From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Mon, 27 Feb 2017 19:24:47 +0000 (+0000)
Subject: AMDGPU: Add f16 to shader calling conventions
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=29df731fe54e2e0819f98c09e8f92edb77e5fc57;p=llvm

AMDGPU: Add f16 to shader calling conventions

Mostly useful for writing tests for f16 features.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296370 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/AMDGPU/AMDGPUCallingConv.td b/lib/Target/AMDGPU/AMDGPUCallingConv.td
index 47dfa499206..6b9ebcfe0b7 100644
--- a/lib/Target/AMDGPU/AMDGPUCallingConv.td
+++ b/lib/Target/AMDGPU/AMDGPUCallingConv.td
@@ -17,7 +17,7 @@ class CCIfNotInReg<CCAction A> : CCIf<"!ArgFlags.isInReg()", A> {}
 // Calling convention for SI
 def CC_SI : CallingConv<[
 
-  CCIfInReg<CCIfType<[f32, i32] , CCAssignToReg<[
+  CCIfInReg<CCIfType<[f32, i32, f16] , CCAssignToReg<[
     SGPR0, SGPR1, SGPR2, SGPR3, SGPR4, SGPR5, SGPR6, SGPR7,
     SGPR8, SGPR9, SGPR10, SGPR11, SGPR12, SGPR13, SGPR14, SGPR15,
     SGPR16, SGPR17, SGPR18, SGPR19, SGPR20, SGPR21, SGPR22, SGPR23,
@@ -35,7 +35,7 @@ def CC_SI : CallingConv<[
   >>>,
 
   // 32*4 + 4 is the minimum for a fetch shader consumer with 32 inputs.
-  CCIfNotInReg<CCIfType<[f32, i32] , CCAssignToReg<[
+  CCIfNotInReg<CCIfType<[f32, i32, f16] , CCAssignToReg<[
     VGPR0, VGPR1, VGPR2, VGPR3, VGPR4, VGPR5, VGPR6, VGPR7,
     VGPR8, VGPR9, VGPR10, VGPR11, VGPR12, VGPR13, VGPR14, VGPR15,
     VGPR16, VGPR17, VGPR18, VGPR19, VGPR20, VGPR21, VGPR22, VGPR23,
@@ -76,7 +76,7 @@ def RetCC_SI : CallingConv<[
   ]>>,
 
   // 32*4 + 4 is the minimum for a fetch shader with 32 outputs.
-  CCIfType<[f32] , CCAssignToReg<[
+  CCIfType<[f32, f16] , CCAssignToReg<[
     VGPR0, VGPR1, VGPR2, VGPR3, VGPR4, VGPR5, VGPR6, VGPR7,
     VGPR8, VGPR9, VGPR10, VGPR11, VGPR12, VGPR13, VGPR14, VGPR15,
     VGPR16, VGPR17, VGPR18, VGPR19, VGPR20, VGPR21, VGPR22, VGPR23,
diff --git a/test/CodeGen/AMDGPU/calling-conventions.ll b/test/CodeGen/AMDGPU/calling-conventions.ll
index 57adc8be6a9..55c2b503430 100644
--- a/test/CodeGen/AMDGPU/calling-conventions.ll
+++ b/test/CodeGen/AMDGPU/calling-conventions.ll
@@ -1,9 +1,10 @@
-; RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 
 ; Make sure we don't crash or assert on spir_kernel calling convention.
 
-; SI-LABEL: {{^}}kernel:
-; SI: s_endpgm
+; GCN-LABEL: {{^}}kernel:
+; GCN: s_endpgm
 define spir_kernel void @kernel(i32 addrspace(1)* %out) {
 entry:
   store i32 0, i32 addrspace(1)* %out
@@ -11,10 +12,34 @@ entry:
 }
 
 ; FIXME: This is treated like a kernel
-; SI-LABEL: {{^}}func:
-; SI: s_endpgm
+; GCN-LABEL: {{^}}func:
+; GCN: s_endpgm
 define spir_func void @func(i32 addrspace(1)* %out) {
 entry:
   store i32 0, i32 addrspace(1)* %out
   ret void
 }
+
+; GCN-LABEL: {{^}}ps_ret_cc_f16:
+; SI: v_cvt_f16_f32_e32 v0, v0
+; SI: v_cvt_f32_f16_e32 v0, v0
+; SI: v_add_f32_e32 v0, 1.0, v0
+
+; VI: v_add_f16_e32 v0, 1.0, v0
+; VI: ; return
+define amdgpu_ps half @ps_ret_cc_f16(half %arg0) {
+  %add = fadd half %arg0, 1.0
+  ret half %add
+}
+
+; GCN-LABEL: {{^}}ps_ret_cc_inreg_f16:
+; SI: v_cvt_f16_f32_e32 v0, s0
+; SI: v_cvt_f32_f16_e32 v0, v0
+; SI: v_add_f32_e32 v0, 1.0, v0
+
+; VI: v_add_f16_e64 v0, s0, 1.0
+; VI: ; return
+define amdgpu_ps half @ps_ret_cc_inreg_f16(half inreg %arg0) {
+  %add = fadd half %arg0, 1.0
+  ret half %add
+}