[llvm] f800368 - [AArch64] Add SVE lowering for vector.reduce.fminimum and fmaximum
David Green via llvm-commits
llvm-commits at lists.llvm.org
Sat Jun 24 03:13:02 PDT 2023
Author: David Green
Date: 2023-06-24T11:12:58+01:00
New Revision: f8003689f12d781f5128feb0655f12af37285085
URL: https://github.com/llvm/llvm-project/commit/f8003689f12d781f5128feb0655f12af37285085
DIFF: https://github.com/llvm/llvm-project/commit/f8003689f12d781f5128feb0655f12af37285085.diff
LOG: [AArch64] Add SVE lowering for vector.reduce.fminimum and fmaximum
Following what is already performed for vector.reduce.fmin/fmax, this adds
lowering for the new vector.reduce.fminimum/fmaximum nodes to the SVE fminv
and fmaxv instructions via the existing FMINV_PRED/FMAXV_PRED nodes.
Differential Revision: https://reviews.llvm.org/D153288
Added:
Modified:
llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
llvm/test/CodeGen/AArch64/sve-fixed-length-fp-reduce.ll
llvm/test/CodeGen/AArch64/sve-fp-reduce.ll
llvm/test/CodeGen/AArch64/sve-split-fp-reduce.ll
llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-fp-reduce.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index c8a4727744feb..8331104e49439 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -1447,6 +1447,8 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
setOperationAction(ISD::VECREDUCE_FADD, VT, Custom);
setOperationAction(ISD::VECREDUCE_FMAX, VT, Custom);
setOperationAction(ISD::VECREDUCE_FMIN, VT, Custom);
+ setOperationAction(ISD::VECREDUCE_FMAXIMUM, VT, Custom);
+ setOperationAction(ISD::VECREDUCE_FMINIMUM, VT, Custom);
setOperationAction(ISD::VECREDUCE_SEQ_FADD, VT, Custom);
setOperationAction(ISD::VECTOR_SPLICE, VT, Custom);
setOperationAction(ISD::VECTOR_DEINTERLEAVE, VT, Custom);
@@ -1858,6 +1860,8 @@ void AArch64TargetLowering::addTypeForFixedLengthSVE(MVT VT,
setOperationAction(ISD::VECREDUCE_FADD, VT, Custom);
setOperationAction(ISD::VECREDUCE_FMAX, VT, Custom);
setOperationAction(ISD::VECREDUCE_FMIN, VT, Custom);
+ setOperationAction(ISD::VECREDUCE_FMAXIMUM, VT, Custom);
+ setOperationAction(ISD::VECREDUCE_FMINIMUM, VT, Custom);
setOperationAction(ISD::VECREDUCE_OR, VT, Custom);
setOperationAction(ISD::VECREDUCE_SEQ_FADD, VT, Custom);
setOperationAction(ISD::VECREDUCE_SMAX, VT, Custom);
@@ -5979,6 +5983,8 @@ SDValue AArch64TargetLowering::LowerOperation(SDValue Op,
case ISD::VECREDUCE_FADD:
case ISD::VECREDUCE_FMAX:
case ISD::VECREDUCE_FMIN:
+ case ISD::VECREDUCE_FMAXIMUM:
+ case ISD::VECREDUCE_FMINIMUM:
return LowerVECREDUCE(Op, DAG);
case ISD::ATOMIC_LOAD_SUB:
return LowerATOMIC_LOAD_SUB(Op, DAG);
@@ -13574,6 +13580,10 @@ SDValue AArch64TargetLowering::LowerVECREDUCE(SDValue Op,
return LowerReductionToSVE(AArch64ISD::FMAXNMV_PRED, Op, DAG);
case ISD::VECREDUCE_FMIN:
return LowerReductionToSVE(AArch64ISD::FMINNMV_PRED, Op, DAG);
+ case ISD::VECREDUCE_FMAXIMUM:
+ return LowerReductionToSVE(AArch64ISD::FMAXV_PRED, Op, DAG);
+ case ISD::VECREDUCE_FMINIMUM:
+ return LowerReductionToSVE(AArch64ISD::FMINV_PRED, Op, DAG);
default:
llvm_unreachable("Unhandled fixed length reduction");
}
diff --git a/llvm/test/CodeGen/AArch64/sve-fixed-length-fp-reduce.ll b/llvm/test/CodeGen/AArch64/sve-fixed-length-fp-reduce.ll
index 39784afe329e3..7fc0ab12e8d28 100644
--- a/llvm/test/CodeGen/AArch64/sve-fixed-length-fp-reduce.ll
+++ b/llvm/test/CodeGen/AArch64/sve-fixed-length-fp-reduce.ll
@@ -54,7 +54,7 @@ define half @fadda_v16f16(half %start, ptr %a) vscale_range(2,0) #0 {
define half @fadda_v32f16(half %start, ptr %a) #0 {
; VBITS_GE_256-LABEL: fadda_v32f16:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #16
+; VBITS_GE_256-NEXT: mov x8, #16 // =0x10
; VBITS_GE_256-NEXT: ptrue p0.h, vl16
; VBITS_GE_256-NEXT: // kill: def $h0 killed $h0 def $z0
; VBITS_GE_256-NEXT: ld1h { z1.h }, p0/z, [x0, x8, lsl #1]
@@ -150,7 +150,7 @@ define float @fadda_v8f32(float %start, ptr %a) vscale_range(2,0) #0 {
define float @fadda_v16f32(float %start, ptr %a) #0 {
; VBITS_GE_256-LABEL: fadda_v16f32:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #8
+; VBITS_GE_256-NEXT: mov x8, #8 // =0x8
; VBITS_GE_256-NEXT: ptrue p0.s, vl8
; VBITS_GE_256-NEXT: // kill: def $s0 killed $s0 def $z0
; VBITS_GE_256-NEXT: ld1w { z1.s }, p0/z, [x0, x8, lsl #2]
@@ -242,7 +242,7 @@ define double @fadda_v4f64(double %start, ptr %a) vscale_range(2,0) #0 {
define double @fadda_v8f64(double %start, ptr %a) #0 {
; VBITS_GE_256-LABEL: fadda_v8f64:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #4
+; VBITS_GE_256-NEXT: mov x8, #4 // =0x4
; VBITS_GE_256-NEXT: ptrue p0.d, vl4
; VBITS_GE_256-NEXT: // kill: def $d0 killed $d0 def $z0
; VBITS_GE_256-NEXT: ld1d { z1.d }, p0/z, [x0, x8, lsl #3]
@@ -339,7 +339,7 @@ define half @faddv_v16f16(half %start, ptr %a) vscale_range(2,0) #0 {
define half @faddv_v32f16(half %start, ptr %a) #0 {
; VBITS_GE_256-LABEL: faddv_v32f16:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #16
+; VBITS_GE_256-NEXT: mov x8, #16 // =0x10
; VBITS_GE_256-NEXT: ptrue p0.h, vl16
; VBITS_GE_256-NEXT: ld1h { z1.h }, p0/z, [x0, x8, lsl #1]
; VBITS_GE_256-NEXT: ld1h { z2.h }, p0/z, [x0]
@@ -426,7 +426,7 @@ define float @faddv_v8f32(float %start, ptr %a) vscale_range(2,0) #0 {
define float @faddv_v16f32(float %start, ptr %a) #0 {
; VBITS_GE_256-LABEL: faddv_v16f32:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #8
+; VBITS_GE_256-NEXT: mov x8, #8 // =0x8
; VBITS_GE_256-NEXT: ptrue p0.s, vl8
; VBITS_GE_256-NEXT: ld1w { z1.s }, p0/z, [x0, x8, lsl #2]
; VBITS_GE_256-NEXT: ld1w { z2.s }, p0/z, [x0]
@@ -510,7 +510,7 @@ define double @faddv_v4f64(double %start, ptr %a) vscale_range(2,0) #0 {
define double @faddv_v8f64(double %start, ptr %a) #0 {
; VBITS_GE_256-LABEL: faddv_v8f64:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #4
+; VBITS_GE_256-NEXT: mov x8, #4 // =0x4
; VBITS_GE_256-NEXT: ptrue p0.d, vl4
; VBITS_GE_256-NEXT: ld1d { z1.d }, p0/z, [x0, x8, lsl #3]
; VBITS_GE_256-NEXT: ld1d { z2.d }, p0/z, [x0]
@@ -558,7 +558,7 @@ define double @faddv_v32f64(double %start, ptr %a) vscale_range(16,0) #0 {
}
;
-; FMAXV
+; FMAXNMV
;
; No NEON 16-bit vector FMAXNMV support. Use SVE.
@@ -597,7 +597,7 @@ define half @fmaxv_v16f16(ptr %a) vscale_range(2,0) #0 {
define half @fmaxv_v32f16(ptr %a) #0 {
; VBITS_GE_256-LABEL: fmaxv_v32f16:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #16
+; VBITS_GE_256-NEXT: mov x8, #16 // =0x10
; VBITS_GE_256-NEXT: ptrue p0.h, vl16
; VBITS_GE_256-NEXT: ld1h { z0.h }, p0/z, [x0, x8, lsl #1]
; VBITS_GE_256-NEXT: ld1h { z1.h }, p0/z, [x0]
@@ -680,7 +680,7 @@ define float @fmaxv_v8f32(ptr %a) vscale_range(2,0) #0 {
define float @fmaxv_v16f32(ptr %a) #0 {
; VBITS_GE_256-LABEL: fmaxv_v16f32:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #8
+; VBITS_GE_256-NEXT: mov x8, #8 // =0x8
; VBITS_GE_256-NEXT: ptrue p0.s, vl8
; VBITS_GE_256-NEXT: ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
; VBITS_GE_256-NEXT: ld1w { z1.s }, p0/z, [x0]
@@ -762,7 +762,7 @@ define double @fmaxv_v4f64(ptr %a) vscale_range(2,0) #0 {
define double @fmaxv_v8f64(ptr %a) #0 {
; VBITS_GE_256-LABEL: fmaxv_v8f64:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #4
+; VBITS_GE_256-NEXT: mov x8, #4 // =0x4
; VBITS_GE_256-NEXT: ptrue p0.d, vl4
; VBITS_GE_256-NEXT: ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
; VBITS_GE_256-NEXT: ld1d { z1.d }, p0/z, [x0]
@@ -810,7 +810,7 @@ define double @fmaxv_v32f64(ptr %a) vscale_range(16,0) #0 {
}
;
-; FMINV
+; FMINNMV
;
; No NEON 16-bit vector FMINNMV support. Use SVE.
@@ -849,7 +849,7 @@ define half @fminv_v16f16(ptr %a) vscale_range(2,0) #0 {
define half @fminv_v32f16(ptr %a) #0 {
; VBITS_GE_256-LABEL: fminv_v32f16:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #16
+; VBITS_GE_256-NEXT: mov x8, #16 // =0x10
; VBITS_GE_256-NEXT: ptrue p0.h, vl16
; VBITS_GE_256-NEXT: ld1h { z0.h }, p0/z, [x0, x8, lsl #1]
; VBITS_GE_256-NEXT: ld1h { z1.h }, p0/z, [x0]
@@ -932,7 +932,7 @@ define float @fminv_v8f32(ptr %a) vscale_range(2,0) #0 {
define float @fminv_v16f32(ptr %a) #0 {
; VBITS_GE_256-LABEL: fminv_v16f32:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #8
+; VBITS_GE_256-NEXT: mov x8, #8 // =0x8
; VBITS_GE_256-NEXT: ptrue p0.s, vl8
; VBITS_GE_256-NEXT: ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
; VBITS_GE_256-NEXT: ld1w { z1.s }, p0/z, [x0]
@@ -1014,7 +1014,7 @@ define double @fminv_v4f64(ptr %a) vscale_range(2,0) #0 {
define double @fminv_v8f64(ptr %a) #0 {
; VBITS_GE_256-LABEL: fminv_v8f64:
; VBITS_GE_256: // %bb.0:
-; VBITS_GE_256-NEXT: mov x8, #4
+; VBITS_GE_256-NEXT: mov x8, #4 // =0x4
; VBITS_GE_256-NEXT: ptrue p0.d, vl4
; VBITS_GE_256-NEXT: ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
; VBITS_GE_256-NEXT: ld1d { z1.d }, p0/z, [x0]
@@ -1061,6 +1061,506 @@ define double @fminv_v32f64(ptr %a) vscale_range(16,0) #0 {
ret double %res
}
+;
+; FMAXV
+;
+
+define half @fmaximumv_v4f16(<4 x half> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fmaximumv_v4f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmaxv h0, v0.4h
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fmaximum.v4f16(<4 x half> %a)
+ ret half %res
+}
+
+define half @fmaximumv_v8f16(<8 x half> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fmaximumv_v8f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmaxv h0, v0.8h
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fmaximum.v8f16(<8 x half> %a)
+ ret half %res
+}
+
+define half @fmaximumv_v16f16(ptr %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fmaximumv_v16f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.h, vl16
+; CHECK-NEXT: ld1h { z0.h }, p0/z, [x0]
+; CHECK-NEXT: fmaxv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <16 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fmaximum.v16f16(<16 x half> %op)
+ ret half %res
+}
+
+define half @fmaximumv_v32f16(ptr %a) #0 {
+; VBITS_GE_256-LABEL: fmaximumv_v32f16:
+; VBITS_GE_256: // %bb.0:
+; VBITS_GE_256-NEXT: mov x8, #16 // =0x10
+; VBITS_GE_256-NEXT: ptrue p0.h, vl16
+; VBITS_GE_256-NEXT: ld1h { z0.h }, p0/z, [x0, x8, lsl #1]
+; VBITS_GE_256-NEXT: ld1h { z1.h }, p0/z, [x0]
+; VBITS_GE_256-NEXT: fmax z0.h, p0/m, z0.h, z1.h
+; VBITS_GE_256-NEXT: fmaxv h0, p0, z0.h
+; VBITS_GE_256-NEXT: // kill: def $h0 killed $h0 killed $z0
+; VBITS_GE_256-NEXT: ret
+;
+; VBITS_GE_512-LABEL: fmaximumv_v32f16:
+; VBITS_GE_512: // %bb.0:
+; VBITS_GE_512-NEXT: ptrue p0.h, vl32
+; VBITS_GE_512-NEXT: ld1h { z0.h }, p0/z, [x0]
+; VBITS_GE_512-NEXT: fmaxv h0, p0, z0.h
+; VBITS_GE_512-NEXT: // kill: def $h0 killed $h0 killed $z0
+; VBITS_GE_512-NEXT: ret
+ %op = load <32 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fmaximum.v32f16(<32 x half> %op)
+ ret half %res
+}
+
+define half @fmaximumv_v64f16(ptr %a) vscale_range(8,0) #0 {
+; CHECK-LABEL: fmaximumv_v64f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.h, vl64
+; CHECK-NEXT: ld1h { z0.h }, p0/z, [x0]
+; CHECK-NEXT: fmaxv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <64 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fmaximum.v64f16(<64 x half> %op)
+ ret half %res
+}
+
+define half @fmaximumv_v128f16(ptr %a) vscale_range(16,0) #0 {
+; CHECK-LABEL: fmaximumv_v128f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.h, vl128
+; CHECK-NEXT: ld1h { z0.h }, p0/z, [x0]
+; CHECK-NEXT: fmaxv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <128 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fmaximum.v128f16(<128 x half> %op)
+ ret half %res
+}
+
+; Don't use SVE for 64-bit f32 vectors.
+define float @fmaximumv_v2f32(<2 x float> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fmaximumv_v2f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmaxp s0, v0.2s
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fmaximum.v2f32(<2 x float> %a)
+ ret float %res
+}
+
+; Don't use SVE for 128-bit f32 vectors.
+define float @fmaximumv_v4f32(<4 x float> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fmaximumv_v4f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmaxv s0, v0.4s
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fmaximum.v4f32(<4 x float> %a)
+ ret float %res
+}
+
+define float @fmaximumv_v8f32(ptr %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fmaximumv_v8f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s, vl8
+; CHECK-NEXT: ld1w { z0.s }, p0/z, [x0]
+; CHECK-NEXT: fmaxv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <8 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fmaximum.v8f32(<8 x float> %op)
+ ret float %res
+}
+
+define float @fmaximumv_v16f32(ptr %a) #0 {
+; VBITS_GE_256-LABEL: fmaximumv_v16f32:
+; VBITS_GE_256: // %bb.0:
+; VBITS_GE_256-NEXT: mov x8, #8 // =0x8
+; VBITS_GE_256-NEXT: ptrue p0.s, vl8
+; VBITS_GE_256-NEXT: ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
+; VBITS_GE_256-NEXT: ld1w { z1.s }, p0/z, [x0]
+; VBITS_GE_256-NEXT: fmax z0.s, p0/m, z0.s, z1.s
+; VBITS_GE_256-NEXT: fmaxv s0, p0, z0.s
+; VBITS_GE_256-NEXT: // kill: def $s0 killed $s0 killed $z0
+; VBITS_GE_256-NEXT: ret
+;
+; VBITS_GE_512-LABEL: fmaximumv_v16f32:
+; VBITS_GE_512: // %bb.0:
+; VBITS_GE_512-NEXT: ptrue p0.s, vl16
+; VBITS_GE_512-NEXT: ld1w { z0.s }, p0/z, [x0]
+; VBITS_GE_512-NEXT: fmaxv s0, p0, z0.s
+; VBITS_GE_512-NEXT: // kill: def $s0 killed $s0 killed $z0
+; VBITS_GE_512-NEXT: ret
+ %op = load <16 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fmaximum.v16f32(<16 x float> %op)
+ ret float %res
+}
+
+define float @fmaximumv_v32f32(ptr %a) vscale_range(8,0) #0 {
+; CHECK-LABEL: fmaximumv_v32f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s, vl32
+; CHECK-NEXT: ld1w { z0.s }, p0/z, [x0]
+; CHECK-NEXT: fmaxv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <32 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fmaximum.v32f32(<32 x float> %op)
+ ret float %res
+}
+
+define float @fmaximumv_v64f32(ptr %a) vscale_range(16,0) #0 {
+; CHECK-LABEL: fmaximumv_v64f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s, vl64
+; CHECK-NEXT: ld1w { z0.s }, p0/z, [x0]
+; CHECK-NEXT: fmaxv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <64 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fmaximum.v64f32(<64 x float> %op)
+ ret float %res
+}
+
+; Nothing to do for single element vectors.
+define double @fmaximumv_v1f64(<1 x double> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fmaximumv_v1f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fmaximum.v1f64(<1 x double> %a)
+ ret double %res
+}
+
+; Don't use SVE for 128-bit f64 vectors.
+define double @fmaximumv_v2f64(<2 x double> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fmaximumv_v2f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmaxp d0, v0.2d
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fmaximum.v2f64(<2 x double> %a)
+ ret double %res
+}
+
+define double @fmaximumv_v4f64(ptr %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fmaximumv_v4f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d, vl4
+; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0]
+; CHECK-NEXT: fmaxv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <4 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fmaximum.v4f64(<4 x double> %op)
+ ret double %res
+}
+
+define double @fmaximumv_v8f64(ptr %a) #0 {
+; VBITS_GE_256-LABEL: fmaximumv_v8f64:
+; VBITS_GE_256: // %bb.0:
+; VBITS_GE_256-NEXT: mov x8, #4 // =0x4
+; VBITS_GE_256-NEXT: ptrue p0.d, vl4
+; VBITS_GE_256-NEXT: ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
+; VBITS_GE_256-NEXT: ld1d { z1.d }, p0/z, [x0]
+; VBITS_GE_256-NEXT: fmax z0.d, p0/m, z0.d, z1.d
+; VBITS_GE_256-NEXT: fmaxv d0, p0, z0.d
+; VBITS_GE_256-NEXT: // kill: def $d0 killed $d0 killed $z0
+; VBITS_GE_256-NEXT: ret
+;
+; VBITS_GE_512-LABEL: fmaximumv_v8f64:
+; VBITS_GE_512: // %bb.0:
+; VBITS_GE_512-NEXT: ptrue p0.d, vl8
+; VBITS_GE_512-NEXT: ld1d { z0.d }, p0/z, [x0]
+; VBITS_GE_512-NEXT: fmaxv d0, p0, z0.d
+; VBITS_GE_512-NEXT: // kill: def $d0 killed $d0 killed $z0
+; VBITS_GE_512-NEXT: ret
+ %op = load <8 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fmaximum.v8f64(<8 x double> %op)
+ ret double %res
+}
+
+define double @fmaximumv_v16f64(ptr %a) vscale_range(8,0) #0 {
+; CHECK-LABEL: fmaximumv_v16f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d, vl16
+; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0]
+; CHECK-NEXT: fmaxv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <16 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fmaximum.v16f64(<16 x double> %op)
+ ret double %res
+}
+
+define double @fmaximumv_v32f64(ptr %a) vscale_range(16,0) #0 {
+; CHECK-LABEL: fmaximumv_v32f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d, vl32
+; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0]
+; CHECK-NEXT: fmaxv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <32 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fmaximum.v32f64(<32 x double> %op)
+ ret double %res
+}
+
+;
+; FMINV
+;
+
+define half @fminimumv_v4f16(<4 x half> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fminimumv_v4f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fminv h0, v0.4h
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fminimum.v4f16(<4 x half> %a)
+ ret half %res
+}
+
+define half @fminimumv_v8f16(<8 x half> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fminimumv_v8f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fminv h0, v0.8h
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fminimum.v8f16(<8 x half> %a)
+ ret half %res
+}
+
+define half @fminimumv_v16f16(ptr %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fminimumv_v16f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.h, vl16
+; CHECK-NEXT: ld1h { z0.h }, p0/z, [x0]
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <16 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fminimum.v16f16(<16 x half> %op)
+ ret half %res
+}
+
+define half @fminimumv_v32f16(ptr %a) #0 {
+; VBITS_GE_256-LABEL: fminimumv_v32f16:
+; VBITS_GE_256: // %bb.0:
+; VBITS_GE_256-NEXT: mov x8, #16 // =0x10
+; VBITS_GE_256-NEXT: ptrue p0.h, vl16
+; VBITS_GE_256-NEXT: ld1h { z0.h }, p0/z, [x0, x8, lsl #1]
+; VBITS_GE_256-NEXT: ld1h { z1.h }, p0/z, [x0]
+; VBITS_GE_256-NEXT: fmin z0.h, p0/m, z0.h, z1.h
+; VBITS_GE_256-NEXT: fminv h0, p0, z0.h
+; VBITS_GE_256-NEXT: // kill: def $h0 killed $h0 killed $z0
+; VBITS_GE_256-NEXT: ret
+;
+; VBITS_GE_512-LABEL: fminimumv_v32f16:
+; VBITS_GE_512: // %bb.0:
+; VBITS_GE_512-NEXT: ptrue p0.h, vl32
+; VBITS_GE_512-NEXT: ld1h { z0.h }, p0/z, [x0]
+; VBITS_GE_512-NEXT: fminv h0, p0, z0.h
+; VBITS_GE_512-NEXT: // kill: def $h0 killed $h0 killed $z0
+; VBITS_GE_512-NEXT: ret
+ %op = load <32 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fminimum.v32f16(<32 x half> %op)
+ ret half %res
+}
+
+define half @fminimumv_v64f16(ptr %a) vscale_range(8,0) #0 {
+; CHECK-LABEL: fminimumv_v64f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.h, vl64
+; CHECK-NEXT: ld1h { z0.h }, p0/z, [x0]
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <64 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fminimum.v64f16(<64 x half> %op)
+ ret half %res
+}
+
+define half @fminimumv_v128f16(ptr %a) vscale_range(16,0) #0 {
+; CHECK-LABEL: fminimumv_v128f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.h, vl128
+; CHECK-NEXT: ld1h { z0.h }, p0/z, [x0]
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <128 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fminimum.v128f16(<128 x half> %op)
+ ret half %res
+}
+
+; Don't use SVE for 64-bit f32 vectors.
+define float @fminimumv_v2f32(<2 x float> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fminimumv_v2f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fminp s0, v0.2s
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fminimum.v2f32(<2 x float> %a)
+ ret float %res
+}
+
+; Don't use SVE for 128-bit f32 vectors.
+define float @fminimumv_v4f32(<4 x float> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fminimumv_v4f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fminv s0, v0.4s
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fminimum.v4f32(<4 x float> %a)
+ ret float %res
+}
+
+define float @fminimumv_v8f32(ptr %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fminimumv_v8f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s, vl8
+; CHECK-NEXT: ld1w { z0.s }, p0/z, [x0]
+; CHECK-NEXT: fminv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <8 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fminimum.v8f32(<8 x float> %op)
+ ret float %res
+}
+
+define float @fminimumv_v16f32(ptr %a) #0 {
+; VBITS_GE_256-LABEL: fminimumv_v16f32:
+; VBITS_GE_256: // %bb.0:
+; VBITS_GE_256-NEXT: mov x8, #8 // =0x8
+; VBITS_GE_256-NEXT: ptrue p0.s, vl8
+; VBITS_GE_256-NEXT: ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
+; VBITS_GE_256-NEXT: ld1w { z1.s }, p0/z, [x0]
+; VBITS_GE_256-NEXT: fmin z0.s, p0/m, z0.s, z1.s
+; VBITS_GE_256-NEXT: fminv s0, p0, z0.s
+; VBITS_GE_256-NEXT: // kill: def $s0 killed $s0 killed $z0
+; VBITS_GE_256-NEXT: ret
+;
+; VBITS_GE_512-LABEL: fminimumv_v16f32:
+; VBITS_GE_512: // %bb.0:
+; VBITS_GE_512-NEXT: ptrue p0.s, vl16
+; VBITS_GE_512-NEXT: ld1w { z0.s }, p0/z, [x0]
+; VBITS_GE_512-NEXT: fminv s0, p0, z0.s
+; VBITS_GE_512-NEXT: // kill: def $s0 killed $s0 killed $z0
+; VBITS_GE_512-NEXT: ret
+ %op = load <16 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fminimum.v16f32(<16 x float> %op)
+ ret float %res
+}
+
+define float @fminimumv_v32f32(ptr %a) vscale_range(8,0) #0 {
+; CHECK-LABEL: fminimumv_v32f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s, vl32
+; CHECK-NEXT: ld1w { z0.s }, p0/z, [x0]
+; CHECK-NEXT: fminv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <32 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fminimum.v32f32(<32 x float> %op)
+ ret float %res
+}
+
+define float @fminimumv_v64f32(ptr %a) vscale_range(16,0) #0 {
+; CHECK-LABEL: fminimumv_v64f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s, vl64
+; CHECK-NEXT: ld1w { z0.s }, p0/z, [x0]
+; CHECK-NEXT: fminv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <64 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fminimum.v64f32(<64 x float> %op)
+ ret float %res
+}
+
+; Nothing to do for single element vectors.
+define double @fminimumv_v1f64(<1 x double> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fminimumv_v1f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fminimum.v1f64(<1 x double> %a)
+ ret double %res
+}
+
+; Don't use SVE for 128-bit f64 vectors.
+define double @fminimumv_v2f64(<2 x double> %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fminimumv_v2f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fminp d0, v0.2d
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fminimum.v2f64(<2 x double> %a)
+ ret double %res
+}
+
+define double @fminimumv_v4f64(ptr %a) vscale_range(2,0) #0 {
+; CHECK-LABEL: fminimumv_v4f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d, vl4
+; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0]
+; CHECK-NEXT: fminv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <4 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fminimum.v4f64(<4 x double> %op)
+ ret double %res
+}
+
+define double @fminimumv_v8f64(ptr %a) #0 {
+; VBITS_GE_256-LABEL: fminimumv_v8f64:
+; VBITS_GE_256: // %bb.0:
+; VBITS_GE_256-NEXT: mov x8, #4 // =0x4
+; VBITS_GE_256-NEXT: ptrue p0.d, vl4
+; VBITS_GE_256-NEXT: ld1d { z0.d }, p0/z, [x0, x8, lsl #3]
+; VBITS_GE_256-NEXT: ld1d { z1.d }, p0/z, [x0]
+; VBITS_GE_256-NEXT: fmin z0.d, p0/m, z0.d, z1.d
+; VBITS_GE_256-NEXT: fminv d0, p0, z0.d
+; VBITS_GE_256-NEXT: // kill: def $d0 killed $d0 killed $z0
+; VBITS_GE_256-NEXT: ret
+;
+; VBITS_GE_512-LABEL: fminimumv_v8f64:
+; VBITS_GE_512: // %bb.0:
+; VBITS_GE_512-NEXT: ptrue p0.d, vl8
+; VBITS_GE_512-NEXT: ld1d { z0.d }, p0/z, [x0]
+; VBITS_GE_512-NEXT: fminv d0, p0, z0.d
+; VBITS_GE_512-NEXT: // kill: def $d0 killed $d0 killed $z0
+; VBITS_GE_512-NEXT: ret
+ %op = load <8 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fminimum.v8f64(<8 x double> %op)
+ ret double %res
+}
+
+define double @fminimumv_v16f64(ptr %a) vscale_range(8,0) #0 {
+; CHECK-LABEL: fminimumv_v16f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d, vl16
+; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0]
+; CHECK-NEXT: fminv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <16 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fminimum.v16f64(<16 x double> %op)
+ ret double %res
+}
+
+define double @fminimumv_v32f64(ptr %a) vscale_range(16,0) #0 {
+; CHECK-LABEL: fminimumv_v32f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d, vl32
+; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0]
+; CHECK-NEXT: fminv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <32 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fminimum.v32f64(<32 x double> %op)
+ ret double %res
+}
+
attributes #0 = { "target-features"="+sve" }
declare half @llvm.vector.reduce.fadd.v4f16(half, <4 x half>)
@@ -1125,3 +1625,45 @@ declare double @llvm.vector.reduce.fmin.v4f64(<4 x double>)
declare double @llvm.vector.reduce.fmin.v8f64(<8 x double>)
declare double @llvm.vector.reduce.fmin.v16f64(<16 x double>)
declare double @llvm.vector.reduce.fmin.v32f64(<32 x double>)
+
+declare half @llvm.vector.reduce.fmaximum.v4f16(<4 x half>)
+declare half @llvm.vector.reduce.fmaximum.v8f16(<8 x half>)
+declare half @llvm.vector.reduce.fmaximum.v16f16(<16 x half>)
+declare half @llvm.vector.reduce.fmaximum.v32f16(<32 x half>)
+declare half @llvm.vector.reduce.fmaximum.v64f16(<64 x half>)
+declare half @llvm.vector.reduce.fmaximum.v128f16(<128 x half>)
+
+declare float @llvm.vector.reduce.fmaximum.v2f32(<2 x float>)
+declare float @llvm.vector.reduce.fmaximum.v4f32(<4 x float>)
+declare float @llvm.vector.reduce.fmaximum.v8f32(<8 x float>)
+declare float @llvm.vector.reduce.fmaximum.v16f32(<16 x float>)
+declare float @llvm.vector.reduce.fmaximum.v32f32(<32 x float>)
+declare float @llvm.vector.reduce.fmaximum.v64f32(<64 x float>)
+
+declare double @llvm.vector.reduce.fmaximum.v1f64(<1 x double>)
+declare double @llvm.vector.reduce.fmaximum.v2f64(<2 x double>)
+declare double @llvm.vector.reduce.fmaximum.v4f64(<4 x double>)
+declare double @llvm.vector.reduce.fmaximum.v8f64(<8 x double>)
+declare double @llvm.vector.reduce.fmaximum.v16f64(<16 x double>)
+declare double @llvm.vector.reduce.fmaximum.v32f64(<32 x double>)
+
+declare half @llvm.vector.reduce.fminimum.v4f16(<4 x half>)
+declare half @llvm.vector.reduce.fminimum.v8f16(<8 x half>)
+declare half @llvm.vector.reduce.fminimum.v16f16(<16 x half>)
+declare half @llvm.vector.reduce.fminimum.v32f16(<32 x half>)
+declare half @llvm.vector.reduce.fminimum.v64f16(<64 x half>)
+declare half @llvm.vector.reduce.fminimum.v128f16(<128 x half>)
+
+declare float @llvm.vector.reduce.fminimum.v2f32(<2 x float>)
+declare float @llvm.vector.reduce.fminimum.v4f32(<4 x float>)
+declare float @llvm.vector.reduce.fminimum.v8f32(<8 x float>)
+declare float @llvm.vector.reduce.fminimum.v16f32(<16 x float>)
+declare float @llvm.vector.reduce.fminimum.v32f32(<32 x float>)
+declare float @llvm.vector.reduce.fminimum.v64f32(<64 x float>)
+
+declare double @llvm.vector.reduce.fminimum.v1f64(<1 x double>)
+declare double @llvm.vector.reduce.fminimum.v2f64(<2 x double>)
+declare double @llvm.vector.reduce.fminimum.v4f64(<4 x double>)
+declare double @llvm.vector.reduce.fminimum.v8f64(<8 x double>)
+declare double @llvm.vector.reduce.fminimum.v16f64(<16 x double>)
+declare double @llvm.vector.reduce.fminimum.v32f64(<32 x double>)
diff --git a/llvm/test/CodeGen/AArch64/sve-fp-reduce.ll b/llvm/test/CodeGen/AArch64/sve-fp-reduce.ll
index 4183a83ed01b1..2c91d1d34e867 100644
--- a/llvm/test/CodeGen/AArch64/sve-fp-reduce.ll
+++ b/llvm/test/CodeGen/AArch64/sve-fp-reduce.ll
@@ -47,7 +47,7 @@ define half @fadda_nxv6f16(<vscale x 6 x half> %v, half %s) {
; CHECK-NEXT: .cfi_offset w29, -16
; CHECK-NEXT: addvl sp, sp, #-1
; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x08, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 8 * VG
-; CHECK-NEXT: mov w8, #32768
+; CHECK-NEXT: mov w8, #32768 // =0x8000
; CHECK-NEXT: ptrue p0.h
; CHECK-NEXT: ptrue p1.d
; CHECK-NEXT: st1h { z0.h }, p0, [sp]
@@ -72,7 +72,7 @@ define half @fadda_nxv10f16(<vscale x 10 x half> %v, half %s) {
; CHECK-NEXT: .cfi_offset w29, -16
; CHECK-NEXT: addvl sp, sp, #-3
; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x18, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 24 * VG
-; CHECK-NEXT: mov w8, #32768
+; CHECK-NEXT: mov w8, #32768 // =0x8000
; CHECK-NEXT: ptrue p0.h
; CHECK-NEXT: ptrue p1.d
; CHECK-NEXT: st1h { z1.h }, p0, [sp]
@@ -100,7 +100,7 @@ define half @fadda_nxv10f16(<vscale x 10 x half> %v, half %s) {
define half @fadda_nxv12f16(<vscale x 12 x half> %v, half %s) {
; CHECK-LABEL: fadda_nxv12f16:
; CHECK: // %bb.0:
-; CHECK-NEXT: mov w8, #32768
+; CHECK-NEXT: mov w8, #32768 // =0x8000
; CHECK-NEXT: // kill: def $h2 killed $h2 def $z2
; CHECK-NEXT: uunpklo z1.s, z1.h
; CHECK-NEXT: ptrue p0.h
@@ -218,7 +218,7 @@ define double @faddv_nxv2f64(double %init, <vscale x 2 x double> %a) {
ret double %res
}
-; FMAXV
+; FMAXNMV
define half @fmaxv_nxv2f16(<vscale x 2 x half> %a) {
; CHECK-LABEL: fmaxv_nxv2f16:
@@ -286,7 +286,7 @@ define double @fmaxv_nxv2f64(<vscale x 2 x double> %a) {
ret double %res
}
-; FMINV
+; FMINNMV
define half @fminv_nxv2f16(<vscale x 2 x half> %a) {
; CHECK-LABEL: fminv_nxv2f16:
@@ -354,6 +354,145 @@ define double @fminv_nxv2f64(<vscale x 2 x double> %a) {
ret double %res
}
+
+
+
+; FMAXV
+
+define half @fmaximumv_nxv2f16(<vscale x 2 x half> %a) {
+; CHECK-LABEL: fmaximumv_nxv2f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d
+; CHECK-NEXT: fmaxv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fmaximum.nxv2f16(<vscale x 2 x half> %a)
+ ret half %res
+}
+
+define half @fmaximumv_nxv4f16(<vscale x 4 x half> %a) {
+; CHECK-LABEL: fmaximumv_nxv4f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: fmaxv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fmaximum.nxv4f16(<vscale x 4 x half> %a)
+ ret half %res
+}
+
+define half @fmaximumv_nxv8f16(<vscale x 8 x half> %a) {
+; CHECK-LABEL: fmaximumv_nxv8f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.h
+; CHECK-NEXT: fmaxv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fmaximum.nxv8f16(<vscale x 8 x half> %a)
+ ret half %res
+}
+
+define float @fmaximumv_nxv2f32(<vscale x 2 x float> %a) {
+; CHECK-LABEL: fmaximumv_nxv2f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d
+; CHECK-NEXT: fmaxv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fmaximum.nxv2f32(<vscale x 2 x float> %a)
+ ret float %res
+}
+
+define float @fmaximumv_nxv4f32(<vscale x 4 x float> %a) {
+; CHECK-LABEL: fmaximumv_nxv4f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: fmaxv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fmaximum.nxv4f32(<vscale x 4 x float> %a)
+ ret float %res
+}
+
+define double @fmaximumv_nxv2f64(<vscale x 2 x double> %a) {
+; CHECK-LABEL: fmaximumv_nxv2f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d
+; CHECK-NEXT: fmaxv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fmaximum.nxv2f64(<vscale x 2 x double> %a)
+ ret double %res
+}
+
+; FMINV
+
+define half @fminimumv_nxv2f16(<vscale x 2 x half> %a) {
+; CHECK-LABEL: fminimumv_nxv2f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fminimum.nxv2f16(<vscale x 2 x half> %a)
+ ret half %res
+}
+
+define half @fminimumv_nxv4f16(<vscale x 4 x half> %a) {
+; CHECK-LABEL: fminimumv_nxv4f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fminimum.nxv4f16(<vscale x 4 x half> %a)
+ ret half %res
+}
+
+define half @fminimumv_nxv8f16(<vscale x 8 x half> %a) {
+; CHECK-LABEL: fminimumv_nxv8f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.h
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fminimum.nxv8f16(<vscale x 8 x half> %a)
+ ret half %res
+}
+
+define float @fminimumv_nxv2f32(<vscale x 2 x float> %a) {
+; CHECK-LABEL: fminimumv_nxv2f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d
+; CHECK-NEXT: fminv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fminimum.nxv2f32(<vscale x 2 x float> %a)
+ ret float %res
+}
+
+define float @fminimumv_nxv4f32(<vscale x 4 x float> %a) {
+; CHECK-LABEL: fminimumv_nxv4f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: fminv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fminimum.nxv4f32(<vscale x 4 x float> %a)
+ ret float %res
+}
+
+define double @fminimumv_nxv2f64(<vscale x 2 x double> %a) {
+; CHECK-LABEL: fminimumv_nxv2f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d
+; CHECK-NEXT: fminv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fminimum.nxv2f64(<vscale x 2 x double> %a)
+ ret double %res
+}
+
define float @fadd_reduct_reassoc_v4v8f32(<vscale x 4 x float> %a, <vscale x 8 x float> %b) {
; CHECK-LABEL: fadd_reduct_reassoc_v4v8f32:
; CHECK: // %bb.0:
@@ -393,3 +532,17 @@ declare half @llvm.vector.reduce.fmin.nxv8f16(<vscale x 8 x half>)
declare float @llvm.vector.reduce.fmin.nxv2f32(<vscale x 2 x float>)
declare float @llvm.vector.reduce.fmin.nxv4f32(<vscale x 4 x float>)
declare double @llvm.vector.reduce.fmin.nxv2f64(<vscale x 2 x double>)
+
+declare half @llvm.vector.reduce.fmaximum.nxv2f16(<vscale x 2 x half>)
+declare half @llvm.vector.reduce.fmaximum.nxv4f16(<vscale x 4 x half>)
+declare half @llvm.vector.reduce.fmaximum.nxv8f16(<vscale x 8 x half>)
+declare float @llvm.vector.reduce.fmaximum.nxv2f32(<vscale x 2 x float>)
+declare float @llvm.vector.reduce.fmaximum.nxv4f32(<vscale x 4 x float>)
+declare double @llvm.vector.reduce.fmaximum.nxv2f64(<vscale x 2 x double>)
+
+declare half @llvm.vector.reduce.fminimum.nxv2f16(<vscale x 2 x half>)
+declare half @llvm.vector.reduce.fminimum.nxv4f16(<vscale x 4 x half>)
+declare half @llvm.vector.reduce.fminimum.nxv8f16(<vscale x 8 x half>)
+declare float @llvm.vector.reduce.fminimum.nxv2f32(<vscale x 2 x float>)
+declare float @llvm.vector.reduce.fminimum.nxv4f32(<vscale x 4 x float>)
+declare double @llvm.vector.reduce.fminimum.nxv2f64(<vscale x 2 x double>)
diff --git a/llvm/test/CodeGen/AArch64/sve-split-fp-reduce.ll b/llvm/test/CodeGen/AArch64/sve-split-fp-reduce.ll
index 0655db6f2b694..53200c9a56fd5 100644
--- a/llvm/test/CodeGen/AArch64/sve-split-fp-reduce.ll
+++ b/llvm/test/CodeGen/AArch64/sve-split-fp-reduce.ll
@@ -32,7 +32,7 @@ define float @faddv_nxv8f32(float %init, <vscale x 8 x float> %a) {
ret float %res
}
-; FMAXV
+; FMAXNMV
define double @fmaxv_nxv8f64(<vscale x 8 x double> %a) {
; CHECK-LABEL: fmaxv_nxv8f64:
@@ -48,7 +48,7 @@ define double @fmaxv_nxv8f64(<vscale x 8 x double> %a) {
ret double %res
}
-; FMINV
+; FMINNMV
define half @fminv_nxv16f16(<vscale x 16 x half> %a) {
; CHECK-LABEL: fminv_nxv16f16:
@@ -62,9 +62,40 @@ define half @fminv_nxv16f16(<vscale x 16 x half> %a) {
ret half %res
}
+; FMAXV
+
+define double @fmaximumv_nxv8f64(<vscale x 8 x double> %a) {
+; CHECK-LABEL: fmaximumv_nxv8f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.d
+; CHECK-NEXT: fmax z1.d, p0/m, z1.d, z3.d
+; CHECK-NEXT: fmax z0.d, p0/m, z0.d, z2.d
+; CHECK-NEXT: fmax z0.d, p0/m, z0.d, z1.d
+; CHECK-NEXT: fmaxv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fmaximum.nxv8f64(<vscale x 8 x double> %a)
+ ret double %res
+}
+
+; FMINV
+
+define half @fminimumv_nxv16f16(<vscale x 16 x half> %a) {
+; CHECK-LABEL: fminimumv_nxv16f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ptrue p0.h
+; CHECK-NEXT: fmin z0.h, p0/m, z0.h, z1.h
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fminimum.nxv16f16(<vscale x 16 x half> %a)
+ ret half %res
+}
+
declare double @llvm.vector.reduce.fadd.nxv8f64(double, <vscale x 8 x double>)
declare float @llvm.vector.reduce.fadd.nxv8f32(float, <vscale x 8 x float>)
declare double @llvm.vector.reduce.fmax.nxv8f64(<vscale x 8 x double>)
-
declare half @llvm.vector.reduce.fmin.nxv16f16(<vscale x 16 x half>)
+declare double @llvm.vector.reduce.fmaximum.nxv8f64(<vscale x 8 x double>)
+declare half @llvm.vector.reduce.fminimum.nxv16f16(<vscale x 16 x half>)
diff --git a/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-fp-reduce.ll b/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-fp-reduce.ll
index dbc5817bd2d9f..77a448c934af9 100644
--- a/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-fp-reduce.ll
+++ b/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-fp-reduce.ll
@@ -244,7 +244,7 @@ define double @faddv_v4f64(double %start, ptr %a) #0 {
}
;
-; FMAXV
+; FMAXNMV
;
define half @fmaxv_v4f16(<4 x half> %a) #0 {
@@ -360,7 +360,7 @@ define double @fmaxv_v4f64(ptr %a) #0 {
}
;
-; FMINV
+; FMINNMV
;
define half @fminv_v4f16(<4 x half> %a) #0 {
@@ -475,6 +475,238 @@ define double @fminv_v4f64(ptr %a) #0 {
ret double %res
}
+;
+; FMAXV
+;
+
+define half @fmaximumv_v4f16(<4 x half> %a) #0 {
+; CHECK-LABEL: fmaximumv_v4f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $z0
+; CHECK-NEXT: ptrue p0.h, vl4
+; CHECK-NEXT: fmaxv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fmaximum.v4f16(<4 x half> %a)
+ ret half %res
+}
+
+define half @fmaximumv_v8f16(<8 x half> %a) #0 {
+; CHECK-LABEL: fmaximumv_v8f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $q0 killed $q0 def $z0
+; CHECK-NEXT: ptrue p0.h, vl8
+; CHECK-NEXT: fmaxv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fmaximum.v8f16(<8 x half> %a)
+ ret half %res
+}
+
+define half @fmaximumv_v16f16(ptr %a) #0 {
+; CHECK-LABEL: fmaximumv_v16f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ldp q1, q0, [x0]
+; CHECK-NEXT: ptrue p0.h, vl8
+; CHECK-NEXT: fmax z0.h, p0/m, z0.h, z1.h
+; CHECK-NEXT: fmaxv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <16 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fmaximum.v16f16(<16 x half> %op)
+ ret half %res
+}
+
+define float @fmaximumv_v2f32(<2 x float> %a) #0 {
+; CHECK-LABEL: fmaximumv_v2f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $z0
+; CHECK-NEXT: ptrue p0.s, vl2
+; CHECK-NEXT: fmaxv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fmaximum.v2f32(<2 x float> %a)
+ ret float %res
+}
+
+define float @fmaximumv_v4f32(<4 x float> %a) #0 {
+; CHECK-LABEL: fmaximumv_v4f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $q0 killed $q0 def $z0
+; CHECK-NEXT: ptrue p0.s, vl4
+; CHECK-NEXT: fmaxv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fmaximum.v4f32(<4 x float> %a)
+ ret float %res
+}
+
+define float @fmaximumv_v8f32(ptr %a) #0 {
+; CHECK-LABEL: fmaximumv_v8f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ldp q1, q0, [x0]
+; CHECK-NEXT: ptrue p0.s, vl4
+; CHECK-NEXT: fmax z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT: fmaxv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <8 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fmaximum.v8f32(<8 x float> %op)
+ ret float %res
+}
+
+define double @fmaximumv_v1f64(<1 x double> %a) #0 {
+; CHECK-LABEL: fmaximumv_v1f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $z0
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fmaximum.v1f64(<1 x double> %a)
+ ret double %res
+}
+
+define double @fmaximumv_v2f64(<2 x double> %a) #0 {
+; CHECK-LABEL: fmaximumv_v2f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $q0 killed $q0 def $z0
+; CHECK-NEXT: ptrue p0.d, vl2
+; CHECK-NEXT: fmaxv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fmaximum.v2f64(<2 x double> %a)
+ ret double %res
+}
+
+define double @fmaximumv_v4f64(ptr %a) #0 {
+; CHECK-LABEL: fmaximumv_v4f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ldp q1, q0, [x0]
+; CHECK-NEXT: ptrue p0.d, vl2
+; CHECK-NEXT: fmax z0.d, p0/m, z0.d, z1.d
+; CHECK-NEXT: fmaxv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <4 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fmaximum.v4f64(<4 x double> %op)
+ ret double %res
+}
+
+;
+; FMINV
+;
+
+define half @fminimumv_v4f16(<4 x half> %a) #0 {
+; CHECK-LABEL: fminimumv_v4f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $z0
+; CHECK-NEXT: ptrue p0.h, vl4
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fminimum.v4f16(<4 x half> %a)
+ ret half %res
+}
+
+define half @fminimumv_v8f16(<8 x half> %a) #0 {
+; CHECK-LABEL: fminimumv_v8f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $q0 killed $q0 def $z0
+; CHECK-NEXT: ptrue p0.h, vl8
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %res = call half @llvm.vector.reduce.fminimum.v8f16(<8 x half> %a)
+ ret half %res
+}
+
+define half @fminimumv_v16f16(ptr %a) #0 {
+; CHECK-LABEL: fminimumv_v16f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ldp q1, q0, [x0]
+; CHECK-NEXT: ptrue p0.h, vl8
+; CHECK-NEXT: fmin z0.h, p0/m, z0.h, z1.h
+; CHECK-NEXT: fminv h0, p0, z0.h
+; CHECK-NEXT: // kill: def $h0 killed $h0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <16 x half>, ptr %a
+ %res = call half @llvm.vector.reduce.fminimum.v16f16(<16 x half> %op)
+ ret half %res
+}
+
+define float @fminimumv_v2f32(<2 x float> %a) #0 {
+; CHECK-LABEL: fminimumv_v2f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $z0
+; CHECK-NEXT: ptrue p0.s, vl2
+; CHECK-NEXT: fminv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fminimum.v2f32(<2 x float> %a)
+ ret float %res
+}
+
+define float @fminimumv_v4f32(<4 x float> %a) #0 {
+; CHECK-LABEL: fminimumv_v4f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $q0 killed $q0 def $z0
+; CHECK-NEXT: ptrue p0.s, vl4
+; CHECK-NEXT: fminv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %res = call float @llvm.vector.reduce.fminimum.v4f32(<4 x float> %a)
+ ret float %res
+}
+
+define float @fminimumv_v8f32(ptr %a) #0 {
+; CHECK-LABEL: fminimumv_v8f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ldp q1, q0, [x0]
+; CHECK-NEXT: ptrue p0.s, vl4
+; CHECK-NEXT: fmin z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT: fminv s0, p0, z0.s
+; CHECK-NEXT: // kill: def $s0 killed $s0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <8 x float>, ptr %a
+ %res = call float @llvm.vector.reduce.fminimum.v8f32(<8 x float> %op)
+ ret float %res
+}
+
+define double @fminimumv_v1f64(<1 x double> %a) #0 {
+; CHECK-LABEL: fminimumv_v1f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $d0 killed $d0 def $z0
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fminimum.v1f64(<1 x double> %a)
+ ret double %res
+}
+
+define double @fminimumv_v2f64(<2 x double> %a) #0 {
+; CHECK-LABEL: fminimumv_v2f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: // kill: def $q0 killed $q0 def $z0
+; CHECK-NEXT: ptrue p0.d, vl2
+; CHECK-NEXT: fminv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %res = call double @llvm.vector.reduce.fminimum.v2f64(<2 x double> %a)
+ ret double %res
+}
+
+define double @fminimumv_v4f64(ptr %a) #0 {
+; CHECK-LABEL: fminimumv_v4f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: ldp q1, q0, [x0]
+; CHECK-NEXT: ptrue p0.d, vl2
+; CHECK-NEXT: fmin z0.d, p0/m, z0.d, z1.d
+; CHECK-NEXT: fminv d0, p0, z0.d
+; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-NEXT: ret
+ %op = load <4 x double>, ptr %a
+ %res = call double @llvm.vector.reduce.fminimum.v4f64(<4 x double> %op)
+ ret double %res
+}
+
attributes #0 = { "target-features"="+sve" }
declare half @llvm.vector.reduce.fadd.v4f16(half, <4 x half>)
@@ -512,3 +744,27 @@ declare float @llvm.vector.reduce.fmin.v8f32(<8 x float>)
declare double @llvm.vector.reduce.fmin.v1f64(<1 x double>)
declare double @llvm.vector.reduce.fmin.v2f64(<2 x double>)
declare double @llvm.vector.reduce.fmin.v4f64(<4 x double>)
+
+declare half @llvm.vector.reduce.fmaximum.v4f16(<4 x half>)
+declare half @llvm.vector.reduce.fmaximum.v8f16(<8 x half>)
+declare half @llvm.vector.reduce.fmaximum.v16f16(<16 x half>)
+
+declare float @llvm.vector.reduce.fmaximum.v2f32(<2 x float>)
+declare float @llvm.vector.reduce.fmaximum.v4f32(<4 x float>)
+declare float @llvm.vector.reduce.fmaximum.v8f32(<8 x float>)
+
+declare double @llvm.vector.reduce.fmaximum.v1f64(<1 x double>)
+declare double @llvm.vector.reduce.fmaximum.v2f64(<2 x double>)
+declare double @llvm.vector.reduce.fmaximum.v4f64(<4 x double>)
+
+declare half @llvm.vector.reduce.fminimum.v4f16(<4 x half>)
+declare half @llvm.vector.reduce.fminimum.v8f16(<8 x half>)
+declare half @llvm.vector.reduce.fminimum.v16f16(<16 x half>)
+
+declare float @llvm.vector.reduce.fminimum.v2f32(<2 x float>)
+declare float @llvm.vector.reduce.fminimum.v4f32(<4 x float>)
+declare float @llvm.vector.reduce.fminimum.v8f32(<8 x float>)
+
+declare double @llvm.vector.reduce.fminimum.v1f64(<1 x double>)
+declare double @llvm.vector.reduce.fminimum.v2f64(<2 x double>)
+declare double @llvm.vector.reduce.fminimum.v4f64(<4 x double>)
More information about the llvm-commits
mailing list