[AArch64][GlobalISel] Add support for ldexp (llvm#160517)

Author: HolyMolyCowMan

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

@@ -444,7 +444,7 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
  // Regardless of FP16 support, widen 16-bit elements to 32-bits.
  .minScalar(0, s32)
  .libcallFor({s32, s64, s128});
- getActionDefinitionsBuilder(G_FPOWI)
+ getActionDefinitionsBuilder({G_FPOWI, G_FLDEXP})
  .scalarize(0)
  .minScalar(0, s32)
  .libcallFor({{s32, s32}, {s64, s32}, {s128, s32}});

llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir

@@ -540,8 +540,9 @@
 # DEBUG-NEXT: .. the first uncovered type index: 1, OK
 # DEBUG-NEXT: .. the first uncovered imm index: 0, OK
 # DEBUG-NEXT: G_FLDEXP (opcode {{[0-9]+}}): 2 type indices, 0 imm indices
-# DEBUG-NEXT:.. type index coverage check SKIPPED: no rules defined
-# DEBUG-NEXT:.. imm index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT: .. opcode {{[0-9]+}} is aliased to {{[0-9]+}}
+# DEBUG-NEXT: .. the first uncovered type index: 2, OK
+# DEBUG-NEXT: .. the first uncovered imm index: 0, OK
 # DEBUG-NEXT: G_FFREXP (opcode {{[0-9]+}}): 2 type indices, 0 imm indices
 # DEBUG-NEXT:.. type index coverage check SKIPPED: no rules defined
 # DEBUG-NEXT:.. imm index coverage check SKIPPED: no rules defined

llvm/test/CodeGen/AArch64/ldexp.ll

@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
 ; RUN: llc -mtriple=aarch64 -mattr=+sve < %s -o - | FileCheck -check-prefixes=SVE,SVELINUX %s
+; RUN: llc -mtriple=aarch64 -global-isel < %s -o - | FileCheck -check-prefixes=GISEL %s
 ; RUN: llc -mtriple=aarch64-windows-msvc -mattr=+sve < %s -o - | FileCheck -check-prefixes=SVE,SVEWINDOWS %s
 ; RUN: llc -mtriple=aarch64-windows-msvc < %s -o - | FileCheck -check-prefixes=WINDOWS %s
 
@@ -15,6 +16,10 @@ define double @testExp(double %val, i32 %a) {
 ; SVE-NEXT: // kill: def $d0 killed $d0 killed $z0
 ; SVE-NEXT: ret
 ;
+; GISEL-LABEL: testExp:
+; GISEL: // %bb.0: // %entry
+; GISEL-NEXT: b ldexp
+;
 ; WINDOWS-LABEL: testExp:
 ; WINDOWS: // %bb.0: // %entry
 ; WINDOWS-NEXT: b ldexp
@@ -37,6 +42,10 @@ define double @testExpIntrinsic(double %val, i32 %a) {
 ; SVE-NEXT: // kill: def $d0 killed $d0 killed $z0
 ; SVE-NEXT: ret
 ;
+; GISEL-LABEL: testExpIntrinsic:
+; GISEL: // %bb.0: // %entry
+; GISEL-NEXT: b ldexp
+;
 ; WINDOWS-LABEL: testExpIntrinsic:
 ; WINDOWS: // %bb.0: // %entry
 ; WINDOWS-NEXT: b ldexp
@@ -55,6 +64,10 @@ define float @testExpf(float %val, i32 %a) {
 ; SVELINUX-NEXT: // kill: def $s0 killed $s0 killed $z0
 ; SVELINUX-NEXT: ret
 ;
+; GISEL-LABEL: testExpf:
+; GISEL: // %bb.0: // %entry
+; GISEL-NEXT: b ldexpf
+;
 ; SVEWINDOWS-LABEL: testExpf:
 ; SVEWINDOWS: // %bb.0: // %entry
 ; SVEWINDOWS-NEXT: b ldexpf
@@ -77,6 +90,10 @@ define float @testExpfIntrinsic(float %val, i32 %a) {
 ; SVE-NEXT: // kill: def $s0 killed $s0 killed $z0
 ; SVE-NEXT: ret
 ;
+; GISEL-LABEL: testExpfIntrinsic:
+; GISEL: // %bb.0: // %entry
+; GISEL-NEXT: b ldexpf
+;
 ; WINDOWS-LABEL: testExpfIntrinsic:
 ; WINDOWS: .seh_proc testExpfIntrinsic
 ; WINDOWS-NEXT: // %bb.0: // %entry
@@ -98,6 +115,90 @@ entry:
  ret float %call
 }
 
+define <2 x float> @test_ldexp_v2f32_v2i32(<2 x float> %Val, <2 x i32> %Exp) {
+; SVE-LABEL: test_ldexp_v2f32_v2i32:
+; SVE: // %bb.0:
+; SVE-NEXT: // kill: def $d1 killed $d1 def $z1
+; SVE-NEXT: mov w8, v1.s[1]
+; SVE-NEXT: // kill: def $d0 killed $d0 def $z0
+; SVE-NEXT: mov s2, v0.s[1]
+; SVE-NEXT: ptrue p0.s
+; SVE-NEXT: fscale z0.s, p0/m, z0.s, z1.s
+; SVE-NEXT: fmov s3, w8
+; SVE-NEXT: fscale z2.s, p0/m, z2.s, z3.s
+; SVE-NEXT: mov v0.s[1], v2.s[0]
+; SVE-NEXT: // kill: def $d0 killed $d0 killed $z0
+; SVE-NEXT: ret
+;
+; GISEL-LABEL: test_ldexp_v2f32_v2i32:
+; GISEL: // %bb.0:
+; GISEL-NEXT: sub sp, sp, #48
+; GISEL-NEXT: stp d9, d8, [sp, #16] // 16-byte Folded Spill
+; GISEL-NEXT: str x30, [sp, #32] // 8-byte Folded Spill
+; GISEL-NEXT: .cfi_def_cfa_offset 48
+; GISEL-NEXT: .cfi_offset w30, -16
+; GISEL-NEXT: .cfi_offset b8, -24
+; GISEL-NEXT: .cfi_offset b9, -32
+; GISEL-NEXT: // kill: def $d1 killed $d1 def $q1
+; GISEL-NEXT: fmov w0, s1
+; GISEL-NEXT: // kill: def $d0 killed $d0 def $q0
+; GISEL-NEXT: mov s8, v0.s[1]
+; GISEL-NEXT: mov s9, v1.s[1]
+; GISEL-NEXT: // kill: def $s0 killed $s0 killed $q0
+; GISEL-NEXT: bl ldexpf
+; GISEL-NEXT: // kill: def $s0 killed $s0 def $q0
+; GISEL-NEXT: str q0, [sp] // 16-byte Folded Spill
+; GISEL-NEXT: fmov w0, s9
+; GISEL-NEXT: fmov s0, s8
+; GISEL-NEXT: bl ldexpf
+; GISEL-NEXT: ldr q1, [sp] // 16-byte Folded Reload
+; GISEL-NEXT: // kill: def $s0 killed $s0 def $q0
+; GISEL-NEXT: ldp d9, d8, [sp, #16] // 16-byte Folded Reload
+; GISEL-NEXT: ldr x30, [sp, #32] // 8-byte Folded Reload
+; GISEL-NEXT: mov v1.s[1], v0.s[0]
+; GISEL-NEXT: fmov d0, d1
+; GISEL-NEXT: add sp, sp, #48
+; GISEL-NEXT: ret
+;
+; WINDOWS-LABEL: test_ldexp_v2f32_v2i32:
+; WINDOWS: .seh_proc test_ldexp_v2f32_v2i32
+; WINDOWS-NEXT: // %bb.0:
+; WINDOWS-NEXT: sub sp, sp, #48
+; WINDOWS-NEXT: .seh_stackalloc 48
+; WINDOWS-NEXT: str x30, [sp, #32] // 8-byte Folded Spill
+; WINDOWS-NEXT: .seh_save_reg x30, 32
+; WINDOWS-NEXT: .seh_endprologue
+; WINDOWS-NEXT: // kill: def $d0 killed $d0 def $q0
+; WINDOWS-NEXT: mov s2, v0.s[1]
+; WINDOWS-NEXT: // kill: def $d1 killed $d1 def $q1
+; WINDOWS-NEXT: stp q1, q0, [sp] // 32-byte Folded Spill
+; WINDOWS-NEXT: mov w0, v1.s[1]
+; WINDOWS-NEXT: fcvt d0, s2
+; WINDOWS-NEXT: bl ldexp
+; WINDOWS-NEXT: ldr q1, [sp, #16] // 16-byte Folded Reload
+; WINDOWS-NEXT: fcvt s0, d0
+; WINDOWS-NEXT: fcvt d1, s1
+; WINDOWS-NEXT: str q0, [sp, #16] // 16-byte Folded Spill
+; WINDOWS-NEXT: ldr q0, [sp] // 16-byte Folded Reload
+; WINDOWS-NEXT: fmov w0, s0
+; WINDOWS-NEXT: fmov d0, d1
+; WINDOWS-NEXT: bl ldexp
+; WINDOWS-NEXT: fcvt s0, d0
+; WINDOWS-NEXT: ldr q1, [sp, #16] // 16-byte Folded Reload
+; WINDOWS-NEXT: mov v0.s[1], v1.s[0]
+; WINDOWS-NEXT: // kill: def $d0 killed $d0 killed $q0
+; WINDOWS-NEXT: .seh_startepilogue
+; WINDOWS-NEXT: ldr x30, [sp, #32] // 8-byte Folded Reload
+; WINDOWS-NEXT: .seh_save_reg x30, 32
+; WINDOWS-NEXT: add sp, sp, #48
+; WINDOWS-NEXT: .seh_stackalloc 48
+; WINDOWS-NEXT: .seh_endepilogue
+; WINDOWS-NEXT: ret
+; WINDOWS-NEXT: .seh_endfunclet
+; WINDOWS-NEXT: .seh_endproc
+ %result = call <2 x float> @llvm.ldexp.v2f32.v2i32(<2 x float> %Val, <2 x i32> %Exp)
+ ret <2 x float> %result
+}
 
 declare float @ldexpf(float, i32) memory(none)
 
@@ -106,6 +207,10 @@ define fp128 @testExpl(fp128 %val, i32 %a) {
 ; SVE: // %bb.0: // %entry
 ; SVE-NEXT: b ldexpl
 ;
+; GISEL-LABEL: testExpl:
+; GISEL: // %bb.0: // %entry
+; GISEL-NEXT: b ldexpl
+;
 ; WINDOWS-LABEL: testExpl:
 ; WINDOWS: // %bb.0: // %entry
 ; WINDOWS-NEXT: b ldexpl
@@ -126,6 +231,17 @@ define half @testExpf16(half %val, i32 %a) {
 ; SVE-NEXT: fcvt h0, s0
 ; SVE-NEXT: ret
 ;
+; GISEL-LABEL: testExpf16:
+; GISEL: // %bb.0: // %entry
+; GISEL-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
+; GISEL-NEXT: .cfi_def_cfa_offset 16
+; GISEL-NEXT: .cfi_offset w30, -16
+; GISEL-NEXT: fcvt s0, h0
+; GISEL-NEXT: bl ldexpf
+; GISEL-NEXT: fcvt h0, s0
+; GISEL-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
+; GISEL-NEXT: ret
+;
 ; WINDOWS-LABEL: testExpf16:
 ; WINDOWS: .seh_proc testExpf16
 ; WINDOWS-NEXT: // %bb.0: // %entry