[PATCH] D65074: AMDGPU: Don't assert on v4f16 arguments to shader calling conventions
Matt Arsenault via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Jul 22 05:46:21 PDT 2019
arsenm created this revision.
arsenm added a reviewer: nhaehnle.
Herald added subscribers: t-tye, tpr, dstuttard, yaxunl, wdng, jvesely, kzhuravl.
https://reviews.llvm.org/D65074
Files:
lib/Target/AMDGPU/SIISelLowering.cpp
test/CodeGen/AMDGPU/calling-conventions.ll
Index: test/CodeGen/AMDGPU/calling-conventions.ll
===================================================================
--- test/CodeGen/AMDGPU/calling-conventions.ll
+++ test/CodeGen/AMDGPU/calling-conventions.ll
@@ -199,6 +199,37 @@
ret void
}
+; FIXME: Differenet ABI for VI+
+; GCN-LABEL: {{^}}ps_mesa_v4f16:
+; SI: v_cvt_f16_f32_e32 v3, v3
+; SI: v_cvt_f16_f32_e32 v2, v2
+; SI: v_cvt_f16_f32_e32 v1, v1
+; SI: v_cvt_f16_f32_e32 v0, v0
+
+; VI: v_add_f16_e32 v2, 1.0, v1
+; VI: v_add_f16_sdwa v1, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; VI: v_add_f16_e32 v4, 1.0, v0
+; VI: v_add_f16_sdwa v0, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+define amdgpu_ps <4 x half> @ps_mesa_v4f16(<4 x half> %arg0) {
+ %add = fadd <4 x half> %arg0, <half 1.0, half 1.0, half 1.0, half 1.0>
+ ret <4 x half> %add
+}
+
+; GCN-LABEL: {{^}}ps_mesa_inreg_v4f16:
+; SI: v_cvt_f16_f32_e32 v{{[0-9]+}}, s3
+; SI: v_cvt_f16_f32_e32 v{{[0-9]+}}, s2
+; SI: v_cvt_f16_f32_e32 v{{[0-9]+}}, s1
+; SI: v_cvt_f16_f32_e32 v{{[0-9]+}}, s0
+
+; VI: v_add_f16_e64
+; VI: v_add_f16_sdwa
+; VI: v_add_f16_e64
+; VI: v_add_f16_sdwa
+define amdgpu_ps <4 x half> @ps_mesa_inreg_v4f16(<4 x half> inreg %arg0) {
+ %add = fadd <4 x half> %arg0, <half 1.0, half 1.0, half 1.0, half 1.0>
+ ret <4 x half> %add
+}
+
; GCN-LABEL: {{^}}ps_mesa_inreg_v3i32:
; GCN-DAG: s_add_i32 s0, s0, 1
; GCN-DAG: s_add_i32 s{{[0-9]*}}, s1, 2
Index: lib/Target/AMDGPU/SIISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/SIISelLowering.cpp
+++ lib/Target/AMDGPU/SIISelLowering.cpp
@@ -1579,7 +1579,8 @@
// entire split argument.
if (Arg->Flags.isSplit()) {
while (!Arg->Flags.isSplitEnd()) {
- assert(!Arg->VT.isVector() &&
+ assert((!Arg->VT.isVector() ||
+ Arg->VT.getScalarSizeInBits() == 16) &&
"unexpected vector split in ps argument type");
if (!SkipArg)
Splits.push_back(*Arg);
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D65074.211057.patch
Type: text/x-patch
Size: 2062 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190722/ff6d2e55/attachment.bin>
More information about the llvm-commits
mailing list