diff --git a/src/nouveau/mme/mme_builder.h b/src/nouveau/mme/mme_builder.h
index 0a5eb321541..2db219e72b4 100644
--- a/src/nouveau/mme/mme_builder.h
+++ b/src/nouveau/mme/mme_builder.h
@@ -131,6 +131,14 @@ mme_free_reg(struct mme_builder *b, struct mme_value val)
    mme_reg_alloc_free(&b->reg_alloc, val);
 }
 
+static inline struct mme_value64
+mme_alloc_reg64(struct mme_builder *b)
+{
+   struct mme_value lo = mme_alloc_reg(b);
+   struct mme_value hi = mme_alloc_reg(b);
+   return mme_value64(lo, hi);
+}
+
 static inline void
 mme_free_reg64(struct mme_builder *b, struct mme_value64 val)
 {
@@ -295,6 +303,26 @@ mme_sub64(struct mme_builder *b,
    return mme_alu64(b, MME_ALU_OP_SUB, MME_ALU_OP_SUBB, x, y);
 }
 
+static inline struct mme_value
+mme_mul_32x32_32_free_srcs(struct mme_builder *b,
+                           struct mme_value x, struct mme_value y)
+{
+   assert(x.type == MME_VALUE_TYPE_REG);
+   assert(y.type == MME_VALUE_TYPE_REG);
+   if (b->devinfo->cls_eng3d >= MME_CLS_TURING) {
+      struct mme_value dst = mme_mul(b, x, y);
+      mme_free_reg(b, x);
+      mme_free_reg(b, y);
+      return dst;
+   } else if (b->devinfo->cls_eng3d >= MME_CLS_FERMI) {
+      struct mme_value dst = mme_alloc_reg(b);
+      mme_fermi_umul_32x32_32_to_free_srcs(b, dst, x, y);
+      return dst;
+   } else {
+      unreachable("Unsupported GPU class");
+   }
+}
+
 static inline void
 mme_imul_32x32_64_to(struct mme_builder *b, struct mme_value64 dst,
                      struct mme_value x, struct mme_value y)
@@ -317,6 +345,7 @@ static inline void
 mme_umul_32x32_64_to(struct mme_builder *b, struct mme_value64 dst,
                      struct mme_value x, struct mme_value y)
 {
+   assert(b->devinfo->cls_eng3d >= MME_CLS_TURING);
    mme_alu64_to(b, dst, MME_ALU_OP_MULU, MME_ALU_OP_MULH,
                 mme_value64(x, mme_zero()),
                 mme_value64(y, mme_zero()));
@@ -326,11 +355,56 @@ static inline struct mme_value64
 mme_umul_32x32_64(struct mme_builder *b,
                   struct mme_value x, struct mme_value y)
 {
+   assert(b->devinfo->cls_eng3d >= MME_CLS_TURING);
    return mme_alu64(b, MME_ALU_OP_MULU, MME_ALU_OP_MULH,
                     mme_value64(x, mme_zero()),
                     mme_value64(y, mme_zero()));
 }
 
+static inline struct mme_value64
+mme_umul_32x32_64_free_srcs(struct mme_builder *b,
+                            struct mme_value x, struct mme_value y)
+{
+   assert(x.type == MME_VALUE_TYPE_REG);
+   assert(y.type == MME_VALUE_TYPE_REG);
+   if (b->devinfo->cls_eng3d >= MME_CLS_TURING) {
+      struct mme_value64 dst = mme_umul_32x32_64(b, x, y);
+      mme_free_reg(b, x);
+      mme_free_reg(b, y);
+      return dst;
+   } else if (b->devinfo->cls_eng3d >= MME_CLS_FERMI) {
+      struct mme_value y_hi = mme_mov(b, mme_zero());
+      struct mme_value64 dst = mme_alloc_reg64(b);
+      mme_fermi_umul_32x64_64_to_free_srcs(b, dst, x, mme_value64(y, y_hi));
+      return dst;
+   } else {
+      unreachable("Unsupported GPU class");
+   }
+}
+
+static inline struct mme_value64
+mme_umul_32x64_64_free_srcs(struct mme_builder *b,
+                            struct mme_value x, struct mme_value64 y)
+{
+   assert(x.type == MME_VALUE_TYPE_REG);
+   assert(y.lo.type == MME_VALUE_TYPE_REG);
+   assert(y.hi.type == MME_VALUE_TYPE_REG);
+   if (b->devinfo->cls_eng3d >= MME_CLS_TURING) {
+      struct mme_value64 dst = mme_umul_32x32_64(b, x, y.lo);
+      struct mme_value tmp = mme_mul(b, x, y.hi);
+      mme_add64_to(b, dst, dst, mme_value64(mme_zero(), tmp));
+      mme_free_reg(b, x);
+      mme_free_reg64(b, y);
+      return dst;
+   } else if (b->devinfo->cls_eng3d >= MME_CLS_FERMI) {
+      struct mme_value64 dst = mme_alloc_reg64(b);
+      mme_fermi_umul_32x64_64_to_free_srcs(b, dst, x, y);
+      return dst;
+   } else {
+      unreachable("Unsupported GPU class");
+   }
+}
+
 static inline struct mme_value64
 mme_mul64(struct mme_builder *b,
           struct mme_value64 x, struct mme_value64 y)
diff --git a/src/nouveau/mme/mme_fermi_builder.c b/src/nouveau/mme/mme_fermi_builder.c
index 676c0057c6e..2c60785bb8f 100644
--- a/src/nouveau/mme/mme_fermi_builder.c
+++ b/src/nouveau/mme/mme_fermi_builder.c
@@ -338,6 +338,28 @@ mme_fermi_umul_32x32_32_to_free_srcs(struct mme_builder *b,
       mme_srl_to(b, x, x, mme_imm(1u));
       mme_sll_to(b, y, y, mme_imm(1u));
    }
+   mme_free_reg(b, x);
+   mme_free_reg(b, y);
+}
+
+void
+mme_fermi_umul_32x64_64_to_free_srcs(struct mme_builder *b,
+                                     struct mme_value64 dst,
+                                     struct mme_value x,
+                                     struct mme_value64 y)
+{
+   mme_while (b, ine, x, mme_zero()) {
+      struct mme_value lsb = mme_and(b, x, mme_imm(1));
+      mme_if (b, ine, lsb, mme_zero()) {
+         mme_add64_to(b, dst, dst, y);
+      }
+      mme_free_reg(b, lsb);
+      mme_srl_to(b, x, x, mme_imm(1u));
+      /* y = y << 1 */
+      mme_add64_to(b, y, y, y);
+   }
+   mme_free_reg(b, x);
+   mme_free_reg64(b, y);
 }
 
 static struct mme_value
diff --git a/src/nouveau/mme/mme_fermi_builder.h b/src/nouveau/mme/mme_fermi_builder.h
index 0e955e7f090..d0e4ef7e042 100644
--- a/src/nouveau/mme/mme_fermi_builder.h
+++ b/src/nouveau/mme/mme_fermi_builder.h
@@ -115,6 +115,12 @@ mme_fermi_umul_32x32_32_to_free_srcs(struct mme_builder *b,
                                      struct mme_value x,
                                      struct mme_value y);
 
+void
+mme_fermi_umul_32x64_64_to_free_srcs(struct mme_builder *b,
+                                     struct mme_value64 dst,
+                                     struct mme_value x,
+                                     struct mme_value64 y);
+
 void
 mme_fermi_merge_to(struct mme_builder *b, struct mme_value dst,
                    struct mme_value x, struct mme_value y,
diff --git a/src/nouveau/mme/tests/mme_builder_test.cpp b/src/nouveau/mme/tests/mme_builder_test.cpp
index 0ccc55dd9e8..aa85769d443 100644
--- a/src/nouveau/mme/tests/mme_builder_test.cpp
+++ b/src/nouveau/mme/tests/mme_builder_test.cpp
@@ -205,7 +205,7 @@ static const uint32_t mul_cases[] = {
    0xfffe0000,
 };
 
-TEST_F(mme_builder_test, mul)
+TEST_F(mme_builder_test, mul_32x32_32)
 {
    for (auto sim : sims) {
       mme_builder b;
@@ -214,7 +214,7 @@ TEST_F(mme_builder_test, mul)
       mme_value x = mme_load(&b);
       mme_value y = mme_load(&b);
 
-      sim->mme_store_data(&b, 0, mme_mul(&b, x, y));
+      sim->mme_store_data(&b, 0, mme_mul_32x32_32_free_srcs(&b, x, y));
 
       auto macro = mme_builder_finish_vec(&b);
 
@@ -231,6 +231,81 @@ TEST_F(mme_builder_test, mul)
    }
 }
 
+TEST_F(mme_builder_test, umul_32x32_64)
+{
+   for (auto sim : sims) {
+      mme_builder b;
+      mme_builder_init(&b, sim->devinfo);
+
+      mme_value x = mme_load(&b);
+      mme_value y = mme_load(&b);
+
+      struct mme_value64 d = mme_umul_32x32_64_free_srcs(&b, x, y);
+
+      sim->mme_store_data(&b, 0, d.lo);
+      sim->mme_store_data(&b, 1, d.hi);
+
+      auto macro = mme_builder_finish_vec(&b);
+
+      for (uint32_t i = 0; i < ARRAY_SIZE(mul_cases); i++) {
+         for (uint32_t j = 0; j < ARRAY_SIZE(mul_cases); j++) {
+            std::vector<uint32_t> params;
+            params.push_back(mul_cases[i]);
+            params.push_back(mul_cases[j]);
+
+            sim->run_macro(macro, params);
+
+            uint64_t d = (uint64_t)mul_cases[i] * (uint64_t)mul_cases[j];
+            ASSERT_EQ(sim->data[0], (uint32_t)d);
+            ASSERT_EQ(sim->data[1], (uint32_t)(d >> 32));
+         }
+      }
+   }
+}
+
+TEST_F(mme_builder_test, umul_32x64_64)
+{
+   for (auto sim : sims) {
+      mme_builder b;
+      mme_builder_init(&b, sim->devinfo);
+
+      mme_value x = mme_load(&b);
+      struct mme_value64 y;
+      y.lo = mme_load(&b);
+      y.hi = mme_load(&b);
+
+      struct mme_value64 d = mme_umul_32x64_64_free_srcs(&b, x, y);
+
+      sim->mme_store_data(&b, 0, d.lo);
+      sim->mme_store_data(&b, 1, d.hi);
+
+      auto macro = mme_builder_finish_vec(&b);
+
+      for (uint32_t i = 0; i < ARRAY_SIZE(mul_cases); i++) {
+         for (uint32_t j = 0; j < ARRAY_SIZE(mul_cases); j++) {
+            for (uint32_t k = 0; k < ARRAY_SIZE(mul_cases); k++) {
+               std::vector<uint32_t> params;
+               params.push_back(mul_cases[i]);
+               params.push_back(mul_cases[j]);
+               params.push_back(mul_cases[k]);
+
+               sim->run_macro(macro, params);
+
+               uint32_t x = mul_cases[i];
+               uint32_t y_lo = mul_cases[j];
+               uint32_t y_hi = mul_cases[k];
+               uint64_t y = ((uint64_t)y_hi << 32) | (uint64_t)y_lo;
+
+               uint64_t d = x * y;
+
+               ASSERT_EQ(sim->data[0], (uint32_t)d);
+               ASSERT_EQ(sim->data[1], (uint32_t)(d >> 32));
+            }
+         }
+      }
+   }
+}
+
 TEST_F(mme_builder_test, sll_srl)
 {
    static const uint32_t x = 0xac406fe1;