100 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 42; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck --check-prefixes=GFX1250,GFX1250-SDAG %s3; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck --check-prefixes=GFX1250,GFX1250-GISEL %s4 5define <2 x half> @fmaximum3_v2f16(<2 x half> %a, <2 x half> %b, <2 x half> %c) {6; GFX1250-SDAG-LABEL: fmaximum3_v2f16:7; GFX1250-SDAG: ; %bb.0: ; %entry8; GFX1250-SDAG-NEXT: s_wait_loadcnt_dscnt 0x09; GFX1250-SDAG-NEXT: s_wait_kmcnt 0x010; GFX1250-SDAG-NEXT: v_pk_maximum3_f16 v0, v2, v0, v111; GFX1250-SDAG-NEXT: s_set_pc_i64 s[30:31]12;13; GFX1250-GISEL-LABEL: fmaximum3_v2f16:14; GFX1250-GISEL: ; %bb.0: ; %entry15; GFX1250-GISEL-NEXT: s_wait_loadcnt_dscnt 0x016; GFX1250-GISEL-NEXT: s_wait_kmcnt 0x017; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v0, v0, v1, v118; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)19; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v0, v2, v0, v020; GFX1250-GISEL-NEXT: s_set_pc_i64 s[30:31]21entry:22 %min = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> %b)23 %res = call <2 x half> @llvm.maximum.v2f16(<2 x half> %c, <2 x half> %min)24 ret <2 x half> %res25}26 27define <2 x half> @fmaximum3_v2f16_vss(<2 x half> %a, <2 x half> inreg %b, <2 x half> inreg %c) {28; GFX1250-SDAG-LABEL: fmaximum3_v2f16_vss:29; GFX1250-SDAG: ; %bb.0: ; %entry30; GFX1250-SDAG-NEXT: s_wait_loadcnt_dscnt 0x031; GFX1250-SDAG-NEXT: s_wait_kmcnt 0x032; GFX1250-SDAG-NEXT: v_pk_maximum3_f16 v0, s1, v0, s033; GFX1250-SDAG-NEXT: s_set_pc_i64 s[30:31]34;35; GFX1250-GISEL-LABEL: fmaximum3_v2f16_vss:36; GFX1250-GISEL: ; %bb.0: ; %entry37; GFX1250-GISEL-NEXT: s_wait_loadcnt_dscnt 0x038; GFX1250-GISEL-NEXT: s_wait_kmcnt 0x039; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v0, v0, s0, s040; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)41; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v0, s1, v0, v042; GFX1250-GISEL-NEXT: s_set_pc_i64 s[30:31]43entry:44 %min = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> %b)45 %res = call <2 x half> @llvm.maximum.v2f16(<2 x half> %c, <2 x half> %min)46 ret <2 x half> %res47}48 49define <3 x half> @fmaximum3_v3f16(<3 x half> %a, <3 x half> %b, <3 x half> %c) {50; GFX1250-SDAG-LABEL: fmaximum3_v3f16:51; GFX1250-SDAG: ; %bb.0: ; %entry52; GFX1250-SDAG-NEXT: s_wait_loadcnt_dscnt 0x053; GFX1250-SDAG-NEXT: s_wait_kmcnt 0x054; GFX1250-SDAG-NEXT: v_pk_maximum3_f16 v0, v4, v0, v255; GFX1250-SDAG-NEXT: v_pk_maximum3_f16 v1, v5, v1, v356; GFX1250-SDAG-NEXT: s_set_pc_i64 s[30:31]57;58; GFX1250-GISEL-LABEL: fmaximum3_v3f16:59; GFX1250-GISEL: ; %bb.0: ; %entry60; GFX1250-GISEL-NEXT: s_wait_loadcnt_dscnt 0x061; GFX1250-GISEL-NEXT: s_wait_kmcnt 0x062; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v0, v0, v2, v263; GFX1250-GISEL-NEXT: v_maximum_f16 v1, v1, v364; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)65; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v0, v4, v0, v066; GFX1250-GISEL-NEXT: v_maximum_f16 v1, v5, v167; GFX1250-GISEL-NEXT: s_set_pc_i64 s[30:31]68entry:69 %min = call <3 x half> @llvm.maximum.v3f16(<3 x half> %a, <3 x half> %b)70 %res = call <3 x half> @llvm.maximum.v3f16(<3 x half> %c, <3 x half> %min)71 ret <3 x half> %res72}73 74define <4 x half> @fmaximum3_v4f16(<4 x half> %a, <4 x half> %b, <4 x half> %c) {75; GFX1250-SDAG-LABEL: fmaximum3_v4f16:76; GFX1250-SDAG: ; %bb.0: ; %entry77; GFX1250-SDAG-NEXT: s_wait_loadcnt_dscnt 0x078; GFX1250-SDAG-NEXT: s_wait_kmcnt 0x079; GFX1250-SDAG-NEXT: v_pk_maximum3_f16 v0, v4, v0, v280; GFX1250-SDAG-NEXT: v_pk_maximum3_f16 v1, v5, v1, v381; GFX1250-SDAG-NEXT: s_set_pc_i64 s[30:31]82;83; GFX1250-GISEL-LABEL: fmaximum3_v4f16:84; GFX1250-GISEL: ; %bb.0: ; %entry85; GFX1250-GISEL-NEXT: s_wait_loadcnt_dscnt 0x086; GFX1250-GISEL-NEXT: s_wait_kmcnt 0x087; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v0, v0, v2, v288; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v1, v1, v3, v389; GFX1250-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)90; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v0, v4, v0, v091; GFX1250-GISEL-NEXT: v_pk_maximum3_f16 v1, v5, v1, v192; GFX1250-GISEL-NEXT: s_set_pc_i64 s[30:31]93entry:94 %min = call <4 x half> @llvm.maximum.v4f16(<4 x half> %a, <4 x half> %b)95 %res = call <4 x half> @llvm.maximum.v4f16(<4 x half> %c, <4 x half> %min)96 ret <4 x half> %res97}98;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:99; GFX1250: {{.*}}100