diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index 953b67f7e2c..026f0ee5558 100644
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -2638,15 +2638,39 @@ late_optimizations = [
 
    # nir_lower_to_source_mods will collapse this, but its existence during the
    # optimization loop can prevent other optimizations.
-   (('fneg', ('fneg', a)), a),
+   (('fneg', ('fneg', a)), a)
+]
 
-   # re-combine inexact mul+add to ffma. Do this before fsub so that a * b - c
-   # gets combined to fma(a, b, -c).
-   (('~fadd@16', ('fmul(is_only_used_by_fadd)', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma16'),
-   (('~fadd@32', ('fmul(is_only_used_by_fadd)', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma32'),
-   (('~fadd@64', ('fmul(is_only_used_by_fadd)', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma64'),
-   (('~fadd@32', ('fmulz(is_only_used_by_fadd)', a, b), c), ('ffmaz', a, b, c), 'options->fuse_ffma32'),
+# re-combine inexact mul+add to ffma. Do this before fsub so that a * b - c
+# gets combined to fma(a, b, -c).
+for sz, mulz in itertools.product([16, 32, 64], [False, True]):
+    # fmulz/ffmaz only for fp32
+    if mulz and sz != 32:
+        continue
 
+    # Fuse the correct fmul. Only consider fmuls where the only users are fadd
+    # (or fneg/fabs which are assumed to be propagated away), as a heuristic to
+    # avoid fusing in cases where it's harmful.
+    fmul = ('fmulz' if mulz else 'fmul') + '(is_only_used_by_fadd)'
+    ffma = 'ffmaz' if mulz else 'ffma'
+
+    fadd = '~fadd@{}'.format(sz)
+    option = 'options->fuse_ffma{}'.format(sz)
+
+    late_optimizations.extend([
+        ((fadd, (fmul, a, b), c), (ffma, a, b, c), option),
+
+        ((fadd, ('fneg(is_only_used_by_fadd)', (fmul, a, b)), c),
+         (ffma, ('fneg', a), b, c), option),
+
+        ((fadd, ('fabs(is_only_used_by_fadd)', (fmul, a, b)), c),
+         (ffma, ('fabs', a), ('fabs', b), c), option),
+
+        ((fadd, ('fneg(is_only_used_by_fadd)', ('fabs', (fmul, a, b))), c),
+         (ffma, ('fneg', ('fabs', a)), ('fabs', b), c), option),
+    ])
+
+late_optimizations.extend([
    # Subtractions get lowered during optimization, so we need to recombine them
    (('fadd@8', a, ('fneg', 'b')), ('fsub', 'a', 'b'), 'options->has_fsub'),
    (('fadd@16', a, ('fneg', 'b')), ('fsub', 'a', 'b'), 'options->has_fsub'),
@@ -2823,7 +2847,7 @@ late_optimizations = [
    (('extract_i8', ('extract_u8', a, b), 0), ('extract_i8', a, b)),
    (('extract_u8', ('extract_i8', a, b), 0), ('extract_u8', a, b)),
    (('extract_u8', ('extract_u8', a, b), 0), ('extract_u8', a, b)),
-]
+])
 
 # A few more extract cases we'd rather leave late
 for N in [16, 32]:
diff --git a/src/compiler/nir/nir_search_helpers.h b/src/compiler/nir/nir_search_helpers.h
index da1e9b506a8..5308fc41e40 100644
--- a/src/compiler/nir/nir_search_helpers.h
+++ b/src/compiler/nir/nir_search_helpers.h
@@ -433,8 +433,12 @@ is_only_used_by_fadd(const nir_alu_instr *instr)
       const nir_alu_instr *const user_alu = nir_instr_as_alu(user_instr);
       assert(instr != user_alu);
 
-      if (user_alu->op != nir_op_fadd)
+      if (user_alu->op == nir_op_fneg || user_alu->op == nir_op_fabs) {
+         if (!is_only_used_by_fadd(user_alu))
+            return false;
+      } else if (user_alu->op != nir_op_fadd) {
          return false;
+      }
    }
 
    return true;
diff --git a/src/gallium/drivers/virgl/ci/traces-virgl-iris.yml b/src/gallium/drivers/virgl/ci/traces-virgl-iris.yml
index 7ad46c464ed..05aaf2c9541 100644
--- a/src/gallium/drivers/virgl/ci/traces-virgl-iris.yml
+++ b/src/gallium/drivers/virgl/ci/traces-virgl-iris.yml
@@ -30,7 +30,7 @@ traces:
       checksum: 32e8b627a33ad08d416dfdb804920371
   0ad/0ad-v2.trace:
     gl-virgl:
-      checksum: bf22fd7c3fc8baa7b0e9345728626d5f
+      checksum: 638fa405f78a6631ba829a8fc98392a6
   glmark2/buffer:update-fraction=0.5:update-dispersion=0.9:columns=200:update-method=map:interleave=false-v2.trace:
     gl-virgl:
       checksum: 040232e01e394a967dc3320bb9252870
@@ -42,7 +42,7 @@ traces:
       checksum: df21895268db3ab185ae5ffa5b2d7f37
   glmark2/bump:bump-render=height-v2.trace:
     gl-virgl:
-      checksum: cd32f46925906c53fae747372a8f2ed8
+      checksum: cceb2b8d4852b94709684b69c688638c
   glmark2/bump:bump-render=high-poly-v2.trace:
     gl-virgl:
       checksum: 11b7a4820b452934e6f12b57b8910a9a
@@ -126,7 +126,7 @@ traces:
       label: [crash]
   gputest/pixmark-julia-fp32-v2.trace:
     gl-virgl:
-      checksum: 0aa3a82a5b849cb83436e52c4e3e95ac
+      checksum: fbf5e44a6f46684b84e5bb5ad6d36c67
   gputest/pixmark-julia-fp64-v2.trace:
     gl-virgl:
       checksum: 1760aea00af985b8cd902128235b08f6
diff --git a/src/gallium/drivers/virgl/ci/traces-virgl.yml b/src/gallium/drivers/virgl/ci/traces-virgl.yml
index 7da834d4f83..2be64e826d5 100644
--- a/src/gallium/drivers/virgl/ci/traces-virgl.yml
+++ b/src/gallium/drivers/virgl/ci/traces-virgl.yml
@@ -123,7 +123,7 @@ traces:
       label: [crash]
   gputest/pixmark-julia-fp32-v2.trace:
     gl-virgl:
-      checksum: 25f938c726c68c08a88193f28f7c4474
+      checksum: 8b3584b1dd8f1d1bb63205564bd78e4e
   gputest/pixmark-julia-fp64-v2.trace:
     gl-virgl:
       checksum: 73ccaff82ea764057fb0f93f0024cf84
@@ -183,7 +183,7 @@ traces:
       checksum: f4af4067b37c00861fa5911e4c0a6629
   supertuxkart/supertuxkart-mansion-egl-gles-v2.trace:
     gl-virgl:
-      checksum: 092e8ca38e58aaa83df2a9f0b7b8aee5
+      checksum: cc7092975dd6c9064aa54cd7f18053b6
   xonotic/xonotic-keybench-high-v2.trace:
     gl-virgl:
       checksum: f3b184bf8858a6ebccd09e7ca032197e