[AMDGPU][True16][CodeGen] true16 codegen pattern for v_pack_b32_f16 #121988
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
broxigarchen
changed the title
broxigarchen
changed the title
broxigarchen
force-pushed
the
main-merge-true16-codegen-V_PACK_B32_F16_
branch
from
f7dc711 to
eb4bd72
Compare
|
@llvm/pr-subscribers-backend-amdgpu Author: Brox Chen (broxigarchen) Changestrue16 codegen pattern for v_pack_b32_f16 Patch is 38.44 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/121988.diff 7 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index e388efe73cddbb..b98c3799332a13 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -3390,6 +3390,9 @@ let SubtargetPredicate = isGFX9Plus in {
let True16Predicate = NotHasTrue16BitInsts in
def : PackB32Pat<V_PACK_B32_F16_e64>;
+let True16Predicate = UseRealTrue16Insts in
+ def : PackB32Pat<V_PACK_B32_F16_t16_e64>;
+
let True16Predicate = UseFakeTrue16Insts in
def : PackB32Pat<V_PACK_B32_F16_fake16_e64>;
} // End SubtargetPredicate = isGFX9Plus
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.ceil.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.ceil.f16.ll
index 84a3a3e88d2383..32d8aa18d97135 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.ceil.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.ceil.f16.ll
@@ -160,14 +160,9 @@ define amdgpu_kernel void @ceil_v2f16(
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-NEXT: v_ceil_f16_e32 v0.l, v0.l
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_ceil_f16_e32 v0.h, v1.l
-; GFX11-NEXT: v_mov_b16_e32 v1.l, v0.l
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT: v_mov_b16_e32 v0.l, v0.h
-; GFX11-NEXT: v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT: v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
; GFX11-NEXT: buffer_store_b32 v0, off, s[4:7], 0
; GFX11-NEXT: s_endpgm
;
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.floor.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.floor.f16.ll
index 9909cfd32b11f7..f6a9fadb33865b 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.floor.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.floor.f16.ll
@@ -161,14 +161,9 @@ define amdgpu_kernel void @floor_v2f16(
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-NEXT: v_floor_f16_e32 v0.l, v0.l
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-NEXT: v_floor_f16_e32 v0.h, v1.l
-; GFX11-NEXT: v_mov_b16_e32 v1.l, v0.l
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT: v_mov_b16_e32 v0.l, v0.h
-; GFX11-NEXT: v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT: v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
; GFX11-NEXT: buffer_store_b32 v0, off, s[4:7], 0
; GFX11-NEXT: s_endpgm
;
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll b/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll
index 53c26cadbf75a6..ff1c3da1d5fe5a 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll
@@ -480,9 +480,8 @@ define <2 x half> @test_ldexp_v2f16_v2i32(<2 x half> %a, <2 x i32> %b) {
; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.h, v3.l, v2.l
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.l, v0.l, v1.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-SDAG-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v2f16_v2i32:
@@ -610,9 +609,7 @@ define <2 x half> @test_ldexp_v2f16_v2i16(<2 x half> %a, <2 x i16> %b) {
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.l, v0.l, v1.l
; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.h, v3.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v2f16_v2i16:
@@ -737,15 +734,13 @@ define <3 x half> @test_ldexp_v3f16_v3i32(<3 x half> %a, <3 x i32> %b) {
; GFX11-SDAG-TRUE16-NEXT: v_lshrrev_b32_e32 v5, 16, v0
; GFX11-SDAG-TRUE16-NEXT: v_med3_i32 v3, v3, s0, 0x7fff
; GFX11-SDAG-TRUE16-NEXT: v_med3_i32 v2, v2, s0, 0x7fff
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-SDAG-TRUE16-NEXT: v_med3_i32 v4, v4, s0, 0x7fff
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.h, v5.l, v3.l
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.l, v0.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT: v_med3_i32 v2, v4, s0, 0x7fff
; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v3.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v1.l, v1.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT: v_perm_b32 v0, v3, v0, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v1.l, v1.l, v4.l
+; GFX11-SDAG-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v3f16_v3i32:
@@ -891,12 +886,9 @@ define <3 x half> @test_ldexp_v3f16_v3i16(<3 x half> %a, <3 x i16> %b) {
; GFX11-SDAG-TRUE16-NEXT: v_lshrrev_b32_e32 v5, 16, v0
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.l, v0.l, v2.l
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v1.l, v1.l, v3.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.h, v5.l, v4.l
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v2.l, v0.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT: v_perm_b32 v0, v0, v2, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v3f16_v3i16:
@@ -1036,24 +1028,21 @@ define <4 x half> @test_ldexp_v4f16_v4i32(<4 x half> %a, <4 x i32> %b) {
; GFX11-SDAG-TRUE16: ; %bb.0:
; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-TRUE16-NEXT: s_movk_i32 s0, 0x8000
-; GFX11-SDAG-TRUE16-NEXT: v_lshrrev_b32_e32 v6, 16, v0
+; GFX11-SDAG-TRUE16-NEXT: v_lshrrev_b32_e32 v6, 16, v1
; GFX11-SDAG-TRUE16-NEXT: v_med3_i32 v5, v5, s0, 0x7fff
; GFX11-SDAG-TRUE16-NEXT: v_med3_i32 v3, v3, s0, 0x7fff
-; GFX11-SDAG-TRUE16-NEXT: v_lshrrev_b32_e32 v7, 16, v1
+; GFX11-SDAG-TRUE16-NEXT: v_lshrrev_b32_e32 v7, 16, v0
; GFX11-SDAG-TRUE16-NEXT: v_med3_i32 v2, v2, s0, 0x7fff
; GFX11-SDAG-TRUE16-NEXT: v_med3_i32 v4, v4, s0, 0x7fff
+; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v1.h, v6.l, v5.l
; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.h, v6.l, v3.l
-; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v1.h, v7.l, v5.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.h, v7.l, v3.l
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.l, v0.l, v2.l
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v1.l, v1.l, v4.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v2.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100
-; GFX11-SDAG-TRUE16-NEXT: v_perm_b32 v1, v3, v1, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX11-SDAG-TRUE16-NEXT: v_pack_b32_f16 v1, v1.l, v1.h
; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v4f16_v4i32:
@@ -1238,20 +1227,14 @@ define <4 x half> @test_ldexp_v4f16_v4i16(<4 x half> %a, <4 x i16> %b) {
; GFX11-SDAG-TRUE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2
; GFX11-SDAG-TRUE16-NEXT: v_lshrrev_b32_e32 v6, 16, v0
; GFX11-SDAG-TRUE16-NEXT: v_lshrrev_b32_e32 v7, 16, v1
+; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v1.l, v1.l, v3.l
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.l, v0.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.h, v1.l, v3.l
; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v1.l, v6.l, v5.l
+; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.h, v6.l, v5.l
; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v1.h, v7.l, v4.l
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v2.l, v0.l
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v3.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, v1.l
-; GFX11-SDAG-TRUE16-NEXT: v_mov_b16_e32 v1.l, v1.h
; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT: v_perm_b32 v0, v0, v2, 0x5040100
-; GFX11-SDAG-TRUE16-NEXT: v_perm_b32 v1, v1, v3, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
+; GFX11-SDAG-TRUE16-NEXT: v_pack_b32_f16 v1, v1.l, v1.h
; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v4f16_v4i16:
diff --git a/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll b/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll
index 245df6684384c6..94b22b79f66323 100644
--- a/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll
@@ -237,14 +237,9 @@ define amdgpu_kernel void @sitofp_v2i16_to_v2f16(
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-TRUE16-NEXT: v_cvt_f16_i16_e32 v0.l, v0.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_cvt_f16_i16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, 0xffff, v1
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
; GFX11-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], 0
; GFX11-TRUE16-NEXT: s_endpgm
;
@@ -338,17 +333,13 @@ define amdgpu_kernel void @sitofp_v2i32_to_v2f16(
; GFX11-TRUE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], 0
; GFX11-TRUE16-NEXT: s_mov_b32 s5, s1
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT: v_cvt_f32_i32_e32 v0, v0
; GFX11-TRUE16-NEXT: v_cvt_f32_i32_e32 v1, v1
+; GFX11-TRUE16-NEXT: v_cvt_f32_i32_e32 v2, v0
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0
-; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v1
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, 0xffff, v1
-; GFX11-TRUE16-NEXT: v_lshl_or_b32 v0, v0, 16, v1
+; 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: buffer_store_b32 v0, off, s[4:7], 0
; GFX11-TRUE16-NEXT: s_endpgm
;
diff --git a/llvm/test/CodeGen/AMDGPU/uitofp.f16.ll b/llvm/test/CodeGen/AMDGPU/uitofp.f16.ll
index bc1b102d33de16..2a2fd93bc2d0b7 100644
--- a/llvm/test/CodeGen/AMDGPU/uitofp.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/uitofp.f16.ll
@@ -237,14 +237,9 @@ define amdgpu_kernel void @uitofp_v2i16_to_v2f16(
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-TRUE16-NEXT: v_cvt_f16_u16_e32 v0.l, v0.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_cvt_f16_u16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, 0xffff, v1
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
; GFX11-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], 0
; GFX11-TRUE16-NEXT: s_endpgm
;
@@ -338,17 +333,13 @@ define amdgpu_kernel void @uitofp_v2i32_to_v2f16(
; GFX11-TRUE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], 0
; GFX11-TRUE16-NEXT: s_mov_b32 s5, s1
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v0
; GFX11-TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v1
+; GFX11-TRUE16-NEXT: v_cvt_f32_u32_e32 v2, v0
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0
-; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v1
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, 0xffff, v1
-; GFX11-TRUE16-NEXT: v_lshl_or_b32 v0, v0, 16, v1
+; 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: buffer_store_b32 v0, off, s[4:7], 0
; GFX11-TRUE16-NEXT: s_endpgm
;
diff --git a/llvm/test/CodeGen/AMDGPU/v_pack.ll b/llvm/test/CodeGen/AMDGPU/v_pack.ll
index 2eba67b06bae1d..072151dd6f5a08 100644
--- a/llvm/test/CodeGen/AMDGPU/v_pack.ll
+++ b/llvm/test/CodeGen/AMDGPU/v_pack.ll
@@ -1,6 +1,10 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1010 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
; RUN: llc -global-isel -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1010 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GISEL %s
+; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=-flat-for-global,-real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-GCN-FAKE16 %s
+; RUN: llc -global-isel -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=-flat-for-global,-real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-GISEL-FAKE16 %s
+; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=-flat-for-global,+real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-GCN-REAL16 %s
+; RUN: llc -global-isel -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=-flat-for-global,+real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-GISEL-REAL16 %s
declare i32 @llvm.amdgcn.workitem.id.x() #1
@@ -38,6 +42,89 @@ define amdgpu_kernel void @v_pack_b32_v2f16(ptr addrspace(1) %in0, ptr addrspace
; GISEL-NEXT: ; use v0
; GISEL-NEXT: ;;#ASMEND
; GISEL-NEXT: s_endpgm
+;
+; GFX11-GCN-FAKE16-LABEL: v_pack_b32_v2f16:
+; GFX11-GCN-FAKE16: ; %bb.0:
+; GFX11-GCN-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-GCN-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; GFX11-GCN-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-GCN-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-GCN-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-GCN-FAKE16-NEXT: global_load_u16 v1, v0, s[0:1] glc dlc
+; GFX11-GCN-FAKE16-NEXT: s_waitcnt vmcnt(0)
+; GFX11-GCN-FAKE16-NEXT: global_load_u16 v0, v0, s[2:3] glc dlc
+; GFX11-GCN-FAKE16-NEXT: s_waitcnt vmcnt(0)
+; GFX11-GCN-FAKE16-NEXT: v_add_f16_e32 v1, 2.0, v1
+; GFX11-GCN-FAKE16-NEXT: v_add_f16_e32 v0, 2.0, v0
+; GFX11-GCN-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-GCN-FAKE16-NEXT: v_pack_b32_f16 v0, v1, v0
+; GFX11-GCN-FAKE16-NEXT: ;;#ASMSTART
+; GFX11-GCN-FAKE16-NEXT: ; use v0
+; GFX11-GCN-FAKE16-NEXT: ;;#ASMEND
+; GFX11-GCN-FAKE16-NEXT: s_endpgm
+;
+; GFX11-GISEL-FAKE16-LABEL: v_pack_b32_v2f16:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-GISEL-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-GISEL-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-GISEL-FAKE16-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-GISEL-FAKE16-NEXT: global_load_u16 v1, v0, s[0:1] glc dlc
+; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0)
+; GFX11-GISEL-FAKE16-NEXT: global_load_u16 v0, v0, s[2:3] glc dlc
+; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0)
+; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v1, 2.0, v1
+; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, 2.0, v0
+; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-GISEL-FAKE16-NEXT: v_pack_b32_f16 v0, v1, v0
+; GFX11-GISEL-FAKE16-NEXT: ;;#ASMSTART
+; GFX11-GISEL-FAKE16-NEXT: ; use v0
+; GFX11-GISEL-FAKE16-NEXT: ;;#ASMEND
+; GFX11-GISEL-FAKE16-NEXT: s_endpgm
+;
+; GFX11-GCN-REAL16-LABEL: v_pack_b32_v2f16:
+; GFX11-GCN-REAL16: ; %bb.0:
+; GFX11-GCN-REAL16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-GCN-REAL16-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; GFX11-GCN-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-GCN-REAL16-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-GCN-REAL16-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-GCN-REAL16-NEXT: global_load_u16 v1, v0, s[0:1] glc dlc
+; GFX11-GCN-REAL16-NEXT: s_waitcnt vmcnt(0)
+; GFX11-GCN-REAL16-NEXT: global_load_u16 v2, v0, s[2:3] glc dlc
+; GFX11-GCN-REAL16-NEXT: s_waitcnt vmcnt(0)
+; GFX11-GCN-REAL16-NEXT: v_mov_b16_e32 v0.l, v1.l
+; GFX11-GCN-REAL16-NEXT: v_mov_b16_e32 v0.h, v2.l
+; GFX11-GCN-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-GCN-REAL16-NEXT: v_add_f16_e32 v0.l, 2.0, v0.l
+; GFX11-GCN-REAL16-NEXT: v_add_f16_e32 v0.h, 2.0, v0.h
+; GFX11-GCN-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-GCN-REAL16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h
+; GFX11-GCN-REAL16-NEXT: ;;#ASMSTART
+; GFX11-GCN-REAL16-NEXT: ; use v0
+; GFX11-GCN-REAL16-NEXT: ;;#ASMEND
+;...
[truncated]
|
arsenm
approved these changes
Jan 10, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/85/builds/4298 Here is the relevant piece of the build log for the reference |
kazutakahirata
pushed a commit
to kazutakahirata/llvm-project
that referenced
this pull request
Jan 13, 2025
…lvm#121988) true16 codegen pattern for v_pack_b32_f16
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
true16 codegen pattern for v_pack_b32_f16