mirror of
https://github.com/RPCSX/llvm.git
synced 2025-01-10 06:00:30 +00:00
6d955a1d6d
PeepholeOptimizer cleans up redundant copies, which makes the operand folding more effective. shader-db stats: Totals: SGPRS: 34200 -> 34336 (0.40 %) VGPRS: 22118 -> 21655 (-2.09 %) Code Size: 632144 -> 633460 (0.21 %) bytes LDS: 11 -> 11 (0.00 %) blocks Scratch: 10240 -> 11264 (10.00 %) bytes per wave Max Waves: 8822 -> 8918 (1.09 %) Wait states: 0 -> 0 (0.00 %) Totals from affected shaders: SGPRS: 7704 -> 7840 (1.77 %) VGPRS: 5169 -> 4706 (-8.96 %) Code Size: 234444 -> 235760 (0.56 %) bytes LDS: 2 -> 2 (0.00 %) blocks Scratch: 0 -> 1024 (0.00 %) bytes per wave Max Waves: 1188 -> 1284 (8.08 %) Wait states: 0 -> 0 (0.00 %) Increases: SGPRS: 35 (0.01 %) VGPRS: 1 (0.00 %) Code Size: 59 (0.02 %) LDS: 0 (0.00 %) Scratch: 1 (0.00 %) Max Waves: 48 (0.02 %) Wait states: 0 (0.00 %) Decreases: SGPRS: 26 (0.01 %) VGPRS: 54 (0.02 %) Code Size: 68 (0.03 %) LDS: 0 (0.00 %) Scratch: 0 (0.00 %) Max Waves: 4 (0.00 %) Wait states: 0 (0.00 %) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@266378 91177308-0d34-0410-b5e6-96231b3b80d8
99 lines
3.5 KiB
LLVM
99 lines
3.5 KiB
LLVM
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
|
|
|
|
declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
|
|
|
|
; SI-LABEL: {{^}}v_uint_to_fp_i64_to_f64
|
|
; SI: buffer_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
|
|
; SI-DAG: v_cvt_f64_u32_e32 [[HI_CONV:v\[[0-9]+:[0-9]+\]]], v[[HI]]
|
|
; SI-DAG: v_cvt_f64_u32_e32 [[LO_CONV:v\[[0-9]+:[0-9]+\]]], v[[LO]]
|
|
; SI-DAG: v_ldexp_f64 [[LDEXP:v\[[0-9]+:[0-9]+\]]], [[HI_CONV]], 32
|
|
; SI: v_add_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[LDEXP]], [[LO_CONV]]
|
|
; SI: buffer_store_dwordx2 [[RESULT]]
|
|
define void @v_uint_to_fp_i64_to_f64(double addrspace(1)* %out, i64 addrspace(1)* %in) {
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
|
|
%gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
|
|
%val = load i64, i64 addrspace(1)* %gep, align 8
|
|
%result = uitofp i64 %val to double
|
|
store double %result, double addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}s_uint_to_fp_i64_to_f64
|
|
define void @s_uint_to_fp_i64_to_f64(double addrspace(1)* %out, i64 %in) {
|
|
%cast = uitofp i64 %in to double
|
|
store double %cast, double addrspace(1)* %out, align 8
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}s_uint_to_fp_v2i64_to_v2f64
|
|
define void @s_uint_to_fp_v2i64_to_v2f64(<2 x double> addrspace(1)* %out, <2 x i64> %in) {
|
|
%cast = uitofp <2 x i64> %in to <2 x double>
|
|
store <2 x double> %cast, <2 x double> addrspace(1)* %out, align 16
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}s_uint_to_fp_v4i64_to_v4f64
|
|
define void @s_uint_to_fp_v4i64_to_v4f64(<4 x double> addrspace(1)* %out, <4 x i64> %in) {
|
|
%cast = uitofp <4 x i64> %in to <4 x double>
|
|
store <4 x double> %cast, <4 x double> addrspace(1)* %out, align 16
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}s_uint_to_fp_i32_to_f64
|
|
; SI: v_cvt_f64_u32_e32
|
|
; SI: s_endpgm
|
|
define void @s_uint_to_fp_i32_to_f64(double addrspace(1)* %out, i32 %in) {
|
|
%cast = uitofp i32 %in to double
|
|
store double %cast, double addrspace(1)* %out, align 8
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}s_uint_to_fp_v2i32_to_v2f64
|
|
; SI: v_cvt_f64_u32_e32
|
|
; SI: v_cvt_f64_u32_e32
|
|
; SI: s_endpgm
|
|
define void @s_uint_to_fp_v2i32_to_v2f64(<2 x double> addrspace(1)* %out, <2 x i32> %in) {
|
|
%cast = uitofp <2 x i32> %in to <2 x double>
|
|
store <2 x double> %cast, <2 x double> addrspace(1)* %out, align 16
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}s_uint_to_fp_v4i32_to_v4f64
|
|
; SI: v_cvt_f64_u32_e32
|
|
; SI: v_cvt_f64_u32_e32
|
|
; SI: v_cvt_f64_u32_e32
|
|
; SI: v_cvt_f64_u32_e32
|
|
; SI: s_endpgm
|
|
define void @s_uint_to_fp_v4i32_to_v4f64(<4 x double> addrspace(1)* %out, <4 x i32> %in) {
|
|
%cast = uitofp <4 x i32> %in to <4 x double>
|
|
store <4 x double> %cast, <4 x double> addrspace(1)* %out, align 16
|
|
ret void
|
|
}
|
|
|
|
; We can't fold the SGPRs into v_cndmask_b32_e32, because it already
|
|
; uses an SGPR (implicit vcc).
|
|
|
|
; SI-LABEL: {{^}}uint_to_fp_i1_to_f64:
|
|
; SI-DAG: v_cmp_eq_i32_e64 vcc
|
|
; SI-DAG: v_cndmask_b32_e32 v[[SEL:[0-9]+]], 0, v{{[0-9]+}}
|
|
; SI-DAG: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0{{$}}
|
|
; SI: buffer_store_dwordx2 v{{\[}}[[ZERO]]:[[SEL]]{{\]}}
|
|
; SI: s_endpgm
|
|
define void @uint_to_fp_i1_to_f64(double addrspace(1)* %out, i32 %in) {
|
|
%cmp = icmp eq i32 %in, 0
|
|
%fp = uitofp i1 %cmp to double
|
|
store double %fp, double addrspace(1)* %out, align 4
|
|
ret void
|
|
}
|
|
|
|
; SI-LABEL: {{^}}uint_to_fp_i1_to_f64_load:
|
|
; SI: v_cndmask_b32_e64 [[IRESULT:v[0-9]]], 0, 1
|
|
; SI-NEXT: v_cvt_f64_u32_e32 [[RESULT:v\[[0-9]+:[0-9]\]]], [[IRESULT]]
|
|
; SI: buffer_store_dwordx2 [[RESULT]]
|
|
; SI: s_endpgm
|
|
define void @uint_to_fp_i1_to_f64_load(double addrspace(1)* %out, i1 %in) {
|
|
%fp = uitofp i1 %in to double
|
|
store double %fp, double addrspace(1)* %out, align 8
|
|
ret void
|
|
}
|