From c9de3d57f70cfbbb96df8f2b124bef6d99b19098 Mon Sep 17 00:00:00 2001
From: Benjamin Lee <benjamin.lee@collabora.com>
Date: Mon, 2 Dec 2024 15:19:45 -0800
Subject: [PATCH] panfrost/va: add FLUSH instruction

This is needed to implement FTZ for intermediate values on valhall.

Signed-off-by: Benjamin Lee <benjamin.lee@collabora.com>
Reviewed-by: Mary Guillemard <mary.guillemard@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32470>
---
 src/panfrost/compiler/compiler.h        |  9 ++++++---
 src/panfrost/compiler/valhall/ISA.xml   | 25 +++++++++++++++++++++++++
 src/panfrost/compiler/valhall/va_pack.c |  9 +++++++++
 3 files changed, 40 insertions(+), 3 deletions(-)

diff --git a/src/panfrost/compiler/compiler.h b/src/panfrost/compiler/compiler.h
index 244e5a7ce86..b4dc34c47b1 100644
--- a/src/panfrost/compiler/compiler.h
+++ b/src/panfrost/compiler/compiler.h
@@ -34,6 +34,7 @@
 #include "util/u_worklist.h"
 #include "bi_opcodes.h"
 #include "bifrost.h"
+#include "valhall_enums.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -473,9 +474,11 @@ typedef struct {
       bool format;               /* LEA_TEX */
 
       struct {
-         enum bi_special special; /* FADD_RSCALE, FMA_RSCALE */
-         enum bi_round round;     /* FMA, converts, FADD, _RSCALE, etc */
-         bool ftz;                /* Flush-to-zero for F16_TO_F32 */
+         enum bi_special special;   /* FADD_RSCALE, FMA_RSCALE */
+         enum bi_round round;       /* FMA, converts, FADD, _RSCALE, etc */
+         bool ftz;                  /* Flush-to-zero for F16_TO_F32 and FLUSH */
+         enum va_nan_mode nan_mode; /* NaN flush mode, for FLUSH */
+         bool flush_inf;            /* Flush infinity to finite, for FLUSH */
       };
 
       struct {
diff --git a/src/panfrost/compiler/valhall/ISA.xml b/src/panfrost/compiler/valhall/ISA.xml
index 5bdecf2a8cd..18e23d02493 100644
--- a/src/panfrost/compiler/valhall/ISA.xml
+++ b/src/panfrost/compiler/valhall/ISA.xml
@@ -778,6 +778,16 @@
     <value desc="Set bottom bit">aor1</value>
   </enum>
 
+  <enum name="NaN mode">
+    <desc>
+      Flush specific NaN values in FLUSH.f32 and FLUSH.v2f16. flush_nan flushes
+      all NaN values to zero. quiet_nan flushes signaling NaNs to quiet NaNs.
+    </desc>
+    <value desc="None">none</value>
+    <value desc="Flush NaN">flush_nan</value>
+    <value desc="Quiet NaN">quiet_nan</value>
+  </enum>
+
   <!-- note that the `unused="true"` annotation here just means that this
        particular entry is unused by the compiler. This may be because the
        instruction isn't generated yet, but it may also be because there
@@ -1691,6 +1701,21 @@
     <subgroup/>
   </ins>
 
+  <group name = "FLUSH" title="Flush floats" dests="1" opcode="0x98" unit="CVT">
+    <ins name="FLUSH.f32" opcode2="0"/>
+    <ins name="FLUSH.v2f16" opcode2="1"/>
+    <desc>
+      Flush special float values. The ftz modifier flushes subnormal values to
+      zero. The flush_inf modifier flushes +inf to the maximum finite value, and
+      -inf to the minimum finite value. nan_mode may flush either all NaN values
+      to zero or signaling NaNs to quiet NaNs depending on the mode.
+    </desc>
+    <va_mod name="nan_mode" start="8" size="2"/>
+    <va_mod name="ftz" start="10" size="1"/>
+    <va_mod name="flush_inf" start="11" size="1"/>
+    <src float="true" absneg="true" swizzle="true"/>
+  </group>
+
   <group name="FREXP" title="Fraction/exponent extract" dests="1" opcode="0x99" unused="true" unit="CVT">
     <ins name="FREXPM.f32" opcode2="0"/>
     <ins name="FREXPM.v2f16" opcode2="1"/>
diff --git a/src/panfrost/compiler/valhall/va_pack.c b/src/panfrost/compiler/valhall/va_pack.c
index 89a653967a5..cea55e83017 100644
--- a/src/panfrost/compiler/valhall/va_pack.c
+++ b/src/panfrost/compiler/valhall/va_pack.c
@@ -427,6 +427,15 @@ va_pack_alu(const bi_instr *I)
          hex |= 1ull << 25;
       break;
 
+   case BI_OPCODE_FLUSH_F32:
+   case BI_OPCODE_FLUSH_V2F16:
+      hex |= I->nan_mode << 8;
+      if (I->ftz)
+         hex |= 1ull << 10;
+      if (I->flush_inf)
+         hex |= 1ull << 11;
+      break;
+
    /* Add mux type */
    case BI_OPCODE_MUX_I32:
    case BI_OPCODE_MUX_V2I16: