[llvm] [AMDGPU][True16][MC] true16 for v_minmax/maxmin_f16 (PR #119586)

via llvm-commits llvm-commits at lists.llvm.org
Wed Dec 11 20:59:48 PST 2024 

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Brox Chen (broxigarchen)

<details>
<summary>Changes</summary>

Support true16 format for v_minmax/maxmin_f16 in MC.

Since we are replacing `v_minmax/maxmin_f16` to `v_minmax/maxmin_f16_t16 / v_minmax/maxmin_f16_fake16` in Post-GFX11, have to update the CodeGen pattern for `v_minmax/maxmin_f16` to get CodeGen test passing.

---

Patch is 144.98 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/119586.diff


9 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/SIInstructions.td (+7-4) 
- (modified) llvm/lib/Target/AMDGPU/VOP3Instructions.td (+4-4) 
- (modified) llvm/lib/Target/AMDGPU/VOPInstructions.td (+6) 
- (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop3.s (+84-60) 
- (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp16.s (+146-56) 
- (modified) llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp8.s (+126-48) 
- (modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3.txt (+168-30) 
- (modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3_dpp16.txt (+172-28) 
- (modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3_dpp8.txt (+156-24) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index bc25d75131cc35..44bc800b0dae81 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -3698,12 +3698,15 @@ def : IntMinMaxPat<V_MAXMIN_U32_e64, umin, umax_oneuse>;
 def : IntMinMaxPat<V_MINMAX_U32_e64, umax, umin_oneuse>;
 def : FPMinMaxPat<V_MINMAX_F32_e64, f32, fmaxnum_like, fminnum_like_oneuse>;
 def : FPMinMaxPat<V_MAXMIN_F32_e64, f32, fminnum_like, fmaxnum_like_oneuse>;
-def : FPMinMaxPat<V_MINMAX_F16_e64, f16, fmaxnum_like, fminnum_like_oneuse>;
-def : FPMinMaxPat<V_MAXMIN_F16_e64, f16, fminnum_like, fmaxnum_like_oneuse>;
 def : FPMinCanonMaxPat<V_MINMAX_F32_e64, f32, fmaxnum_like, fminnum_like_oneuse>;
 def : FPMinCanonMaxPat<V_MAXMIN_F32_e64, f32, fminnum_like, fmaxnum_like_oneuse>;
-def : FPMinCanonMaxPat<V_MINMAX_F16_e64, f16, fmaxnum_like, fminnum_like_oneuse>;
-def : FPMinCanonMaxPat<V_MAXMIN_F16_e64, f16, fminnum_like, fmaxnum_like_oneuse>;
+}
+
+let True16Predicate = UseFakeTrue16Insts in {
+def : FPMinMaxPat<V_MINMAX_F16_fake16_e64, f16, fmaxnum_like, fminnum_like_oneuse>;
+def : FPMinMaxPat<V_MAXMIN_F16_fake16_e64, f16, fminnum_like, fmaxnum_like_oneuse>;
+def : FPMinCanonMaxPat<V_MINMAX_F16_fake16_e64, f16, fmaxnum_like, fminnum_like_oneuse>;
+def : FPMinCanonMaxPat<V_MAXMIN_F16_fake16_e64, f16, fminnum_like, fmaxnum_like_oneuse>;
 }
 
 let OtherPredicates = [isGFX9Plus] in {
diff --git a/llvm/lib/Target/AMDGPU/VOP3Instructions.td b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
index 8a9f8aa3d16d3a..fafa7797e02060 100644
--- a/llvm/lib/Target/AMDGPU/VOP3Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
@@ -1372,8 +1372,8 @@ class VOP3_DOT_Profile_fake16<VOPProfile P, VOP3Features Features = VOP3_REGULAR
 let SubtargetPredicate = isGFX11Plus in {
   defm V_MAXMIN_F32     : VOP3Inst<"v_maxmin_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
   defm V_MINMAX_F32     : VOP3Inst<"v_minmax_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
-  defm V_MAXMIN_F16     : VOP3Inst<"v_maxmin_f16", VOP3_Profile<VOP_F16_F16_F16_F16>>;
-  defm V_MINMAX_F16     : VOP3Inst<"v_minmax_f16", VOP3_Profile<VOP_F16_F16_F16_F16>>;
+  defm V_MAXMIN_F16     : VOP3Inst_t16<"v_maxmin_f16", VOP_F16_F16_F16_F16>;
+  defm V_MINMAX_F16     : VOP3Inst_t16<"v_minmax_f16", VOP_F16_F16_F16_F16>;
   defm V_MAXMIN_U32     : VOP3Inst<"v_maxmin_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
   defm V_MINMAX_U32     : VOP3Inst<"v_minmax_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
   defm V_MAXMIN_I32     : VOP3Inst<"v_maxmin_i32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
@@ -1728,8 +1728,8 @@ defm V_PERMLANE16_B32      : VOP3_Real_Base_gfx11_gfx12<0x25b>;
 defm V_PERMLANEX16_B32     : VOP3_Real_Base_gfx11_gfx12<0x25c>;
 defm V_MAXMIN_F32          : VOP3_Realtriple_gfx11<0x25e>;
 defm V_MINMAX_F32          : VOP3_Realtriple_gfx11<0x25f>;
-defm V_MAXMIN_F16          : VOP3_Realtriple_gfx11<0x260>;
-defm V_MINMAX_F16          : VOP3_Realtriple_gfx11<0x261>;
+defm V_MAXMIN_F16          : VOP3_Realtriple_t16_and_fake16_gfx11<0x260, "v_maxmin_f16">;
+defm V_MINMAX_F16          : VOP3_Realtriple_t16_and_fake16_gfx11<0x261, "v_minmax_f16">;
 defm V_MAXMIN_U32          : VOP3_Realtriple_gfx11_gfx12<0x262>;
 defm V_MINMAX_U32          : VOP3_Realtriple_gfx11_gfx12<0x263>;
 defm V_MAXMIN_I32          : VOP3_Realtriple_gfx11_gfx12<0x264>;
diff --git a/llvm/lib/Target/AMDGPU/VOPInstructions.td b/llvm/lib/Target/AMDGPU/VOPInstructions.td
index 5c4d96d3688b87..633bafa0a202e8 100644
--- a/llvm/lib/Target/AMDGPU/VOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOPInstructions.td
@@ -1905,6 +1905,12 @@ multiclass VOP3_Realtriple_t16_gfx11<bits<10> op, string asmName, string opName
                                      string pseudo_mnemonic = "", bit isSingle = 0> :
   VOP3_Realtriple_with_name<GFX11Gen, op, opName, asmName, pseudo_mnemonic, isSingle>;
 
+multiclass VOP3_Realtriple_t16_and_fake16_gfx11<bits<10> op, string asmName, string opName = NAME,
+                                                string pseudo_mnemonic = "", bit isSingle = 0> {
+  defm _t16: VOP3_Realtriple_t16_gfx11<op, asmName, opName#"_t16", pseudo_mnemonic, isSingle>;
+  defm _fake16: VOP3_Realtriple_t16_gfx11<op, asmName, opName#"_fake16", pseudo_mnemonic, isSingle>;
+}
+
 multiclass VOP3Only_Realtriple_t16_gfx11<bits<10> op, string asmName,
                                      string opName = NAME, string pseudo_mnemonic = "">
   : VOP3_Realtriple_t16_gfx11<op, asmName, opName, pseudo_mnemonic, 1>;
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop3.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop3.s
index b649bab532f262..feca177a420336 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop3.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop3.s
@@ -3572,50 +3572,62 @@ v_max_u16 v5.l, v255.l, v255.h
 v_max_u16 v255.h, 0xfe0b, vcc_hi
 // GFX11: v_max_u16 v255.h, 0xfe0b, vcc_hi op_sel:[0,0,1] ; encoding: [0xff,0x40,0x09,0xd7,0xff,0xd6,0x00,0x00,0x0b,0xfe,0x00,0x00]
 
-v_maxmin_f16 v5, v1, v2, s3
-// GFX11: v_maxmin_f16 v5, v1, v2, s3             ; encoding: [0x05,0x00,0x60,0xd6,0x01,0x05,0x0e,0x00]
+v_maxmin_f16 v5.l, v1.l, v2.l, s3
+// GFX11: v_maxmin_f16 v5.l, v1.l, v2.l, s3       ; encoding: [0x05,0x00,0x60,0xd6,0x01,0x05,0x0e,0x00]
 
-v_maxmin_f16 v5, v255, s2, s105
-// GFX11: v_maxmin_f16 v5, v255, s2, s105         ; encoding: [0x05,0x00,0x60,0xd6,0xff,0x05,0xa4,0x01]
+v_maxmin_f16 v5.l, v255.l, s2, s105
+// GFX11: v_maxmin_f16 v5.l, v255.l, s2, s105     ; encoding: [0x05,0x00,0x60,0xd6,0xff,0x05,0xa4,0x01]
 
-v_maxmin_f16 v5, s1, v255, exec_hi
-// GFX11: v_maxmin_f16 v5, s1, v255, exec_hi      ; encoding: [0x05,0x00,0x60,0xd6,0x01,0xfe,0xff,0x01]
+v_maxmin_f16 v5.l, s1, v255.l, exec_hi
+// GFX11: v_maxmin_f16 v5.l, s1, v255.l, exec_hi  ; encoding: [0x05,0x00,0x60,0xd6,0x01,0xfe,0xff,0x01]
 
-v_maxmin_f16 v5, s105, s105, exec_lo
-// GFX11: v_maxmin_f16 v5, s105, s105, exec_lo    ; encoding: [0x05,0x00,0x60,0xd6,0x69,0xd2,0xf8,0x01]
+v_maxmin_f16 v5.l, s105, s105, exec_lo
+// GFX11: v_maxmin_f16 v5.l, s105, s105, exec_lo  ; encoding: [0x05,0x00,0x60,0xd6,0x69,0xd2,0xf8,0x01]
 
-v_maxmin_f16 v5, vcc_lo, ttmp15, v3
-// GFX11: v_maxmin_f16 v5, vcc_lo, ttmp15, v3     ; encoding: [0x05,0x00,0x60,0xd6,0x6a,0xf6,0x0c,0x04]
+v_maxmin_f16 v5.l, vcc_lo, ttmp15, v3.l
+// GFX11: v_maxmin_f16 v5.l, vcc_lo, ttmp15, v3.l ; encoding: [0x05,0x00,0x60,0xd6,0x6a,0xf6,0x0c,0x04]
 
-v_maxmin_f16 v5, vcc_hi, 0xfe0b, v255
-// GFX11: v_maxmin_f16 v5, vcc_hi, 0xfe0b, v255   ; encoding: [0x05,0x00,0x60,0xd6,0x6b,0xfe,0xfd,0x07,0x0b,0xfe,0x00,0x00]
+v_maxmin_f16 v5.l, vcc_hi, 0xfe0b, v255.l
+// GFX11: v_maxmin_f16 v5.l, vcc_hi, 0xfe0b, v255.l ; encoding: [0x05,0x00,0x60,0xd6,0x6b,0xfe,0xfd,0x07,0x0b,0xfe,0x00,0x00]
 
-v_maxmin_f16 v5, -|ttmp15|, -|src_scc|, -|ttmp15|
-// GFX11: v_maxmin_f16 v5, -|ttmp15|, -|src_scc|, -|ttmp15| ; encoding: [0x05,0x07,0x60,0xd6,0x7b,0xfa,0xed,0xe1]
+v_maxmin_f16 v5.l, -|ttmp15|, -|src_scc|, -|ttmp15|
+// GFX11: v_maxmin_f16 v5.l, -|ttmp15|, -|src_scc|, -|ttmp15| ; encoding: [0x05,0x07,0x60,0xd6,0x7b,0xfa,0xed,0xe1]
 
-v_maxmin_f16 v5, m0, 0.5, m0
-// GFX11: v_maxmin_f16 v5, m0, 0.5, m0            ; encoding: [0x05,0x00,0x60,0xd6,0x7d,0xe0,0xf5,0x01]
+v_maxmin_f16 v5.l, m0, 0.5, m0
+// GFX11: v_maxmin_f16 v5.l, m0, 0.5, m0          ; encoding: [0x05,0x00,0x60,0xd6,0x7d,0xe0,0xf5,0x01]
 
-v_maxmin_f16 v5, |exec_lo|, -1, vcc_hi
-// GFX11: v_maxmin_f16 v5, |exec_lo|, -1, vcc_hi  ; encoding: [0x05,0x01,0x60,0xd6,0x7e,0x82,0xad,0x01]
+v_maxmin_f16 v5.l, |exec_lo|, -1, vcc_hi
+// GFX11: v_maxmin_f16 v5.l, |exec_lo|, -1, vcc_hi ; encoding: [0x05,0x01,0x60,0xd6,0x7e,0x82,0xad,0x01]
 
-v_maxmin_f16 v5, -|exec_hi|, null, -|vcc_lo|
-// GFX11: v_maxmin_f16 v5, -|exec_hi|, null, -|vcc_lo| ; encoding: [0x05,0x05,0x60,0xd6,0x7f,0xf8,0xa8,0xa1]
+v_maxmin_f16 v5.l, -|exec_hi|, null, -|vcc_lo|
+// GFX11: v_maxmin_f16 v5.l, -|exec_hi|, null, -|vcc_lo| ; encoding: [0x05,0x05,0x60,0xd6,0x7f,0xf8,0xa8,0xa1]
 
-v_maxmin_f16 v5, null, exec_lo, -|0xfe0b|
-// GFX11: v_maxmin_f16 v5, null, exec_lo, -|0xfe0b| ; encoding: [0x05,0x04,0x60,0xd6,0x7c,0xfc,0xfc,0x83,0x0b,0xfe,0x00,0x00]
+v_maxmin_f16 v5.l, null, exec_lo, -|0xfe0b|
+// GFX11: v_maxmin_f16 v5.l, null, exec_lo, -|0xfe0b| ; encoding: [0x05,0x04,0x60,0xd6,0x7c,0xfc,0xfc,0x83,0x0b,0xfe,0x00,0x00]
 
-v_maxmin_f16 v5, -1, -|exec_hi|, -|src_scc|
-// GFX11: v_maxmin_f16 v5, -1, -|exec_hi|, -|src_scc| ; encoding: [0x05,0x06,0x60,0xd6,0xc1,0xfe,0xf4,0xc3]
+v_maxmin_f16 v5.l, -1, -|exec_hi|, -|src_scc|
+// GFX11: v_maxmin_f16 v5.l, -1, -|exec_hi|, -|src_scc| ; encoding: [0x05,0x06,0x60,0xd6,0xc1,0xfe,0xf4,0xc3]
 
-v_maxmin_f16 v5, 0.5, -m0, 0.5 mul:2
-// GFX11: v_maxmin_f16 v5, 0.5, -m0, 0.5 mul:2    ; encoding: [0x05,0x00,0x60,0xd6,0xf0,0xfa,0xc0,0x4b]
+v_maxmin_f16 v5.l, 0.5, -m0, 0.5 mul:2
+// GFX11: v_maxmin_f16 v5.l, 0.5, -m0, 0.5 mul:2  ; encoding: [0x05,0x00,0x60,0xd6,0xf0,0xfa,0xc0,0x4b]
 
-v_maxmin_f16 v5, -src_scc, |vcc_lo|, -1 mul:4
-// GFX11: v_maxmin_f16 v5, -src_scc, |vcc_lo|, -1 mul:4 ; encoding: [0x05,0x02,0x60,0xd6,0xfd,0xd4,0x04,0x33]
+v_maxmin_f16 v5.l, -src_scc, |vcc_lo|, -1 mul:4
+// GFX11: v_maxmin_f16 v5.l, -src_scc, |vcc_lo|, -1 mul:4 ; encoding: [0x05,0x02,0x60,0xd6,0xfd,0xd4,0x04,0x33]
 
-v_maxmin_f16 v255, -|0xfe0b|, -|vcc_hi|, null clamp div:2
-// GFX11: v_maxmin_f16 v255, -|0xfe0b|, -|vcc_hi|, null clamp div:2 ; encoding: [0xff,0x83,0x60,0xd6,0xff,0xd6,0xf0,0x79,0x0b,0xfe,0x00,0x00]
+v_maxmin_f16 v255.l, -|0xfe0b|, -|vcc_hi|, null clamp div:2
+// GFX11: v_maxmin_f16 v255.l, -|0xfe0b|, -|vcc_hi|, null clamp div:2 ; encoding: [0xff,0x83,0x60,0xd6,0xff,0xd6,0xf0,0x79,0x0b,0xfe,0x00,0x00]
+
+v_maxmin_f16 v5.l, v255.h, s2, s105
+// GFX11: v_maxmin_f16 v5.l, v255.h, s2, s105 op_sel:[1,0,0,0] ; encoding: [0x05,0x08,0x60,0xd6,0xff,0x05,0xa4,0x01]
+
+v_maxmin_f16 v5.l, s1, v255.h, exec_hi
+// GFX11: v_maxmin_f16 v5.l, s1, v255.h, exec_hi op_sel:[0,1,0,0] ; encoding: [0x05,0x10,0x60,0xd6,0x01,0xfe,0xff,0x01]
+
+v_maxmin_f16 v5.l, vcc_hi, 0xfe0b, v255.h
+// GFX11: v_maxmin_f16 v5.l, vcc_hi, 0xfe0b, v255.h op_sel:[0,0,1,0] ; encoding: [0x05,0x20,0x60,0xd6,0x6b,0xfe,0xfd,0x07,0x0b,0xfe,0x00,0x00]
+
+v_maxmin_f16 v255.h, -|0xfe0b|, -|vcc_hi|, null clamp div:2
+// GFX11: v_maxmin_f16 v255.h, -|0xfe0b|, -|vcc_hi|, null op_sel:[0,0,0,1] clamp div:2 ; encoding: [0xff,0xc3,0x60,0xd6,0xff,0xd6,0xf0,0x79,0x0b,0xfe,0x00,0x00]
 
 v_maxmin_f32 v5, v1, v2, s3
 // GFX11: v_maxmin_f32 v5, v1, v2, s3             ; encoding: [0x05,0x00,0x5e,0xd6,0x01,0x05,0x0e,0x00]
@@ -4532,50 +4544,62 @@ v_min_u16 v5.l, v255.l, v255.h
 v_min_u16 v255.h, 0xfe0b, vcc_hi
 // GFX11: v_min_u16 v255.h, 0xfe0b, vcc_hi op_sel:[0,0,1] ; encoding: [0xff,0x40,0x0b,0xd7,0xff,0xd6,0x00,0x00,0x0b,0xfe,0x00,0x00]
 
-v_minmax_f16 v5, v1, v2, s3
-// GFX11: v_minmax_f16 v5, v1, v2, s3             ; encoding: [0x05,0x00,0x61,0xd6,0x01,0x05,0x0e,0x00]
+v_minmax_f16 v5.l, v1.l, v2.l, s3
+// GFX11: v_minmax_f16 v5.l, v1.l, v2.l, s3       ; encoding: [0x05,0x00,0x61,0xd6,0x01,0x05,0x0e,0x00]
+
+v_minmax_f16 v5.l, v255.l, s2, s105
+// GFX11: v_minmax_f16 v5.l, v255.l, s2, s105     ; encoding: [0x05,0x00,0x61,0xd6,0xff,0x05,0xa4,0x01]
+
+v_minmax_f16 v5.l, s1, v255.l, exec_hi
+// GFX11: v_minmax_f16 v5.l, s1, v255.l, exec_hi  ; encoding: [0x05,0x00,0x61,0xd6,0x01,0xfe,0xff,0x01]
+
+v_minmax_f16 v5.l, s105, s105, exec_lo
+// GFX11: v_minmax_f16 v5.l, s105, s105, exec_lo  ; encoding: [0x05,0x00,0x61,0xd6,0x69,0xd2,0xf8,0x01]
+
+v_minmax_f16 v5.l, vcc_lo, ttmp15, v3.l
+// GFX11: v_minmax_f16 v5.l, vcc_lo, ttmp15, v3.l ; encoding: [0x05,0x00,0x61,0xd6,0x6a,0xf6,0x0c,0x04]
 
-v_minmax_f16 v5, v255, s2, s105
-// GFX11: v_minmax_f16 v5, v255, s2, s105         ; encoding: [0x05,0x00,0x61,0xd6,0xff,0x05,0xa4,0x01]
+v_minmax_f16 v5.l, vcc_hi, 0xfe0b, v255.l
+// GFX11: v_minmax_f16 v5.l, vcc_hi, 0xfe0b, v255.l ; encoding: [0x05,0x00,0x61,0xd6,0x6b,0xfe,0xfd,0x07,0x0b,0xfe,0x00,0x00]
 
-v_minmax_f16 v5, s1, v255, exec_hi
-// GFX11: v_minmax_f16 v5, s1, v255, exec_hi      ; encoding: [0x05,0x00,0x61,0xd6,0x01,0xfe,0xff,0x01]
+v_minmax_f16 v5.l, -|ttmp15|, -|src_scc|, -|ttmp15|
+// GFX11: v_minmax_f16 v5.l, -|ttmp15|, -|src_scc|, -|ttmp15| ; encoding: [0x05,0x07,0x61,0xd6,0x7b,0xfa,0xed,0xe1]
 
-v_minmax_f16 v5, s105, s105, exec_lo
-// GFX11: v_minmax_f16 v5, s105, s105, exec_lo    ; encoding: [0x05,0x00,0x61,0xd6,0x69,0xd2,0xf8,0x01]
+v_minmax_f16 v5.l, m0, 0.5, m0
+// GFX11: v_minmax_f16 v5.l, m0, 0.5, m0          ; encoding: [0x05,0x00,0x61,0xd6,0x7d,0xe0,0xf5,0x01]
 
-v_minmax_f16 v5, vcc_lo, ttmp15, v3
-// GFX11: v_minmax_f16 v5, vcc_lo, ttmp15, v3     ; encoding: [0x05,0x00,0x61,0xd6,0x6a,0xf6,0x0c,0x04]
+v_minmax_f16 v5.l, |exec_lo|, -1, vcc_hi
+// GFX11: v_minmax_f16 v5.l, |exec_lo|, -1, vcc_hi ; encoding: [0x05,0x01,0x61,0xd6,0x7e,0x82,0xad,0x01]
 
-v_minmax_f16 v5, vcc_hi, 0xfe0b, v255
-// GFX11: v_minmax_f16 v5, vcc_hi, 0xfe0b, v255   ; encoding: [0x05,0x00,0x61,0xd6,0x6b,0xfe,0xfd,0x07,0x0b,0xfe,0x00,0x00]
+v_minmax_f16 v5.l, -|exec_hi|, null, -|vcc_lo|
+// GFX11: v_minmax_f16 v5.l, -|exec_hi|, null, -|vcc_lo| ; encoding: [0x05,0x05,0x61,0xd6,0x7f,0xf8,0xa8,0xa1]
 
-v_minmax_f16 v5, -|ttmp15|, -|src_scc|, -|ttmp15|
-// GFX11: v_minmax_f16 v5, -|ttmp15|, -|src_scc|, -|ttmp15| ; encoding: [0x05,0x07,0x61,0xd6,0x7b,0xfa,0xed,0xe1]
+v_minmax_f16 v5.l, null, exec_lo, -|0xfe0b|
+// GFX11: v_minmax_f16 v5.l, null, exec_lo, -|0xfe0b| ; encoding: [0x05,0x04,0x61,0xd6,0x7c,0xfc,0xfc,0x83,0x0b,0xfe,0x00,0x00]
 
-v_minmax_f16 v5, m0, 0.5, m0
-// GFX11: v_minmax_f16 v5, m0, 0.5, m0            ; encoding: [0x05,0x00,0x61,0xd6,0x7d,0xe0,0xf5,0x01]
+v_minmax_f16 v5.l, -1, -|exec_hi|, -|src_scc|
+// GFX11: v_minmax_f16 v5.l, -1, -|exec_hi|, -|src_scc| ; encoding: [0x05,0x06,0x61,0xd6,0xc1,0xfe,0xf4,0xc3]
 
-v_minmax_f16 v5, |exec_lo|, -1, vcc_hi
-// GFX11: v_minmax_f16 v5, |exec_lo|, -1, vcc_hi  ; encoding: [0x05,0x01,0x61,0xd6,0x7e,0x82,0xad,0x01]
+v_minmax_f16 v5.l, 0.5, -m0, 0.5 mul:2
+// GFX11: v_minmax_f16 v5.l, 0.5, -m0, 0.5 mul:2  ; encoding: [0x05,0x00,0x61,0xd6,0xf0,0xfa,0xc0,0x4b]
 
-v_minmax_f16 v5, -|exec_hi|, null, -|vcc_lo|
-// GFX11: v_minmax_f16 v5, -|exec_hi|, null, -|vcc_lo| ; encoding: [0x05,0x05,0x61,0xd6,0x7f,0xf8,0xa8,0xa1]
+v_minmax_f16 v5.l, -src_scc, |vcc_lo|, -1 mul:4
+// GFX11: v_minmax_f16 v5.l, -src_scc, |vcc_lo|, -1 mul:4 ; encoding: [0x05,0x02,0x61,0xd6,0xfd,0xd4,0x04,0x33]
 
-v_minmax_f16 v5, null, exec_lo, -|0xfe0b|
-// GFX11: v_minmax_f16 v5, null, exec_lo, -|0xfe0b| ; encoding: [0x05,0x04,0x61,0xd6,0x7c,0xfc,0xfc,0x83,0x0b,0xfe,0x00,0x00]
+v_minmax_f16 v255.l, -|0xfe0b|, -|vcc_hi|, null clamp div:2
+// GFX11: v_minmax_f16 v255.l, -|0xfe0b|, -|vcc_hi|, null clamp div:2 ; encoding: [0xff,0x83,0x61,0xd6,0xff,0xd6,0xf0,0x79,0x0b,0xfe,0x00,0x00]
 
-v_minmax_f16 v5, -1, -|exec_hi|, -|src_scc|
-// GFX11: v_minmax_f16 v5, -1, -|exec_hi|, -|src_scc| ; encoding: [0x05,0x06,0x61,0xd6,0xc1,0xfe,0xf4,0xc3]
+v_minmax_f16 v5.l, v255.h, s2, s105
+// GFX11: v_minmax_f16 v5.l, v255.h, s2, s105 op_sel:[1,0,0,0] ; encoding: [0x05,0x08,0x61,0xd6,0xff,0x05,0xa4,0x01]
 
-v_minmax_f16 v5, 0.5, -m0, 0.5 mul:2
-// GFX11: v_minmax_f16 v5, 0.5, -m0, 0.5 mul:2    ; encoding: [0x05,0x00,0x61,0xd6,0xf0,0xfa,0xc0,0x4b]
+v_minmax_f16 v5.l, s1, v255.h, exec_hi
+// GFX11: v_minmax_f16 v5.l, s1, v255.h, exec_hi op_sel:[0,1,0,0] ; encoding: [0x05,0x10,0x61,0xd6,0x01,0xfe,0xff,0x01]
 
-v_minmax_f16 v5, -src_scc, |vcc_lo|, -1 mul:4
-// GFX11: v_minmax_f16 v5, -src_scc, |vcc_lo|, -1 mul:4 ; encoding: [0x05,0x02,0x61,0xd6,0xfd,0xd4,0x04,0x33]
+v_minmax_f16 v5.l, vcc_hi, 0xfe0b, v255.h
+// GFX11: v_minmax_f16 v5.l, vcc_hi, 0xfe0b, v255.h op_sel:[0,0,1,0] ; encoding: [0x05,0x20,0x61,0xd6,0x6b,0xfe,0xfd,0x07,0x0b,0xfe,0x00,0x00]
 
-v_minmax_f16 v255, -|0xfe0b|, -|vcc_hi|, null clamp div:2
-// GFX11: v_minmax_f16 v255, -|0xfe0b|, -|vcc_hi|, null clamp div:2 ; encoding: [0xff,0x83,0x61,0xd6,0xff,0xd6,0xf0,0x79,0x0b,0xfe,0x00,0x00]
+v_minmax_f16 v255.h, -|0xfe0b|, -|vcc_hi|, null clamp div:2
+// GFX11: v_minmax_f16 v255.h, -|0xfe0b|, -|vcc_hi|, null op_sel:[0,0,0,1] clamp div:2 ; encoding: [0xff,0xc3,0x61,0xd6,0xff,0xd6,0xf0,0x79,0x0b,0xfe,0x00,0x00]
 
 v_minmax_f32 v5, v1, v2, s3
 // GFX11: v_minmax_f32 v5, v1, v2, s3             ; encoding: [0x05,0x00,0x5f,0xd6,0x01,0x05,0x0e,0x00]
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp16.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp16.s
index e6f868d2b40e7e..cfaca994b793fc 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp16.s
@@ -2420,47 +2420,92 @@ v_max_u16_e64_dpp v5.l, v1.l, v2.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_
 v_max_u16_e64_dpp v255.h, v255.l, v255.l row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
 // GFX11: v_max_u16_e64_dpp v255.h, v255.l, v255.l op_sel:[0,0,1] row_xmask:15 row_mask:0x3 bank_mask:0x0 fi:1 ; encoding: [0xff,0x40,0x09,0xd7,0xfa,0xfe,0x03,0x00,0xff,0x6f,0x05,0x30]
 
-v_maxmin_f16_e64_dpp v5, v1, v2, v3 quad_perm:[3,2,1,0]
-// GFX11: v_maxmin_f16_e64_dpp v5, v1, v2, v3 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0x0e,0x04,0x01,0x1b,0x00,0xff]
+v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, v3.l quad_perm:[3,2,1,0]
+// GFX11: v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, v3.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0x0e,0x04,0x01,0x1b,0x00,0xff]
 
-v_maxmin_f16_e64_dpp v5, v1, v2, v3 quad_perm:[0,1,2,3]
-// GFX11: v_maxmin_f16_e64_dpp v5, v1, v2, v3 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x00,0xff]
+v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, v3.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf
+// GFX11: v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, v3.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0x0e,0x04,0x01,0xe4,0x00,0xff]
 
-v_maxmin_f16_e64_dpp v5, v1, v2, v3 row_mirror
-// GFX11: v_maxmin_f16_e64_dpp v5, v1, v2, v3 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0x0e,0x04,0x01,0x40,0x01,0xff]
+v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, v3.l row_mirror row_mask:0xf bank_mask:0xf
+// GFX11: v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, v3.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0x0e,0x04,0x01,0x40,0x01,0xff]
 
-v_maxmin_f16_e64_dpp v5, v1, v2, v255 row_half_mirror
-// GFX11: v_maxmin_f16_e64_dpp v5, v1, v2, v255 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0xfe,0x07,0x01,0x41,0x01,0xff]
+v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, v255.l row_half_mirror row_mask:0xf bank_mask:0xf
+// GFX11: v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, v255.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0xfe,0x07,0x01,0x41,0x01,0xff]
 
-v_maxmin_f16_e64_dpp v5, v1, v2, s105 row_shl:1
-// GFX11: v_maxmin_f16_e64_dpp v5, v1, v2, s105 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0xa6,0x01,0x01,0x01,0x01,0xff]
+v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, s105 row_shl:1 row_mask:0xf bank_mask:0xf
+// GFX11: v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, s105 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0xa6,0x01,0x01,0x01,0x01,0xff]
 
-v_maxmin_f16_e64_dpp v5, v1, v2, vcc_hi row_shl:15
-// GFX11: v_maxmin_f16_e64_dpp v5, v1, v2, vcc_hi row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0xae,0x01,0x01,0x0f,0x01,0xff]
+v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, vcc_hi row_shl:15
+// GFX11: v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, vcc_hi row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0xae,0x01,0x01,0x0f,0x01,0xff]
 
-v_maxmin_f16_e64_dpp v5, v1, v2, vcc_lo row_shr:1
-// GFX11: v_maxmin_f16_e64_dpp v5, v1, v2, vcc_lo row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0xaa,0x01,0x01,0x11,0x01,0xff]
+v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, vcc_lo row_shr:1
+// GFX11: v_maxmin_f16_e64_dpp v5.l, v1.l, v2.l, vcc_lo row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x00,0x60,0xd6,0xfa,0x04,0xaa,0x01,0x01,0x11,0x01,0xff]
 
-v_maxmin_f16_e64_dpp v5, |v1|, v2, -ttmp15 row_shr:15
-// GFX11: v_maxmin_f16_e64_dpp v5, |v1|, v2, -ttmp15 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0x05,0x01,0x60,0xd6,0xfa,0x04,0xee,0x81,0x01,0x1f,0x01,0xff]
+v_maxmin_f16_e64_dpp v5.l, |v1.l|, v2.l, -ttmp15 row_shr:15 row_mask:0xf bank_...
[truncated]

``````````

</details>


https://github.com/llvm/llvm-project/pull/119586

More information about the llvm-commits mailing list