscv-call) &__mulsi3, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; LIBCALL-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; LIBCALL-NEXT: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[COPY2]], [[C2]](s32)
+    ; LIBCALL-NEXT: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C2]](s32)
+    ; LIBCALL-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[COPY2]](s32), [[ASHR2]]
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY2]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; LIBCALL-NEXT: PseudoRET implicit $x10, implicit $x11
+    %2:_(s32) = COPY $x10
+    %0:_(s16) = G_TRUNC %2(s32)
+    %3:_(s32) = COPY $x11
+    %1:_(s16) = G_TRUNC %3(s32)
+    %4:_(s16), %5:_(s1) = G_SMULO %0, %1
+    %6:_(s32) = G_ANYEXT %4(s16)
+    %7:_(s32) = G_ANYEXT %5(s1)
+    $x10 = COPY %6(s32)
+    $x11 = COPY %7(s32)
+    PseudoRET implicit $x10, implicit $x11
+
+...
+---
+name:            smulo_i32
+body:             |
+  bb.1:
+    liveins: $x10, $x11
+
+    ; CHECK-LABEL: name: smulo_i32
+    ; CHECK: liveins: $x10, $x11
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-NEXT: [[SMULH:%[0-9]+]]:_(s32) = G_SMULH [[COPY]], [[COPY1]]
+    ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY]], [[COPY1]]
+    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
+    ; CHECK-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[MUL]], [[C]](s32)
+    ; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[SMULH]](s32), [[ASHR]]
+    ; CHECK-NEXT: $x10 = COPY [[MUL]](s32)
+    ; CHECK-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; CHECK-NEXT: PseudoRET implicit $x10, implicit $x11
+    ;
+    ; LIBCALL-LABEL: name: smulo_i32
+    ; LIBCALL: liveins: $x10, $x11
+    ; LIBCALL-NEXT: {{  $}}
+    ; LIBCALL-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; LIBCALL-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
+    ; LIBCALL-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[COPY]], [[C]](s32)
+    ; LIBCALL-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
+    ; LIBCALL-NEXT: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[COPY1]], [[C1]](s32)
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[ASHR]](s32)
+    ; LIBCALL-NEXT: $x12 = COPY [[COPY1]](s32)
+    ; LIBCALL-NEXT: $x13 = COPY [[ASHR1]](s32)
+    ; LIBCALL-NEXT: PseudoCALL target-flags(riscv-call) &__muldi3, implicit-def $x1, implicit $x10, implicit $x11, implicit $x12, implicit $x13, implicit-def $x10, implicit-def $x11
+    ; LIBCALL-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $x11
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[COPY1]](s32)
+    ; LIBCALL-NEXT: PseudoCALL target-flags(riscv-call) &__mulsi3, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; LIBCALL-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
+    ; LIBCALL-NEXT: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[COPY3]], [[C2]](s32)
+    ; LIBCALL-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[COPY2]](s32), [[ASHR2]]
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY3]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; LIBCALL-NEXT: PseudoRET implicit $x10, implicit $x11
+    %0:_(s32) = COPY $x10
+    %1:_(s32) = COPY $x11
+    %2:_(s32), %3:_(s1) = G_SMULO %0, %1
+    %4:_(s32) = G_ANYEXT %3(s1)
+    $x10 = COPY %2(s32)
+    $x11 = COPY %4(s32)
+    PseudoRET implicit $x10, implicit $x11
+
+...
+---
+name:            umulo_i8
+body:             |
+  bb.1:
+    liveins: $x10, $x11
+
+    ; CHECK-LABEL: name: umulo_i8
+    ; CHECK: liveins: $x10, $x11
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CHECK-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C]]
+    ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CHECK-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
+    ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; CHECK-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CHECK-NEXT: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C2]]
+    ; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; CHECK-NEXT: $x10 = COPY [[MUL]](s32)
+    ; CHECK-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; CHECK-NEXT: PseudoRET implicit $x10, implicit $x11
+    ;
+    ; LIBCALL-LABEL: name: umulo_i8
+    ; LIBCALL: liveins: $x10, $x11
+    ; LIBCALL-NEXT: {{  $}}
+    ; LIBCALL-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; LIBCALL-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; LIBCALL-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C]]
+    ; LIBCALL-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; LIBCALL-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
+    ; LIBCALL-NEXT: $x10 = COPY [[AND]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[AND1]](s32)
+    ; LIBCALL-NEXT: PseudoCALL target-flags(riscv-call) &__mulsi3, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; LIBCALL-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; LIBCALL-NEXT: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C2]]
+    ; LIBCALL-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[COPY2]](s32), [[AND2]]
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY2]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; LIBCALL-NEXT: PseudoRET implicit $x10, implicit $x11
+    %2:_(s32) = COPY $x10
+    %0:_(s8) = G_TRUNC %2(s32)
+    %3:_(s32) = COPY $x11
+    %1:_(s8) = G_TRUNC %3(s32)
+    %4:_(s8), %5:_(s1) = G_UMULO %0, %1
+    %6:_(s32) = G_ANYEXT %4(s8)
+    %7:_(s32) = G_ANYEXT %5(s1)
+    $x10 = COPY %6(s32)
+    $x11 = COPY %7(s32)
+    PseudoRET implicit $x10, implicit $x11
+
+...
+---
+name:            umulo_i16
+body:             |
+  bb.1:
+    liveins: $x10, $x11
+
+    ; CHECK-LABEL: name: umulo_i16
+    ; CHECK: liveins: $x10, $x11
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; CHECK-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C]]
+    ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; CHECK-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
+    ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; CHECK-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; CHECK-NEXT: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C2]]
+    ; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; CHECK-NEXT: $x10 = COPY [[MUL]](s32)
+    ; CHECK-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; CHECK-NEXT: PseudoRET implicit $x10, implicit $x11
+    ;
+    ; LIBCALL-LABEL: name: umulo_i16
+    ; LIBCALL: liveins: $x10, $x11
+    ; LIBCALL-NEXT: {{  $}}
+    ; LIBCALL-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; LIBCALL-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; LIBCALL-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C]]
+    ; LIBCALL-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; LIBCALL-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
+    ; LIBCALL-NEXT: $x10 = COPY [[AND]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[AND1]](s32)
+    ; LIBCALL-NEXT: PseudoCALL target-flags(riscv-call) &__mulsi3, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; LIBCALL-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; LIBCALL-NEXT: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C2]]
+    ; LIBCALL-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[COPY2]](s32), [[AND2]]
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY2]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; LIBCALL-NEXT: PseudoRET implicit $x10, implicit $x11
+    %2:_(s32) = COPY $x10
+    %0:_(s16) = G_TRUNC %2(s32)
+    %3:_(s32) = COPY $x11
+    %1:_(s16) = G_TRUNC %3(s32)
+    %4:_(s16), %5:_(s1) = G_UMULO %0, %1
+    %6:_(s32) = G_ANYEXT %4(s16)
+    %7:_(s32) = G_ANYEXT %5(s1)
+    $x10 = COPY %6(s32)
+    $x11 = COPY %7(s32)
+    PseudoRET implicit $x10, implicit $x11
+
+...
+---
+name:            umulo_i32
+body:             |
+  bb.1:
+    liveins: $x10, $x11
+
+    ; CHECK-LABEL: name: umulo_i32
+    ; CHECK: liveins: $x10, $x11
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; CHECK-NEXT: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[COPY]], [[COPY1]]
+    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY]], [[COPY1]]
+    ; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[UMULH]](s32), [[C]]
+    ; CHECK-NEXT: $x10 = COPY [[MUL]](s32)
+    ; CHECK-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; CHECK-NEXT: PseudoRET implicit $x10, implicit $x11
+    ;
+    ; LIBCALL-LABEL: name: umulo_i32
+    ; LIBCALL: liveins: $x10, $x11
+    ; LIBCALL-NEXT: {{  $}}
+    ; LIBCALL-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $x11
+    ; LIBCALL-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; LIBCALL-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[C]](s32)
+    ; LIBCALL-NEXT: $x12 = COPY [[COPY1]](s32)
+    ; LIBCALL-NEXT: $x13 = COPY [[C1]](s32)
+    ; LIBCALL-NEXT: PseudoCALL target-flags(riscv-call) &__muldi3, implicit-def $x1, implicit $x10, implicit $x11, implicit $x12, implicit $x13, implicit-def $x10, implicit-def $x11
+    ; LIBCALL-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $x11
+    ; LIBCALL-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[COPY1]](s32)
+    ; LIBCALL-NEXT: PseudoCALL target-flags(riscv-call) &__mulsi3, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; LIBCALL-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $x10
+    ; LIBCALL-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[COPY2]](s32), [[C2]]
+    ; LIBCALL-NEXT: $x10 = COPY [[COPY3]](s32)
+    ; LIBCALL-NEXT: $x11 = COPY [[ICMP]](s32)
+    ; LIBCALL-NEXT: PseudoRET implicit $x10, implicit $x11
+    %0:_(s32) = COPY $x10
+    %1:_(s32) = COPY $x11
+    %2:_(s32), %3:_(s1) = G_UMULO %0, %1
+    %4:_(s32) = G_ANYEXT %3(s1)
+    $x10 = COPY %2(s32)
+    $x11 = COPY %4(s32)
+    PseudoRET implicit $x10, implicit $x11
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/rv64/legalize-mulo.mir b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/rv64/legalize-mulo.mir
new file mode 100644
index 000000000000000..219635914efdd1d
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/rv64/legalize-mulo.mir
@@ -0,0 +1,442 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mattr=+m -mtriple=riscv64 -run-pass=legalizer %s -o - \
+# RUN: | FileCheck %s
+# RUN: llc -mattr=+zmmul -mtriple=riscv64 -run-pass=legalizer %s -o - \
+# RUN: | FileCheck %s
+# RUN: llc -mtriple=riscv64 -run-pass=legalizer %s -o - \
+# RUN: | FileCheck %s --check-prefix=LIBCALL
+
+---
+name:            smulo_i8
+body:             |
+  bb.1:
+    liveins: $x10, $x11
+
+    ; CHECK-LABEL: name: smulo_i8
+    ; CHECK: liveins: $x10, $x11
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x11
+    ; CHECK-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 56
+    ; CHECK-NEXT: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[COPY]], [[C]](s64)
+    ; CHECK-NEXT: [[ASHR:%[0-9]+]]:_(s64) = G_ASHR [[SHL]], [[C]](s64)
+    ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 56
+    ; CHECK-NEXT: [[SHL1:%[0-9]+]]:_(s64) = G_SHL [[COPY1]], [[C1]](s64)
+    ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:_(s64) = G_ASHR [[SHL1]], [[C1]](s64)
+    ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(s64) = G_MUL [[ASHR]], [[ASHR1]]
+    ; CHECK-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 56
+    ; CHECK-NEXT: [[SHL2:%[0-9]+]]:_(s64) = G_SHL [[MUL]], [[C2]](s64)
+    ; CHECK-NEXT: [[ASHR2:%[0-9]+]]:_(s64) = G_ASHR [[SHL2]], [[C2]](s64)
+    ; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s64) = G_ICMP intpred(ne), [[MUL]](s64), [[ASHR2]]
+    ; CHECK-NEXT: $x10 = COPY [[MUL]](s64)
+    ; CHECK-NEXT: $x11 = COPY [[ICMP]](s64)
+    ; CHECK-NEXT: PseudoRET implicit $x10, implicit $x11
+    ;
+    ; LIBCALL-LABEL: name: smulo_i8
+    ; LIBCALL: liveins: $x10, $x11
+    ; LIBCALL-NEXT: {{  $}}
+    ; LIBCALL-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; LIBCALL-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x11
+    ; LIBCALL-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 56
+    ; LIBCALL-NEXT: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[COPY]], [[C]](s64)
+    ; LIBCALL-NEXT: [[ASHR:%[0-9]+]]:_(s64) = G_ASHR [[SHL]], [[C]](s64)
+    ; LIBCALL-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 56
+    ; LIBCALL-NEXT: [[SHL1:%[0-9]+]]:_(s64) = G_SHL [[COPY1]], [[C1]](s64)
+    ; LIBCALL-NEXT: [[ASHR1:%[0-9]+]]:_(s64) = G_ASHR [[SHL1]], [[C1]](s64)
+    ; LIBCALL-NEXT: $x10 = COPY [[ASHR]](s64)
+    ; LIBCALL-NEXT: $x11 = COPY [[ASHR1]](s64)
+    ; LIBCALL-NEXT: PseudoCALL target-flags(riscv-call) &__muldi3, implicit-def $x1, implicit $x10, implicit $x11, implicit-def $x10
+    ; LIBCALL-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY $x10
+    ; LIBCALL-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 56
+    ; LIBCALL-NEXT: [[SHL2:%[0-9]+...
[truncated]

tschuett · 2023-09-28T15:37:49Z

llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp

@@ -145,13 +145,21 @@ RISCVLegalizerInfo::RISCVLegalizerInfo(const RISCVSubtarget &ST) {
        .legalFor({XLenLLT})
        .lower();
    // clang-format on
+
+    getActionDefinitionsBuilder({G_SMULO, G_UMULO})


I am maybe blind, but could you hoist them above the if? They look similar.

Yeah they're the same. They were briefly different while I tried and failed to fix the 2 libcall issue.

I have a feeling that you are going for coverage and not performance. Could you instead do a custom legalization, i.e. RISC-V specific and not generic lower?

The widening of small types creates an XLenLLT sized G_SMULO even though that can't overflow if the small type was half of XLen or less. I don't know how to distinquish that G_SMULO from the original. So I ended up with a double XLen libcall no matter what which is not really what I wanted.

I guess I could write custom legalization for everything, but didn't seem ideal.

github-actions · 2023-09-29T18:43:48Z

✅ With the latest revision this PR passed the C/C++ code formatter.

topperc · 2023-10-13T21:44:42Z

Ping

aemerson

LGTM.

aemerson · 2023-10-13T23:59:26Z

llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp

+        // the low bits for the mul result and high bits to do the overflow
+        // check.
+        .widenScalarIf(
+            [=, &ST](const LegalityQuery &Query) {


No need for ST?

Thanks. Copy and pasted from elsewhere. Will remove.

[RISCV][GISel] Legalize G_SMULO/G_UMULO

843549e

I'm unhappy with the XLen case when we need to make a libcall. We get two libcalls.

topperc requested review from aemerson, arsenm, mshockwave, michaelmaitland and nitinjohnraj September 28, 2023 06:35

llvmbot added backend:RISC-V llvm:globalisel labels Sep 28, 2023

tschuett reviewed Sep 28, 2023

View reviewed changes

!fixup reduce code duplication

56b7ccb

topperc added 2 commits September 29, 2023 12:49

!fixup clang-format

b2137ec

Fixes to get a single libcall for XLenLLT.

9132427

aemerson approved these changes Oct 14, 2023

View reviewed changes

!fixup remove unused lambda capture

2ca370d

topperc merged commit 3750558 into llvm:main Oct 14, 2023
2 of 3 checks passed

topperc deleted the pr/gisel-mulo branch October 14, 2023 03:34

madhur13490 mentioned this pull request Oct 20, 2023

Revert commit ba8565fbcb975e2d067ce3ae5a7dbaae4953edd3 madhur13490/llvm-project#3

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[RISCV][GISel] Legalize G_SMULO/G_UMULO #67635

[RISCV][GISel] Legalize G_SMULO/G_UMULO #67635

topperc commented Sep 28, 2023

llvmbot commented Sep 28, 2023 •

edited

Loading

tschuett Sep 28, 2023 •

edited

Loading

topperc Sep 28, 2023

tschuett Sep 28, 2023

topperc Sep 28, 2023

github-actions bot commented Sep 29, 2023 •

edited

Loading

topperc commented Oct 13, 2023

aemerson left a comment

aemerson Oct 13, 2023

topperc Oct 14, 2023

[RISCV][GISel] Legalize G_SMULO/G_UMULO #67635

[RISCV][GISel] Legalize G_SMULO/G_UMULO #67635

Conversation

topperc commented Sep 28, 2023

llvmbot commented Sep 28, 2023 • edited Loading

tschuett Sep 28, 2023 • edited Loading

Choose a reason for hiding this comment

topperc Sep 28, 2023

Choose a reason for hiding this comment

tschuett Sep 28, 2023

Choose a reason for hiding this comment

topperc Sep 28, 2023

Choose a reason for hiding this comment

github-actions bot commented Sep 29, 2023 • edited Loading

topperc commented Oct 13, 2023

aemerson left a comment

Choose a reason for hiding this comment

aemerson Oct 13, 2023

Choose a reason for hiding this comment

topperc Oct 14, 2023

Choose a reason for hiding this comment

llvmbot commented Sep 28, 2023 •

edited

Loading

tschuett Sep 28, 2023 •

edited

Loading

github-actions bot commented Sep 29, 2023 •

edited

Loading