shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/fpow.ll
Brox Chen 6dbc01e801
[AMDGPU][True16][CodeGen] update GFX11Plus codegen test with true16 flag (#135078) 
This is a NFC patch.

This patch run a bulk update on CodeGen tests that are impacted by the
true16 features. This patch applies:
1. duplicate GFX11plus runlines and apply them with
"+mattr=+real-true16" and "+mattr=-real-true16"
2. update the test with the update script

For some GISEL runlines, the current CodeGen do not fully support the
true16 version. Still update the runlines, but comment out the failing
one, and added a "FIXME-TRUE16" comment to that test for easier
tracking. These test will be fixed in the following patches.

This is in a transition state that we support both
"+real-true16/-real-true16" in our code base. We plan to move to
"+real-true16" as default, and finally remove "-real-true16" mode and
test lines.
2025-04-23 13:06:52 -04:00

1144 lines
46 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=GFX6 %s
; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -check-prefix=GFX8 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx90a < %s | FileCheck -check-prefix=GFX90A %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s
define float @v_pow_f32(float %x, float %y) {
; GFX6-LABEL: v_pow_f32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_log_f32_e32 v0, v0
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_f32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_log_f32_e32 v0, v0
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_f32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_log_f32_e32 v0, v0
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_f32:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_log_f32_e32 v0, v0
; GFX90A-NEXT: v_mul_legacy_f32 v0, v1, v0
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_f32:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_log_f32_e32 v0, v0
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_pow_f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_log_f32_e32 v0, v0
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_exp_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%pow = call float @llvm.pow.f32(float %x, float %y)
ret float %pow
}
define <2 x float> @v_pow_v2f32(<2 x float> %x, <2 x float> %y) {
; GFX6-LABEL: v_pow_v2f32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_log_f32_e32 v0, v0
; GFX6-NEXT: v_log_f32_e32 v1, v1
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, v2, v0
; GFX6-NEXT: v_mul_legacy_f32_e32 v1, v3, v1
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: v_exp_f32_e32 v1, v1
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_v2f32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_log_f32_e32 v0, v0
; GFX8-NEXT: v_log_f32_e32 v1, v1
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, v2, v0
; GFX8-NEXT: v_mul_legacy_f32_e32 v1, v3, v1
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: v_exp_f32_e32 v1, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_v2f32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_log_f32_e32 v0, v0
; GFX9-NEXT: v_log_f32_e32 v1, v1
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, v2, v0
; GFX9-NEXT: v_mul_legacy_f32_e32 v1, v3, v1
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: v_exp_f32_e32 v1, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_v2f32:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_log_f32_e32 v0, v0
; GFX90A-NEXT: v_log_f32_e32 v1, v1
; GFX90A-NEXT: v_mul_legacy_f32 v0, v2, v0
; GFX90A-NEXT: v_mul_legacy_f32 v1, v3, v1
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: v_exp_f32_e32 v1, v1
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_v2f32:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_log_f32_e32 v0, v0
; GFX10-NEXT: v_log_f32_e32 v1, v1
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, v2, v0
; GFX10-NEXT: v_mul_legacy_f32_e32 v1, v3, v1
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: v_exp_f32_e32 v1, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_pow_v2f32:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_log_f32_e32 v0, v0
; GFX11-NEXT: v_log_f32_e32 v1, v1
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_dual_mul_dx9_zero_f32 v0, v2, v0 :: v_dual_mul_dx9_zero_f32 v1, v3, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_exp_f32_e32 v0, v0
; GFX11-NEXT: v_exp_f32_e32 v1, v1
; GFX11-NEXT: s_setpc_b64 s[30:31]
%pow = call <2 x float> @llvm.pow.v2f32(<2 x float> %x, <2 x float> %y)
ret <2 x float> %pow
}
define half @v_pow_f16(half %x, half %y) {
; GFX6-LABEL: v_pow_f16:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX6-NEXT: v_log_f32_e32 v0, v0
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_f16:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX8-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX8-NEXT: v_log_f32_e32 v0, v0
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX9-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX9-NEXT: v_log_f32_e32 v0, v0
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_f16:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX90A-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX90A-NEXT: v_log_f32_e32 v0, v0
; GFX90A-NEXT: v_mul_legacy_f32 v0, v1, v0
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_f16:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX10-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX10-NEXT: v_log_f32_e32 v0, v0
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_pow_f16:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-TRUE16-NEXT: v_exp_f32_e32 v0, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_pow_f16:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-FAKE16-NEXT: v_exp_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%pow = call half @llvm.pow.f16(half %x, half %y)
ret half %pow
}
define <2 x half> @v_pow_v2f16(<2 x half> %x, <2 x half> %y) {
; GFX6-LABEL: v_pow_v2f16:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX6-NEXT: v_cvt_f32_f16_e32 v2, v2
; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3
; GFX6-NEXT: v_log_f32_e32 v0, v0
; GFX6-NEXT: v_log_f32_e32 v1, v1
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, v2, v0
; GFX6-NEXT: v_mul_legacy_f32_e32 v1, v3, v1
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: v_exp_f32_e32 v1, v1
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_v2f16:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX8-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX8-NEXT: v_log_f32_e32 v2, v2
; GFX8-NEXT: v_log_f32_e32 v0, v0
; GFX8-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX8-NEXT: v_exp_f32_e32 v1, v2
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: v_cvt_f16_f32_sdwa v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT: v_or_b32_e32 v0, v0, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX9-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX9-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX9-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX9-NEXT: v_log_f32_e32 v2, v2
; GFX9-NEXT: v_log_f32_e32 v0, v0
; GFX9-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX9-NEXT: v_exp_f32_e32 v1, v2
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_v2f16:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX90A-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX90A-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX90A-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX90A-NEXT: v_log_f32_e32 v2, v2
; GFX90A-NEXT: v_log_f32_e32 v0, v0
; GFX90A-NEXT: v_mul_legacy_f32 v2, v3, v2
; GFX90A-NEXT: v_mul_legacy_f32 v0, v1, v0
; GFX90A-NEXT: v_exp_f32_e32 v1, v2
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX90A-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX90A-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_v2f16:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX10-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX10-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX10-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX10-NEXT: v_log_f32_e32 v2, v2
; GFX10-NEXT: v_log_f32_e32 v0, v0
; GFX10-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX10-NEXT: v_exp_f32_e32 v1, v2
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX10-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_pow_v2f16:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v2, v0.h
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v1.h
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-TRUE16-NEXT: v_log_f32_e32 v2, v2
; GFX11-TRUE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_mul_dx9_zero_f32_e32 v2, v3, v2
; GFX11-TRUE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_exp_f32_e32 v1, v2
; GFX11-TRUE16-NEXT: v_exp_f32_e32 v2, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v1
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v2
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_pow_v2f16:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v0
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v2, v2
; GFX11-FAKE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v3
; GFX11-FAKE16-NEXT: v_log_f32_e32 v2, v2
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-FAKE16-NEXT: v_mul_dx9_zero_f32_e32 v2, v3, v2
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_exp_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: v_exp_f32_e32 v1, v2
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%pow = call <2 x half> @llvm.pow.v2f16(<2 x half> %x, <2 x half> %y)
ret <2 x half> %pow
}
define <2 x half> @v_pow_v2f16_fneg_lhs(<2 x half> %x, <2 x half> %y) {
; GFX6-LABEL: v_pow_v2f16_fneg_lhs:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; GFX6-NEXT: v_or_b32_e32 v0, v0, v1
; GFX6-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v3
; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX6-NEXT: v_lshrrev_b32_e32 v0, 16, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v2, v2
; GFX6-NEXT: v_log_f32_e32 v3, v3
; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX6-NEXT: v_log_f32_e32 v4, v0
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, v2, v3
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: v_mul_legacy_f32_e32 v1, v1, v4
; GFX6-NEXT: v_exp_f32_e32 v1, v1
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_v2f16_fneg_lhs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_cvt_f32_f16_sdwa v2, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX8-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX8-NEXT: v_log_f32_e32 v2, v2
; GFX8-NEXT: v_log_f32_e32 v0, v0
; GFX8-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX8-NEXT: v_exp_f32_e32 v1, v2
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: v_cvt_f16_f32_sdwa v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT: v_or_b32_e32 v0, v0, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_v2f16_fneg_lhs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cvt_f32_f16_sdwa v2, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX9-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX9-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX9-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX9-NEXT: v_log_f32_e32 v2, v2
; GFX9-NEXT: v_log_f32_e32 v0, v0
; GFX9-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX9-NEXT: v_exp_f32_e32 v1, v2
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_v2f16_fneg_lhs:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_cvt_f32_f16_sdwa v2, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX90A-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX90A-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX90A-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX90A-NEXT: v_log_f32_e32 v2, v2
; GFX90A-NEXT: v_log_f32_e32 v0, v0
; GFX90A-NEXT: v_mul_legacy_f32 v2, v3, v2
; GFX90A-NEXT: v_mul_legacy_f32 v0, v1, v0
; GFX90A-NEXT: v_exp_f32_e32 v1, v2
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX90A-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX90A-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_v2f16_fneg_lhs:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_cvt_f32_f16_sdwa v2, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX10-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX10-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX10-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX10-NEXT: v_log_f32_e32 v2, v2
; GFX10-NEXT: v_log_f32_e32 v0, v0
; GFX10-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX10-NEXT: v_exp_f32_e32 v1, v2
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX10-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_pow_v2f16_fneg_lhs:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e64 v2, -v0.h
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e64 v0, -v0.l
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v1.h
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-TRUE16-NEXT: v_log_f32_e32 v2, v2
; GFX11-TRUE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_mul_dx9_zero_f32_e32 v2, v3, v2
; GFX11-TRUE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_exp_f32_e32 v1, v2
; GFX11-TRUE16-NEXT: v_exp_f32_e32 v2, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v1
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v2
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_pow_v2f16_fneg_lhs:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v0
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e64 v2, -v2
; GFX11-FAKE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v3
; GFX11-FAKE16-NEXT: v_log_f32_e32 v2, v2
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-FAKE16-NEXT: v_mul_dx9_zero_f32_e32 v2, v3, v2
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_exp_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: v_exp_f32_e32 v1, v2
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%x.fneg = fneg <2 x half> %x
%pow = call <2 x half> @llvm.pow.v2f16(<2 x half> %x.fneg, <2 x half> %y)
ret <2 x half> %pow
}
define <2 x half> @v_pow_v2f16_fneg_rhs(<2 x half> %x, <2 x half> %y) {
; GFX6-LABEL: v_pow_v2f16_fneg_rhs:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3
; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX6-NEXT: v_or_b32_e32 v2, v2, v3
; GFX6-NEXT: v_xor_b32_e32 v2, 0x80008000, v2
; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v2
; GFX6-NEXT: v_cvt_f32_f16_e32 v2, v2
; GFX6-NEXT: v_log_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3
; GFX6-NEXT: v_log_f32_e32 v1, v1
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, v2, v0
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: v_mul_legacy_f32_e32 v1, v3, v1
; GFX6-NEXT: v_exp_f32_e32 v1, v1
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_v2f16_fneg_rhs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX8-NEXT: v_cvt_f32_f16_sdwa v3, -v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX8-NEXT: v_log_f32_e32 v2, v2
; GFX8-NEXT: v_log_f32_e32 v0, v0
; GFX8-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX8-NEXT: v_exp_f32_e32 v1, v2
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: v_cvt_f16_f32_sdwa v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT: v_or_b32_e32 v0, v0, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_v2f16_fneg_rhs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX9-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX9-NEXT: v_cvt_f32_f16_sdwa v3, -v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX9-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX9-NEXT: v_log_f32_e32 v2, v2
; GFX9-NEXT: v_log_f32_e32 v0, v0
; GFX9-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX9-NEXT: v_exp_f32_e32 v1, v2
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_v2f16_fneg_rhs:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX90A-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX90A-NEXT: v_cvt_f32_f16_sdwa v3, -v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX90A-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX90A-NEXT: v_log_f32_e32 v2, v2
; GFX90A-NEXT: v_log_f32_e32 v0, v0
; GFX90A-NEXT: v_mul_legacy_f32 v2, v3, v2
; GFX90A-NEXT: v_mul_legacy_f32 v0, v1, v0
; GFX90A-NEXT: v_exp_f32_e32 v1, v2
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX90A-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX90A-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_v2f16_fneg_rhs:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX10-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX10-NEXT: v_cvt_f32_f16_sdwa v3, -v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX10-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX10-NEXT: v_log_f32_e32 v2, v2
; GFX10-NEXT: v_log_f32_e32 v0, v0
; GFX10-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX10-NEXT: v_exp_f32_e32 v1, v2
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX10-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_pow_v2f16_fneg_rhs:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v2, v0.h
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e64 v3, -v1.h
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e64 v1, -v1.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-TRUE16-NEXT: v_log_f32_e32 v2, v2
; GFX11-TRUE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_mul_dx9_zero_f32_e32 v2, v3, v2
; GFX11-TRUE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_exp_f32_e32 v1, v2
; GFX11-TRUE16-NEXT: v_exp_f32_e32 v2, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v1
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v2
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_pow_v2f16_fneg_rhs:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v0
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v2, v2
; GFX11-FAKE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e64 v3, -v3
; GFX11-FAKE16-NEXT: v_log_f32_e32 v2, v2
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-FAKE16-NEXT: v_mul_dx9_zero_f32_e32 v2, v3, v2
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_exp_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: v_exp_f32_e32 v1, v2
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%y.fneg = fneg <2 x half> %y
%pow = call <2 x half> @llvm.pow.v2f16(<2 x half> %x, <2 x half> %y.fneg)
ret <2 x half> %pow
}
define <2 x half> @v_pow_v2f16_fneg_lhs_rhs(<2 x half> %x, <2 x half> %y) {
; GFX6-LABEL: v_pow_v2f16_fneg_lhs_rhs:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; GFX6-NEXT: v_or_b32_e32 v0, v0, v1
; GFX6-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
; GFX6-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3
; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v1
; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0
; GFX6-NEXT: v_or_b32_e32 v2, v2, v3
; GFX6-NEXT: v_xor_b32_e32 v2, 0x80008000, v2
; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v2
; GFX6-NEXT: v_cvt_f32_f16_e32 v2, v2
; GFX6-NEXT: v_log_f32_e32 v0, v0
; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3
; GFX6-NEXT: v_log_f32_e32 v1, v1
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, v2, v0
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: v_mul_legacy_f32_e32 v1, v3, v1
; GFX6-NEXT: v_exp_f32_e32 v1, v1
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_v2f16_fneg_lhs_rhs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_cvt_f32_f16_sdwa v2, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX8-NEXT: v_cvt_f32_f16_sdwa v3, -v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX8-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX8-NEXT: v_log_f32_e32 v2, v2
; GFX8-NEXT: v_log_f32_e32 v0, v0
; GFX8-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX8-NEXT: v_exp_f32_e32 v1, v2
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: v_cvt_f16_f32_sdwa v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT: v_or_b32_e32 v0, v0, v1
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_v2f16_fneg_lhs_rhs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cvt_f32_f16_sdwa v2, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX9-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX9-NEXT: v_cvt_f32_f16_sdwa v3, -v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX9-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX9-NEXT: v_log_f32_e32 v2, v2
; GFX9-NEXT: v_log_f32_e32 v0, v0
; GFX9-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX9-NEXT: v_exp_f32_e32 v1, v2
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_v2f16_fneg_lhs_rhs:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_cvt_f32_f16_sdwa v2, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX90A-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX90A-NEXT: v_cvt_f32_f16_sdwa v3, -v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX90A-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX90A-NEXT: v_log_f32_e32 v2, v2
; GFX90A-NEXT: v_log_f32_e32 v0, v0
; GFX90A-NEXT: v_mul_legacy_f32 v2, v3, v2
; GFX90A-NEXT: v_mul_legacy_f32 v0, v1, v0
; GFX90A-NEXT: v_exp_f32_e32 v1, v2
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX90A-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX90A-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_v2f16_fneg_lhs_rhs:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_cvt_f32_f16_sdwa v2, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX10-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX10-NEXT: v_cvt_f32_f16_sdwa v3, -v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; GFX10-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX10-NEXT: v_log_f32_e32 v2, v2
; GFX10-NEXT: v_log_f32_e32 v0, v0
; GFX10-NEXT: v_mul_legacy_f32_e32 v2, v3, v2
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX10-NEXT: v_exp_f32_e32 v1, v2
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX10-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_pow_v2f16_fneg_lhs_rhs:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e64 v2, -v0.h
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e64 v0, -v0.l
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e64 v3, -v1.h
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e64 v1, -v1.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-TRUE16-NEXT: v_log_f32_e32 v2, v2
; GFX11-TRUE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_mul_dx9_zero_f32_e32 v2, v3, v2
; GFX11-TRUE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_exp_f32_e32 v1, v2
; GFX11-TRUE16-NEXT: v_exp_f32_e32 v2, v0
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v1
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v2
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_pow_v2f16_fneg_lhs_rhs:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v0
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e64 v0, -v0
; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e64 v1, -v1
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e64 v2, -v2
; GFX11-FAKE16-NEXT: v_log_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e64 v3, -v3
; GFX11-FAKE16-NEXT: v_log_f32_e32 v2, v2
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-FAKE16-NEXT: v_mul_dx9_zero_f32_e32 v2, v3, v2
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_exp_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: v_exp_f32_e32 v1, v2
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
%x.fneg = fneg <2 x half> %x
%y.fneg = fneg <2 x half> %y
%pow = call <2 x half> @llvm.pow.v2f16(<2 x half> %x.fneg, <2 x half> %y.fneg)
ret <2 x half> %pow
}
; FIXME
; define double @v_pow_f64(double %x, double %y) {
; %pow = call double @llvm.pow.f64(double %x, double %y)
; ret double %pow
; }
define float @v_pow_f32_fabs_lhs(float %x, float %y) {
; GFX6-LABEL: v_pow_f32_fabs_lhs:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_log_f32_e64 v0, |v0|
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_f32_fabs_lhs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_log_f32_e64 v0, |v0|
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_f32_fabs_lhs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_log_f32_e64 v0, |v0|
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_f32_fabs_lhs:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_log_f32_e64 v0, |v0|
; GFX90A-NEXT: v_mul_legacy_f32 v0, v1, v0
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_f32_fabs_lhs:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_log_f32_e64 v0, |v0|
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, v1, v0
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_pow_f32_fabs_lhs:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_log_f32_e64 v0, |v0|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_mul_dx9_zero_f32_e32 v0, v1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_exp_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x)
%pow = call float @llvm.pow.f32(float %fabs.x, float %y)
ret float %pow
}
define float @v_pow_f32_fabs_rhs(float %x, float %y) {
; GFX6-LABEL: v_pow_f32_fabs_rhs:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_log_f32_e32 v0, v0
; GFX6-NEXT: v_mul_legacy_f32_e64 v0, |v1|, v0
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_f32_fabs_rhs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_log_f32_e32 v0, v0
; GFX8-NEXT: v_mul_legacy_f32_e64 v0, |v1|, v0
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_f32_fabs_rhs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_log_f32_e32 v0, v0
; GFX9-NEXT: v_mul_legacy_f32_e64 v0, |v1|, v0
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_f32_fabs_rhs:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_log_f32_e32 v0, v0
; GFX90A-NEXT: v_mul_legacy_f32 v0, |v1|, v0
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_f32_fabs_rhs:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_log_f32_e32 v0, v0
; GFX10-NEXT: v_mul_legacy_f32_e64 v0, |v1|, v0
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_pow_f32_fabs_rhs:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_log_f32_e32 v0, v0
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_mul_dx9_zero_f32_e64 v0, |v1|, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_exp_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%fabs.y = call float @llvm.fabs.f32(float %y)
%pow = call float @llvm.pow.f32(float %x, float %fabs.y)
ret float %pow
}
define float @v_pow_f32_fabs_lhs_rhs(float %x, float %y) {
; GFX6-LABEL: v_pow_f32_fabs_lhs_rhs:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_log_f32_e64 v0, |v0|
; GFX6-NEXT: v_mul_legacy_f32_e64 v0, |v1|, v0
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_pow_f32_fabs_lhs_rhs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_log_f32_e64 v0, |v0|
; GFX8-NEXT: v_mul_legacy_f32_e64 v0, |v1|, v0
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_pow_f32_fabs_lhs_rhs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_log_f32_e64 v0, |v0|
; GFX9-NEXT: v_mul_legacy_f32_e64 v0, |v1|, v0
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX90A-LABEL: v_pow_f32_fabs_lhs_rhs:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX90A-NEXT: v_log_f32_e64 v0, |v0|
; GFX90A-NEXT: v_mul_legacy_f32 v0, |v1|, v0
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_pow_f32_fabs_lhs_rhs:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_log_f32_e64 v0, |v0|
; GFX10-NEXT: v_mul_legacy_f32_e64 v0, |v1|, v0
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_pow_f32_fabs_lhs_rhs:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_log_f32_e64 v0, |v0|
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_mul_dx9_zero_f32_e64 v0, |v1|, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_exp_f32_e32 v0, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x)
%fabs.y = call float @llvm.fabs.f32(float %y)
%pow = call float @llvm.pow.f32(float %fabs.x, float %fabs.y)
ret float %pow
}
define amdgpu_ps float @v_pow_f32_sgpr_vgpr(float inreg %x, float %y) {
; GFX6-LABEL: v_pow_f32_sgpr_vgpr:
; GFX6: ; %bb.0:
; GFX6-NEXT: v_log_f32_e32 v1, s0
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, v0, v1
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: ; return to shader part epilog
;
; GFX8-LABEL: v_pow_f32_sgpr_vgpr:
; GFX8: ; %bb.0:
; GFX8-NEXT: v_log_f32_e32 v1, s0
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, v0, v1
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: v_pow_f32_sgpr_vgpr:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_log_f32_e32 v1, s0
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, v0, v1
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: ; return to shader part epilog
;
; GFX90A-LABEL: v_pow_f32_sgpr_vgpr:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: v_log_f32_e32 v1, s0
; GFX90A-NEXT: v_mul_legacy_f32 v0, v0, v1
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: v_pow_f32_sgpr_vgpr:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_log_f32_e32 v1, s0
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, v0, v1
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: ; return to shader part epilog
;
; GFX11-LABEL: v_pow_f32_sgpr_vgpr:
; GFX11: ; %bb.0:
; GFX11-NEXT: v_log_f32_e32 v1, s0
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_mul_dx9_zero_f32_e32 v0, v0, v1
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_exp_f32_e32 v0, v0
; GFX11-NEXT: ; return to shader part epilog
%pow = call float @llvm.pow.f32(float %x, float %y)
ret float %pow
}
define amdgpu_ps float @v_pow_f32_vgpr_sgpr(float %x, float inreg %y) {
; GFX6-LABEL: v_pow_f32_vgpr_sgpr:
; GFX6: ; %bb.0:
; GFX6-NEXT: v_log_f32_e32 v0, v0
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, s0, v0
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: ; return to shader part epilog
;
; GFX8-LABEL: v_pow_f32_vgpr_sgpr:
; GFX8: ; %bb.0:
; GFX8-NEXT: v_log_f32_e32 v0, v0
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, s0, v0
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: v_pow_f32_vgpr_sgpr:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_log_f32_e32 v0, v0
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, s0, v0
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: ; return to shader part epilog
;
; GFX90A-LABEL: v_pow_f32_vgpr_sgpr:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: v_log_f32_e32 v0, v0
; GFX90A-NEXT: v_mul_legacy_f32 v0, s0, v0
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: v_pow_f32_vgpr_sgpr:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_log_f32_e32 v0, v0
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, s0, v0
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: ; return to shader part epilog
;
; GFX11-LABEL: v_pow_f32_vgpr_sgpr:
; GFX11: ; %bb.0:
; GFX11-NEXT: v_log_f32_e32 v0, v0
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_mul_dx9_zero_f32_e32 v0, s0, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_exp_f32_e32 v0, v0
; GFX11-NEXT: ; return to shader part epilog
%pow = call float @llvm.pow.f32(float %x, float %y)
ret float %pow
}
define amdgpu_ps float @v_pow_f32_sgpr_sgpr(float inreg %x, float inreg %y) {
; GFX6-LABEL: v_pow_f32_sgpr_sgpr:
; GFX6: ; %bb.0:
; GFX6-NEXT: v_log_f32_e32 v0, s0
; GFX6-NEXT: v_mul_legacy_f32_e32 v0, s1, v0
; GFX6-NEXT: v_exp_f32_e32 v0, v0
; GFX6-NEXT: ; return to shader part epilog
;
; GFX8-LABEL: v_pow_f32_sgpr_sgpr:
; GFX8: ; %bb.0:
; GFX8-NEXT: v_log_f32_e32 v0, s0
; GFX8-NEXT: v_mul_legacy_f32_e32 v0, s1, v0
; GFX8-NEXT: v_exp_f32_e32 v0, v0
; GFX8-NEXT: ; return to shader part epilog
;
; GFX9-LABEL: v_pow_f32_sgpr_sgpr:
; GFX9: ; %bb.0:
; GFX9-NEXT: v_log_f32_e32 v0, s0
; GFX9-NEXT: v_mul_legacy_f32_e32 v0, s1, v0
; GFX9-NEXT: v_exp_f32_e32 v0, v0
; GFX9-NEXT: ; return to shader part epilog
;
; GFX90A-LABEL: v_pow_f32_sgpr_sgpr:
; GFX90A: ; %bb.0:
; GFX90A-NEXT: v_log_f32_e32 v0, s0
; GFX90A-NEXT: v_mul_legacy_f32 v0, s1, v0
; GFX90A-NEXT: v_exp_f32_e32 v0, v0
; GFX90A-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: v_pow_f32_sgpr_sgpr:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_log_f32_e32 v0, s0
; GFX10-NEXT: v_mul_legacy_f32_e32 v0, s1, v0
; GFX10-NEXT: v_exp_f32_e32 v0, v0
; GFX10-NEXT: ; return to shader part epilog
;
; GFX11-LABEL: v_pow_f32_sgpr_sgpr:
; GFX11: ; %bb.0:
; GFX11-NEXT: v_log_f32_e32 v0, s0
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_mul_dx9_zero_f32_e32 v0, s1, v0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_exp_f32_e32 v0, v0
; GFX11-NEXT: ; return to shader part epilog
%pow = call float @llvm.pow.f32(float %x, float %y)
ret float %pow
}
declare half @llvm.pow.f16(half, half)
declare float @llvm.pow.f32(float, float)
declare double @llvm.pow.f64(double, double)
declare half @llvm.fabs.f16(half)
declare float @llvm.fabs.f32(float)
declare <2 x half> @llvm.pow.v2f16(<2 x half>, <2 x half>)
declare <2 x float> @llvm.pow.v2f32(<2 x float>, <2 x float>)
Reference in New Issue View Git Blame Copy Permalink