AMDGPU: Remove dead arguments in test and add SGPR variants

Also cleanup to avoid the memory noise by using return values
in the trivial cases.

Author: arsenm

llvm/test/CodeGen/AMDGPU/llvm.fptrunc.round.ll

@@ -3,42 +3,38 @@
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck %s
 ; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx1030 -verify-machineinstrs < %s | FileCheck %s
 
-define amdgpu_gs void @test_fptrunc_round_upward(float %a, i32 %data0, <4 x i32> %data1, ptr addrspace(1) %out) {
-; CHECK-LABEL: test_fptrunc_round_upward:
+define amdgpu_gs half @v_fptrunc_round_f32_to_f16_upward(float %a) {
+; CHECK-LABEL: v_fptrunc_round_f32_to_f16_upward:
 ; CHECK:       ; %bb.0:
 ; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 1), 1
 ; CHECK-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; CHECK-NEXT:    global_store_short v[6:7], v0, off
-; CHECK-NEXT:    s_endpgm
+; CHECK-NEXT:    ; return to shader part epilog
   %res = call half @llvm.fptrunc.round.f16.f32(float %a, metadata !"round.upward")
-  store half %res, ptr addrspace(1) %out, align 4
-  ret void
+  ret half %res
 }
 
-define amdgpu_gs void @test_fptrunc_round_downward(float %a, i32 %data0, <4 x i32> %data1, ptr addrspace(1) %out) {
-; CHECK-LABEL: test_fptrunc_round_downward:
+define amdgpu_gs half @v_fptrunc_round_f32_to_f16_downward(float %a) {
+; CHECK-LABEL: v_fptrunc_round_f32_to_f16_downward:
 ; CHECK:       ; %bb.0:
 ; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 3, 1), 1
 ; CHECK-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; CHECK-NEXT:    global_store_short v[6:7], v0, off
-; CHECK-NEXT:    s_endpgm
+; CHECK-NEXT:    ; return to shader part epilog
   %res = call half @llvm.fptrunc.round.f16.f32(float %a, metadata !"round.downward")
-  store half %res, ptr addrspace(1) %out, align 4
-  ret void
+  ret half %res
 }
 
-define amdgpu_gs void @test_fptrunc_round_upward_multiple_calls(float %a, float %b, i32 %data0, <4 x i32> %data1, ptr addrspace(1) %out) {
-; CHECK-LABEL: test_fptrunc_round_upward_multiple_calls:
+define amdgpu_gs void @v_fptrunc_round_f32_to_f16_upward_multiple_calls(float %a, float %b, ptr addrspace(1) %out) {
+; CHECK-LABEL: v_fptrunc_round_f32_to_f16_upward_multiple_calls:
 ; CHECK:       ; %bb.0:
 ; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 1), 1
 ; CHECK-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; CHECK-NEXT:    v_cvt_f16_f32_e32 v2, v1
+; CHECK-NEXT:    v_cvt_f16_f32_e32 v4, v1
 ; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 2
 ; CHECK-NEXT:    v_cvt_f16_f32_e32 v1, v1
 ; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 3, 1), 0
-; CHECK-NEXT:    v_add_f16_e32 v0, v0, v2
+; CHECK-NEXT:    v_add_f16_e32 v0, v0, v4
 ; CHECK-NEXT:    v_add_f16_e32 v0, v1, v0
-; CHECK-NEXT:    global_store_short v[7:8], v0, off
+; CHECK-NEXT:    global_store_short v[2:3], v0, off
 ; CHECK-NEXT:    s_endpgm
   %res1 = call half @llvm.fptrunc.round.f16.f32(float %a, metadata !"round.upward")
   %res2 = call half @llvm.fptrunc.round.f16.f32(float %b, metadata !"round.upward")
@@ -49,4 +45,56 @@ define amdgpu_gs void @test_fptrunc_round_upward_multiple_calls(float %a, float
   ret void
 }
 
-declare half @llvm.fptrunc.round.f16.f32(float, metadata)
+define amdgpu_gs i32 @s_fptrunc_round_f32_to_f16_upward(float inreg %a, ptr addrspace(1) %out) {
+; CHECK-LABEL: s_fptrunc_round_f32_to_f16_upward:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    v_mov_b32_e32 v0, s0
+; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 1), 1
+; CHECK-NEXT:    v_cvt_f16_f32_e32 v0, v0
+; CHECK-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; CHECK-NEXT:    v_readfirstlane_b32 s0, v0
+; CHECK-NEXT:    ; return to shader part epilog
+  %res = call half @llvm.fptrunc.round.f16.f32(float %a, metadata !"round.upward")
+  %bitcast = bitcast half %res to i16
+  %ret = zext i16 %bitcast to i32
+  ret i32 %ret
+}
+
+define amdgpu_gs i32 @s_fptrunc_round_f32_to_f16_downward(float inreg %a, ptr addrspace(1) %out) {
+; CHECK-LABEL: s_fptrunc_round_f32_to_f16_downward:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    v_mov_b32_e32 v0, s0
+; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 3, 1), 1
+; CHECK-NEXT:    v_cvt_f16_f32_e32 v0, v0
+; CHECK-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; CHECK-NEXT:    v_readfirstlane_b32 s0, v0
+; CHECK-NEXT:    ; return to shader part epilog
+  %res = call half @llvm.fptrunc.round.f16.f32(float %a, metadata !"round.downward")
+  %bitcast = bitcast half %res to i16
+  %ret = zext i16 %bitcast to i32
+  ret i32 %ret
+}
+
+define amdgpu_gs void @s_fptrunc_round_f32_to_f16_upward_multiple_calls(float inreg %a, float inreg %b, ptr addrspace(1) %out) {
+; CHECK-LABEL: s_fptrunc_round_f32_to_f16_upward_multiple_calls:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    v_mov_b32_e32 v2, s0
+; CHECK-NEXT:    v_mov_b32_e32 v3, s1
+; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 1), 1
+; CHECK-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; CHECK-NEXT:    v_cvt_f16_f32_e32 v4, v3
+; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 2
+; CHECK-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 3, 1), 0
+; CHECK-NEXT:    v_add_f16_e32 v2, v2, v4
+; CHECK-NEXT:    v_add_f16_e32 v2, v3, v2
+; CHECK-NEXT:    global_store_short v[0:1], v2, off
+; CHECK-NEXT:    s_endpgm
+  %res1 = call half @llvm.fptrunc.round.f16.f32(float %a, metadata !"round.upward")
+  %res2 = call half @llvm.fptrunc.round.f16.f32(float %b, metadata !"round.upward")
+  %res3 = call half @llvm.fptrunc.round.f16.f32(float %b, metadata !"round.downward")
+  %res4 = fadd half %res1, %res2
+  %res5 = fadd half %res3, %res4
+  store half %res5, ptr addrspace(1) %out, align 4
+  ret void
+}