AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

added packed math fp16 native device functions

Browse Source 
1. Added SDWA implementation inside IR file
2. Added device functions to header + used them in test

Change-Id: Ib4e059a58eee201cc82438689e3e9bc5f9d26653
This commit is contained in:
Aditya Atluri
2017-01-12 14:10:51 -06:00
parent c286bf6f8a
commit eeef055469
3 changed files with 106 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
hipamd/src/hip_ir.ll
+23
View File
@@ -61,6 +61,29 @@ define half @__hip_hc_ir_hfma_half(half %a, half %b, half %c) #1 {
ret half %9
}
define i32 @__hip_hc_ir_hadd2_int(i32 %a, i32 %b) #1 {
%1 = tail call i32 asm sideeffect "v_add_f16 $0, $1, $2","=v,v,v"(i32 %a, i32 %b)
tail call void asm sideeffect "v_add_f16_sdwa $0, $1, $2 dst_sel:WORD_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 src1_sel:WORD_1","v,v,v"(i32 %1, i32 %a, i32 %b)
ret i32 %1
}
define i32 @__hip_hc_ir_hfma2_int(i32 %a, i32 %b, i32 %c) #1 {
%1 = tail call i32 asm sideeffect "v_mad_f16 $0, $1, $2, $3","=v,v,v,v"(i32 %a, i32 %b, i32 %c)
tail call void asm sideeffect "v_mul_f16_sdwa $0, $1, $2 dst_sel:WORD_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 src1_sel:WORD_1","v,v,v"(i32 %1, i32 %a, i32 %b)
tail call void asm sideeffect "v_add_f16_sdwa $0, $1, $2 dst_sel:WORD_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 src1_sel:WORD_1","v,v,v"(i32 %1, i32 %1, i32 %c)
ret i32 %1
}
define i32 @__hip_hc_ir_hmul2_int(i32 %a, i32 %b) #1 {
%1 = tail call i32 asm sideeffect "v_mul_f16 $0, $1, $2","=v,v,v"(i32 %a, i32 %b)
tail call void asm sideeffect "v_mul_f16_sdwa $0, $1, $2 dst_sel:WORD_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 src1_sel:WORD_1","v,v,v"(i32 %1, i32 %a, i32 %b)
ret i32 %1
}
define i32 @__hip_hc_ir_hsub2_int(i32 %a, i32 %b) #1 {
%1 = tail call i32 asm sideeffect "v_sub_f16 $0, $1, $2","=v,v,v"(i32 %a, i32 %b)
tail call void asm sideeffect "v_sub_f16_sdwa $0, $1, $2 dst_sel:WORD_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 src1_sel:WORD_1","v,v,v"(i32 %1, i32 %a, i32 %b)
ret i32 %1
}
attributes #1 = { alwaysinline nounwind }