[PATCH] D52018: [AMDGPU] Add instruction selection for i1 to f16 conversion
Carl Ritson via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Thu Sep 13 01:05:05 PDT 2018
critson created this revision.
critson added a reviewer: arsenm.
Herald added subscribers: llvm-commits, t-tye, tpr, dstuttard, yaxunl, nhaehnle, wdng, jvesely, kzhuravl.
This is required for GPUs with 16 bit instructions where f16 is a
legal register type and hence uint_to_fp i1 to f16 is not lowered
by legalizing.
Change-Id: Ie4c0fd6ced7cf10ad612023c6879724d9ded5851
Repository:
rL LLVM
https://reviews.llvm.org/D52018
Files:
lib/Target/AMDGPU/SIInstructions.td
test/CodeGen/AMDGPU/uitofp.f16.ll
Index: test/CodeGen/AMDGPU/uitofp.f16.ll
===================================================================
--- test/CodeGen/AMDGPU/uitofp.f16.ll
+++ test/CodeGen/AMDGPU/uitofp.f16.ll
@@ -92,4 +92,23 @@
ret void
}
+; FUNC-LABEL: {{^}}s_uint_to_fp_i1_to_f16:
+; GCN-DAG: v_cmp_le_f32_e32 [[CMP0:vcc]], 1.0, {{v[0-9]+}}
+; GCN-DAG: v_cmp_le_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]], 0, {{v[0-9]+}}
+; GCN: s_xor_b64 [[R_CMP:s\[[0-9]+:[0-9]+\]]], [[CMP1]], [[CMP0]]
+; GCN: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1.0, [[R_CMP]]
+; GCN-NEXT: v_cvt_f16_f32_e32 [[R_F16:v[0-9]+]], [[RESULT]]
+; GCN: buffer_store_short
+; GCN: s_endpgm
+define amdgpu_kernel void @s_uint_to_fp_i1_to_f16(half addrspace(1)* %out, float addrspace(1)* %in0, float addrspace(1)* %in1) {
+ %a = load float, float addrspace(1) * %in0
+ %b = load float, float addrspace(1) * %in1
+ %acmp = fcmp oge float %a, 0.000000e+00
+ %bcmp = fcmp oge float %b, 1.000000e+00
+ %result = xor i1 %acmp, %bcmp
+ %fp = uitofp i1 %result to half
+ store half %fp, half addrspace(1)* %out
+ ret void
+}
+
; f16 = uitofp i64 is in uint_to_fp.i64.ll
Index: lib/Target/AMDGPU/SIInstructions.td
===================================================================
--- lib/Target/AMDGPU/SIInstructions.td
+++ lib/Target/AMDGPU/SIInstructions.td
@@ -1321,6 +1321,11 @@
>;
def : GCNPat <
+ (f16 (uint_to_fp i1:$src)),
+ (V_CVT_F16_F32_e32 (V_CNDMASK_B32_e64 (i32 0), (i32 CONST.FP32_ONE), $src))
+>;
+
+def : GCNPat <
(f32 (sint_to_fp i1:$src)),
(V_CNDMASK_B32_e64 (i32 0), (i32 CONST.FP32_NEG_ONE), $src)
>;
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D52018.165099.patch
Type: text/x-patch
Size: 1580 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180913/10d34bdb/attachment.bin>
More information about the llvm-commits
mailing list