[AArch64][GlobalISel] Widen G_BSWAP before clamping

[paquette]

Summary

This allows us to avoid odd type breakdowns + allows us to legalize types like s88 in the first place.

Add some testcases for known legal types + testcases for s4 and s88.

[jroelofs]

LGTM

llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp

getActionDefinitionsBuilder(G_PHI)
       .clampMaxNumElements(0, s8, 16)
       .clampMaxNumElements(0, s16, 8)
       .clampMaxNumElements(0, s32, 4)
       .clampMaxNumElements(0, s64, 2)
       .clampMaxNumElements(0, p0, 2);
 
   getActionDefinitionsBuilder(G_BSWAP)
       .legalFor({s32, s64, v4s32, v2s32, v2s64})
-      .clampScalar(0, s32, s64)
       .widenScalarToNextPow2(0)
+      .clampScalar(0, s32, s64)
       .customIf(typeIs(0, v2s16)); // custom lower as G_REV32 + G_LSHR
 
   getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL, G_AND, G_OR, G_XOR})
       .legalFor({s32, s64, v2s32, v4s32, v4s16, v8s16, v16s8, v8s8})
       .scalarizeIf(
           [=](const LegalityQuery &Query) {
             return Query.Opcode == G_MUL && Query.Types[0] == v2s64;
           },

llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir

 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -march=aarch64 -run-pass=legalizer %s -o - -verify-machineinstrs | FileCheck %s
---- |
-  target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
-  target triple = "aarch64"
-
-  declare i16 @llvm.bswap.i16(i16) #0
-
-  define i16 @bswap_s16(i16 %a) { ret i16 0 }
-
-  define <2 x i16> @bswap_2xi16(<2 x i16> %a) { ret <2 x i16> <i16 0, i16 0> }
-
-  attributes #0 = { nounwind readnone speculatable willreturn }
 
 ...
 ---
 name:            bswap_s16
-alignment:       4
 tracksRegLiveness: true
-liveins:
-  - { reg: '$w0' }
-frameInfo:
-  maxAlignment:    1
-machineFunctionInfo: {}
 body:             |
   bb.1:
     liveins: $w0
 
     ; CHECK-LABEL: name: bswap_s16
     ; CHECK: liveins: $w0
     ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
     ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
     ; CHECK: [[BSWAP:%[0-9]+]]:_(s32) = G_BSWAP [[COPY1]]
     ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
     ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BSWAP]], [[C]](s64)
     ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
     ; CHECK: $w0 = COPY [[COPY2]](s32)
     ; CHECK: RET_ReallyLR implicit $w0
     %1:_(s32) = COPY $w0
     %0:_(s16) = G_TRUNC %1(s32)
     %2:_(s16) = G_BSWAP %0
     %3:_(s32) = G_ANYEXT %2(s16)
     $w0 = COPY %3(s32)
     RET_ReallyLR implicit $w0
-
 ...
 ---
 name:            bswap_2xi16
-alignment:       4
 tracksRegLiveness: true
-registers:
-  - { id: 0, class: _ }
-  - { id: 1, class: _ }
-liveins:
-  - { reg: '$s0' }
-frameInfo:
-  maxAlignment:    1
-machineFunctionInfo: {}
 body:             |
   bb.1:
     liveins: $s0
 
     ; CHECK-LABEL: name: bswap_2xi16
     ; CHECK: liveins: $s0
     ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $s0
     ; CHECK: [[DEF:%[0-9]+]]:_(<8 x s8>) = G_IMPLICIT_DEF
     ; CHECK: [[INSERT_SUBREG:%[0-9]+]]:_(<8 x s8>) = INSERT_SUBREG [[DEF]](<8 x s8>), [[COPY]](<2 x s16>), %subreg.ssub
     ; CHECK: [[REV32_:%[0-9]+]]:_(<8 x s8>) = G_REV32 [[INSERT_SUBREG]]
     ; CHECK: [[BITCAST:%[0-9]+]]:_(<2 x s32>) = G_BITCAST [[REV32_]](<8 x s8>)
     ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
     ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[C]](s32), [[C]](s32)
     ; CHECK: [[LSHR:%[0-9]+]]:_(<2 x s32>) = G_LSHR [[BITCAST]], [[BUILD_VECTOR]](<2 x s32>)
     ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
     ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[LSHR]](<2 x s32>), [[C1]](s64)
     ; CHECK: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[EVEC]](s32)
     ; CHECK: $s0 = COPY [[BITCAST1]](<2 x s16>)
     ; CHECK: RET_ReallyLR
     %0:_(<2 x s16>) = COPY $s0
     %1:_(<2 x s16>) = G_BSWAP %0
     $s0 = COPY %1(<2 x s16>)
     RET_ReallyLR
-
 ...
+---
+name:            bswap_s32_legal
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $w0
+    ; CHECK-LABEL: name: bswap_s32_legal
+    ; CHECK: liveins: $w0
+    ; CHECK: %copy:_(s32) = COPY $w0
+    ; CHECK: %bswap:_(s32) = G_BSWAP %copy
+    ; CHECK: $w0 = COPY %bswap(s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %copy:_(s32) = COPY $w0
+    %bswap:_(s32) = G_BSWAP %copy
+    $w0 = COPY %bswap(s32)
+    RET_ReallyLR implicit $w0
+...
+---
+name:            bswap_s64_legal
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $x0
+    ; CHECK-LABEL: name: bswap_s64_legal
+    ; CHECK: liveins: $x0
+    ; CHECK: %copy:_(s64) = COPY $x0
+    ; CHECK: %bswap:_(s64) = G_BSWAP %copy
+    ; CHECK: $x0 = COPY %bswap(s64)
+    ; CHECK: RET_ReallyLR implicit $x0
+    %copy:_(s64) = COPY $x0
+    %bswap:_(s64) = G_BSWAP %copy
+    $x0 = COPY %bswap(s64)
+    RET_ReallyLR implicit $x0
+...
+---
+name:            bswap_v4s32_legal
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $q0
+    ; CHECK-LABEL: name: bswap_v4s32_legal
+    ; CHECK: liveins: $q0
+    ; CHECK: %copy:_(<4 x s32>) = COPY $q0
+    ; CHECK: %bswap:_(<4 x s32>) = G_BSWAP %copy
+    ; CHECK: $q0 = COPY %bswap(<4 x s32>)
+    ; CHECK: RET_ReallyLR implicit $q0
+    %copy:_(<4 x s32>) = COPY $q0
+    %bswap:_(<4 x s32>) = G_BSWAP %copy
+    $q0 = COPY %bswap(<4 x s32>)
+    RET_ReallyLR implicit $q0
+...
+---
+name:            bswap_v2s32_legal
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $d0
+    ; CHECK-LABEL: name: bswap_v2s32_legal
+    ; CHECK: liveins: $d0
+    ; CHECK: %copy:_(<2 x s32>) = COPY $d0
+    ; CHECK: %bswap:_(<2 x s32>) = G_BSWAP %copy
+    ; CHECK: $d0 = COPY %bswap(<2 x s32>)
+    ; CHECK: RET_ReallyLR implicit $d0
+    %copy:_(<2 x s32>) = COPY $d0
+    %bswap:_(<2 x s32>) = G_BSWAP %copy
+    $d0 = COPY %bswap(<2 x s32>)
+    RET_ReallyLR implicit $d0
+...
+---
+name:            bswap_v2s64_legal
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $q0
+    ; CHECK-LABEL: name: bswap_v2s64_legal
+    ; CHECK: liveins: $q0
+    ; CHECK: %copy:_(<2 x s64>) = COPY $q0
+    ; CHECK: %bswap:_(<2 x s64>) = G_BSWAP %copy
+    ; CHECK: $q0 = COPY %bswap(<2 x s64>)
+    ; CHECK: RET_ReallyLR implicit $q0
+    %copy:_(<2 x s64>) = COPY $q0
+    %bswap:_(<2 x s64>) = G_BSWAP %copy
+    $q0 = COPY %bswap(<2 x s64>)
+    RET_ReallyLR implicit $q0
+...
+---
+name:            bswap_s88
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $x0
+    ; CHECK-LABEL: name: bswap_s88
+    ; CHECK: liveins: $x0
+    ; CHECK: [[DEF:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
+    ; CHECK: [[BSWAP:%[0-9]+]]:_(s64) = G_BSWAP [[DEF]]
+    ; CHECK: [[BSWAP1:%[0-9]+]]:_(s64) = G_BSWAP [[DEF]]
+    ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 40
+    ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
+    ; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 64
+    ; CHECK: [[SUB:%[0-9]+]]:_(s64) = G_SUB [[C]], [[C2]]
+    ; CHECK: [[SUB1:%[0-9]+]]:_(s64) = G_SUB [[C2]], [[C]]
+    ; CHECK: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ult), [[C]](s64), [[C2]]
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s1) = G_TRUNC [[ICMP]](s32)
+    ; CHECK: [[ICMP1:%[0-9]+]]:_(s32) = G_ICMP intpred(eq), [[C]](s64), [[C1]]
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s1) = G_TRUNC [[ICMP1]](s32)
+    ; CHECK: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[BSWAP1]], [[C]](s64)
+    ; CHECK: [[LSHR1:%[0-9]+]]:_(s64) = G_LSHR [[BSWAP]], [[C]](s64)
+    ; CHECK: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[BSWAP1]], [[SUB1]](s64)
+    ; CHECK: [[OR:%[0-9]+]]:_(s64) = G_OR [[LSHR1]], [[SHL]]
+    ; CHECK: [[LSHR2:%[0-9]+]]:_(s64) = G_LSHR [[BSWAP1]], [[SUB]](s64)
+    ; CHECK: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[TRUNC]](s1), [[OR]], [[LSHR2]]
+    ; CHECK: [[SELECT1:%[0-9]+]]:_(s64) = G_SELECT [[TRUNC1]](s1), [[BSWAP]], [[SELECT]]
+    ; CHECK: [[SELECT2:%[0-9]+]]:_(s64) = G_SELECT [[TRUNC]](s1), [[LSHR]], [[C1]]
+    ; CHECK: $x0 = COPY [[SELECT1]](s64)
+    ; CHECK: RET_ReallyLR implicit $x0
+    %val:_(s88) = G_IMPLICIT_DEF
+    %bswap:_(s88) = G_BSWAP %val
+    %trunc:_(s64) = G_TRUNC %bswap
+    $x0 = COPY %trunc(s64)
+    RET_ReallyLR implicit $x0
+...
+---
+name:            bswap_s4
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $x0
+    ; CHECK-LABEL: name: bswap_s4
+    ; CHECK: liveins: $x0
+    ; CHECK: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; CHECK: [[BSWAP:%[0-9]+]]:_(s32) = G_BSWAP [[DEF]]
+    ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 28
+    ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BSWAP]], [[C]](s64)
+    ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 15
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LSHR]](s32)
+    ; CHECK: %ext:_(s64) = G_AND [[ANYEXT]], [[C1]]
+    ; CHECK: $x0 = COPY %ext(s64)
+    ; CHECK: RET_ReallyLR implicit $x0
+    %val:_(s4) = G_IMPLICIT_DEF
+    %bswap:_(s4) = G_BSWAP %val
+    %ext:_(s64) = G_ZEXT %bswap
+    $x0 = COPY %ext(s64)
+    RET_ReallyLR implicit $x0