[llvm] 5be3a1a - [SPARC] Legalize truncation and extension between fp128 and half
via llvm-commits
llvm-commits at lists.llvm.org
Sun Jun 13 11:05:21 PDT 2021
Author: LemonBoy
Date: 2021-06-13T20:05:15+02:00
New Revision: 5be3a1a06483a5958a1a5dbffda386fd4ce579fe
URL: https://github.com/llvm/llvm-project/commit/5be3a1a06483a5958a1a5dbffda386fd4ce579fe
DIFF: https://github.com/llvm/llvm-project/commit/5be3a1a06483a5958a1a5dbffda386fd4ce579fe.diff
LOG: [SPARC] Legalize truncation and extension between fp128 and half
Lower truncations and expansions between fp128 and half values into libcalls.
Expand truncating stores into two separate truncation and a store operations.
Reviewed By: jrtc27
Differential Revision: https://reviews.llvm.org/D104185
Added:
Modified:
llvm/lib/Target/Sparc/SparcISelLowering.cpp
llvm/test/CodeGen/SPARC/fp16-promote.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/Sparc/SparcISelLowering.cpp b/llvm/lib/Target/Sparc/SparcISelLowering.cpp
index e5c7794b7d2fd..cb43f0fad6f1b 100644
--- a/llvm/lib/Target/Sparc/SparcISelLowering.cpp
+++ b/llvm/lib/Target/Sparc/SparcISelLowering.cpp
@@ -1480,6 +1480,7 @@ SparcTargetLowering::SparcTargetLowering(const TargetMachine &TM,
setTruncStoreAction(MVT::f32, MVT::f16, Expand);
setTruncStoreAction(MVT::f64, MVT::f16, Expand);
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+ setTruncStoreAction(MVT::f128, MVT::f16, Expand);
setTruncStoreAction(MVT::f128, MVT::f32, Expand);
setTruncStoreAction(MVT::f128, MVT::f64, Expand);
@@ -1525,6 +1526,8 @@ SparcTargetLowering::SparcTargetLowering(const TargetMachine &TM,
setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand);
setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
+ setOperationAction(ISD::FP16_TO_FP, MVT::f128, Expand);
+ setOperationAction(ISD::FP_TO_FP16, MVT::f128, Expand);
setOperationAction(ISD::BITCAST, MVT::f32, Expand);
setOperationAction(ISD::BITCAST, MVT::i32, Expand);
diff --git a/llvm/test/CodeGen/SPARC/fp16-promote.ll b/llvm/test/CodeGen/SPARC/fp16-promote.ll
index 6d8c25f9f0d42..6afd8f1a2896b 100644
--- a/llvm/test/CodeGen/SPARC/fp16-promote.ll
+++ b/llvm/test/CodeGen/SPARC/fp16-promote.ll
@@ -77,6 +77,90 @@ define double @test_fpextend_double(half* %p) nounwind {
ret double %r
}
+define void @test_fpextend_fp128(half* %p, fp128* %out) nounwind {
+; V8-OPT-LABEL: test_fpextend_fp128:
+; V8-OPT: ! %bb.0:
+; V8-OPT-NEXT: save %sp, -112, %sp
+; V8-OPT-NEXT: call __gnu_h2f_ieee
+; V8-OPT-NEXT: lduh [%i0], %o0
+; V8-OPT-NEXT: st %f0, [%fp+-20]
+; V8-OPT-NEXT: add %fp, -16, %i0
+; V8-OPT-NEXT: st %i0, [%sp+64]
+; V8-OPT-NEXT: call _Q_stoq
+; V8-OPT-NEXT: ld [%fp+-20], %o0
+; V8-OPT-NEXT: unimp 16
+; V8-OPT-NEXT: ldd [%fp+-16], %f0
+; V8-OPT-NEXT: ldd [%fp+-8], %f2
+; V8-OPT-NEXT: std %f2, [%i1+8]
+; V8-OPT-NEXT: std %f0, [%i1]
+; V8-OPT-NEXT: ret
+; V8-OPT-NEXT: restore
+;
+; V8-UNOPT-LABEL: test_fpextend_fp128:
+; V8-UNOPT: ! %bb.0:
+; V8-UNOPT-NEXT: save %sp, -112, %sp
+; V8-UNOPT-NEXT: call __gnu_h2f_ieee
+; V8-UNOPT-NEXT: lduh [%i0], %o0
+; V8-UNOPT-NEXT: st %f0, [%fp+-20]
+; V8-UNOPT-NEXT: add %fp, -16, %i0
+; V8-UNOPT-NEXT: st %i0, [%sp+64]
+; V8-UNOPT-NEXT: call _Q_stoq
+; V8-UNOPT-NEXT: ld [%fp+-20], %o0
+; V8-UNOPT-NEXT: unimp 16
+; V8-UNOPT-NEXT: ldd [%fp+-16], %f4
+; V8-UNOPT-NEXT: ! implicit-def: $q0
+; V8-UNOPT-NEXT: fmovs %f4, %f0
+; V8-UNOPT-NEXT: fmovs %f5, %f1
+; V8-UNOPT-NEXT: ldd [%fp+-8], %f4
+; V8-UNOPT-NEXT: fmovs %f4, %f2
+; V8-UNOPT-NEXT: fmovs %f5, %f3
+; V8-UNOPT-NEXT: fmovs %f2, %f4
+; V8-UNOPT-NEXT: fmovs %f3, %f5
+; V8-UNOPT-NEXT: std %f4, [%i1+8]
+; V8-UNOPT-NEXT: ! kill: def $d0 killed $d0 killed $q0
+; V8-UNOPT-NEXT: std %f0, [%i1]
+; V8-UNOPT-NEXT: ret
+; V8-UNOPT-NEXT: restore
+;
+; V9-LABEL: test_fpextend_fp128:
+; V9: ! %bb.0:
+; V9-NEXT: save %sp, -112, %sp
+; V9-NEXT: call __gnu_h2f_ieee
+; V9-NEXT: lduh [%i0], %o0
+; V9-NEXT: st %f0, [%fp+-20]
+; V9-NEXT: add %fp, -16, %i0
+; V9-NEXT: st %i0, [%sp+64]
+; V9-NEXT: call _Q_stoq
+; V9-NEXT: ld [%fp+-20], %o0
+; V9-NEXT: unimp 16
+; V9-NEXT: ldd [%fp+-16], %f0
+; V9-NEXT: ldd [%fp+-8], %f2
+; V9-NEXT: std %f2, [%i1+8]
+; V9-NEXT: std %f0, [%i1]
+; V9-NEXT: ret
+; V9-NEXT: restore
+;
+; SPARC64-LABEL: test_fpextend_fp128:
+; SPARC64: ! %bb.0:
+; SPARC64-NEXT: save %sp, -192, %sp
+; SPARC64-NEXT: call __gnu_h2f_ieee
+; SPARC64-NEXT: lduh [%i0], %o0
+; SPARC64-NEXT: add %fp, 2031, %o0
+; SPARC64-NEXT: fmovs %f0, %f3
+; SPARC64-NEXT: call _Qp_stoq
+; SPARC64-NEXT: nop
+; SPARC64-NEXT: ldd [%fp+2031], %f0
+; SPARC64-NEXT: ldd [%fp+2039], %f2
+; SPARC64-NEXT: std %f2, [%i1+8]
+; SPARC64-NEXT: std %f0, [%i1]
+; SPARC64-NEXT: ret
+; SPARC64-NEXT: restore
+ %a = load half, half* %p
+ %r = fpext half %a to fp128
+ store fp128 %r, fp128* %out
+ ret void
+}
+
define void @test_fptrunc_float(float %f, half* %p) nounwind {
; V8-OPT-LABEL: test_fptrunc_float:
; V8-OPT: ! %bb.0:
@@ -181,6 +265,66 @@ define void @test_fptrunc_double(double %d, half* %p) nounwind {
ret void
}
+define void @test_fptrunc_fp128(fp128* %dp, half* %p) nounwind {
+; V8-OPT-LABEL: test_fptrunc_fp128:
+; V8-OPT: ! %bb.0:
+; V8-OPT-NEXT: save %sp, -104, %sp
+; V8-OPT-NEXT: ldd [%i0], %f0
+; V8-OPT-NEXT: ldd [%i0+8], %f2
+; V8-OPT-NEXT: std %f2, [%sp+100]
+; V8-OPT-NEXT: call __trunctfhf2
+; V8-OPT-NEXT: std %f0, [%sp+92]
+; V8-OPT-NEXT: sth %o0, [%i1]
+; V8-OPT-NEXT: ret
+; V8-OPT-NEXT: restore
+;
+; V8-UNOPT-LABEL: test_fptrunc_fp128:
+; V8-UNOPT: ! %bb.0:
+; V8-UNOPT-NEXT: save %sp, -104, %sp
+; V8-UNOPT-NEXT: ldd [%i0], %f4
+; V8-UNOPT-NEXT: ! implicit-def: $q0
+; V8-UNOPT-NEXT: fmovs %f4, %f0
+; V8-UNOPT-NEXT: fmovs %f5, %f1
+; V8-UNOPT-NEXT: ldd [%i0+8], %f4
+; V8-UNOPT-NEXT: fmovs %f4, %f2
+; V8-UNOPT-NEXT: fmovs %f5, %f3
+; V8-UNOPT-NEXT: fmovs %f2, %f4
+; V8-UNOPT-NEXT: fmovs %f3, %f5
+; V8-UNOPT-NEXT: std %f4, [%sp+100]
+; V8-UNOPT-NEXT: ! kill: def $d0 killed $d0 killed $q0
+; V8-UNOPT-NEXT: call __trunctfhf2
+; V8-UNOPT-NEXT: std %f0, [%sp+92]
+; V8-UNOPT-NEXT: sth %o0, [%i1]
+; V8-UNOPT-NEXT: ret
+; V8-UNOPT-NEXT: restore
+;
+; V9-LABEL: test_fptrunc_fp128:
+; V9: ! %bb.0:
+; V9-NEXT: save %sp, -104, %sp
+; V9-NEXT: ldd [%i0], %f0
+; V9-NEXT: ldd [%i0+8], %f2
+; V9-NEXT: std %f2, [%sp+100]
+; V9-NEXT: call __trunctfhf2
+; V9-NEXT: std %f0, [%sp+92]
+; V9-NEXT: sth %o0, [%i1]
+; V9-NEXT: ret
+; V9-NEXT: restore
+;
+; SPARC64-LABEL: test_fptrunc_fp128:
+; SPARC64: ! %bb.0:
+; SPARC64-NEXT: save %sp, -176, %sp
+; SPARC64-NEXT: ldd [%i0], %f0
+; SPARC64-NEXT: call __trunctfhf2
+; SPARC64-NEXT: ldd [%i0+8], %f2
+; SPARC64-NEXT: sth %o0, [%i1]
+; SPARC64-NEXT: ret
+; SPARC64-NEXT: restore
+ %d = load fp128, fp128* %dp
+ %a = fptrunc fp128 %d to half
+ store half %a, half* %p
+ ret void
+}
+
define void @test_fadd(half* %p, half* %q) nounwind {
; V8-OPT-LABEL: test_fadd:
; V8-OPT: ! %bb.0:
More information about the llvm-commits
mailing list